+

WO1999007543A1 - Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation - Google Patents

Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation Download PDF

Info

Publication number
WO1999007543A1
WO1999007543A1 PCT/JP1997/002783 JP9702783W WO9907543A1 WO 1999007543 A1 WO1999007543 A1 WO 1999007543A1 JP 9702783 W JP9702783 W JP 9702783W WO 9907543 A1 WO9907543 A1 WO 9907543A1
Authority
WO
WIPO (PCT)
Prior art keywords
tire
design
objective function
value
determined
Prior art date
Application number
PCT/JP1997/002783
Other languages
English (en)
French (fr)
Inventor
Yukio Nakajima
Original Assignee
Bridgestone Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Bridgestone Corporation filed Critical Bridgestone Corporation
Priority to PCT/JP1997/002783 priority Critical patent/WO1999007543A1/ja
Priority to DE69731222T priority patent/DE69731222T2/de
Priority to US09/269,972 priority patent/US7369976B1/en
Priority to EP97934746A priority patent/EP0937570B1/en
Priority to JP51194999A priority patent/JP4393595B2/ja
Priority to ES97934746T priority patent/ES2229377T3/es
Publication of WO1999007543A1 publication Critical patent/WO1999007543A1/ja

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/10Geometric CAD
    • G06F30/15Vehicle, aircraft or watercraft design
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B60VEHICLES IN GENERAL
    • B60CVEHICLE TYRES; TYRE INFLATION; TYRE CHANGING; CONNECTING VALVES TO INFLATABLE ELASTIC BODIES IN GENERAL; DEVICES OR ARRANGEMENTS RELATED TO TYRES
    • B60C3/00Tyres characterised by the transverse section
    • B60C3/04Tyres characterised by the transverse section characterised by the relative dimensions of the section, e.g. low profile

Definitions

  • the present invention relates to a design method for a tire, an optimization analysis device, and a storage medium storing an optimization analysis program.
  • the present invention relates to a storage medium storing a tire design method, an optimization analysis device, and an optimization analysis program.
  • the present invention relates to a tire structure, shape, and pattern design, for example, a tire sidewall shape and a crown.
  • the present invention relates to a tire design method, an optimization analyzer, and a storage medium storing an optimization analysis program which can be used for designing the shape of a part.
  • the structure, shape, and pattern design of a tire is to determine the structure, shape, pattern, manufacturing conditions, and the like for obtaining the desired tire performance.
  • the performance of this tire is a physical quantity obtained by calculation or experiment or an actual vehicle filling evaluation result.
  • Conventional tire design methods such as tire structure, shape, and pattern design consisted of trial and error empirical rules obtained by repeating experiments and numerical experiments using a computer. As a result, the number of prototypes and tests required for development became enormous, which increased development costs and shortened the development period.
  • Finding the optimal solution is like climbing a mountain. At this time, since the altitude of the mountain is related to the performance, the optimal solution corresponds to the top of the mountain.
  • the objective function is simple, its design space (mountain shape) has one peak as shown in Fig. 8, so it is necessary to find the optimal solution using an optimization method based on mathematical programming. Can be.
  • the present invention can design the best mode of a tire under given conditions, and can provide a tire design method, an optimization analysis device, and an optimization method that can increase the efficiency of tire design and development.
  • the purpose is to obtain a storage medium that stores a fragmentation analysis program. Disclosure of the invention
  • the present inventor has made various studies and found that a nonlinear prediction technology that is used in a different field is engineered to model a neural network of a higher animal, for example, a neural network. Focusing on applying the “optimization design method” and the “optimization design method” to the special field of tire design, we tried to study and established it as a concrete tire design method.
  • the tire design method of the present invention includes: (a) a conversion system for associating a non-linear correspondence between a tire design parameter representing a tire cross-sectional shape including an internal structure or a tire structure and a performance of the tire; (B) defining an objective function representing the performance of the tire, and at least reducing the performance of the tire and the manufacturing conditions of the tire. And (c) using the conversion system defined in step (a) to determine an optimal value of the objective function based on the objective function and the constraint condition.
  • the method includes the steps of obtaining a design parameter and designing the tire based on the design parameter of the tire.
  • the values of tire performance are determined by tire design parameters, for example, tire cross-sectional shape including internal structure and tire structure.
  • tire design parameters for example, tire cross-sectional shape including internal structure and tire structure.
  • the tire performance does not change linearly even when the values of the tire cross-sectional shape and the tire structure are changed linearly.
  • a conversion system for associating a non-linear correspondence between the tire design parameters representing the tire cross-sectional shape including the internal structure or the tire structure and the performance of the tire is predetermined.
  • This conversion system can be determined using a non-linear prediction technique that engineeringly models a neural network such as a neural network.
  • an objective function representing the tire performance is determined, and constraints that restrict at least one of the permissible ranges of the tire performance and the tire manufacturing conditions are determined.
  • Objective functions that express tire performance include, for example, superiority and inferiority of tire performance such as tire circumferential belt tension and lateral spring constant when air is filled to improve steering stability, and grounding characteristics in contact with the ground when straight or lateral force is applied. Can be used to control physical quantities.
  • Restrictions that constrain at least one of the permissible ranges of tire performance and tire design parameters include, for example, restrictions on the tire cross-sectional shape and tire structure, such as constraints on force-cass line peripheral values, and upper and lower primary natural frequencies.
  • Constraints belt layer angle constraints, belt layer width, tire dimensions, panel constants, tire deformation, tire weight, stress, strain, strain energy, rolling resistance, etc. It should be noted that the objective function and the constraint conditions are not limited to the above example, and various types can be determined according to the tire design purpose.
  • step (c) using the conversion system determined in step (a), the tire design parameters for giving the optimal value of the objective function are determined based on the objective function and the constraint conditions, and the design parameters are obtained.
  • Design tires based on data.
  • a conversion system that associates the non-linear correspondence between tire design parameters and tire performance is determined, and the conversion system associates the correspondence between the design parameters of multiple tires and their performance. Relationships can be found. Therefore, by obtaining tire design parameters that give the optimum value of the objective function and designing the tire based on the design parameters, a high-performance tire can be designed.
  • this step (C) the value of the design variable that gives the optimal value of the objective function can be obtained while considering the constraints.
  • the design parameters of the tire are defined as design variables, and the design variables that give the optimal value of the objective function using the conversion system determined in step (a) while considering the constraints.
  • the tire can be designed based on the design variables that give the optimal value of the objective function. As described above, by taking into account the constraint conditions, it is possible to consider the allowable range of at least one of the tire performance and the tire design parameter, and to specify the design range in advance or set a desired range.
  • the sensitivity of the objective function which is the ratio of the change of the objective function to the unit change of the design variable, and the change in the constraint condition for the unit change of the design variable
  • the amount of change in the design variable that gives the optimal value of the objective function is predicted while considering the constraint, and the objective when the design variable is changed by an amount corresponding to the predicted amount
  • the tire can be designed by changing the tire design parameters and the like based on the design variables that give the optimum value of the objective function.
  • This transformation system is used in the design space (mountain shape) Indicated by the same level (contour line). That is, various design parameters are related to the tire performance, and it is general that the range of the design parameter is narrowed like a contour line as the tire performance is optimized. In addition, the range of tire design parameters generally includes design constraints and the range that can be actually taken.
  • the relationship between tire performance, which is the objective function, and tire design parameters is shown in Fig. 8. It can be limited by fences along the ridgeline of the mountain as shown. If this is a constraint, the objective function obtains an optimal solution by changing the design variables in the transformation system on the peak of the shape as shown by the contour line so as not to exceed the fence of the constraint. This is equivalent to climbing a mountain shaped as shown in Fig. 8 with the help of optimization techniques such as mathematical programming up to the top.
  • the optimal solution when obtaining the optimal solution by the above steps (a) to (c), the optimal solution can be obtained only after executing the following steps (d) to (f).
  • the step (c) includes a step (d) of selecting one of the tire design parameters included in the conversion system determined in the step (a) as a design variable, and (E) changing the value of a design variable selected from within the conversion system determined in step (a) until the optimal value of the objective function is given using the conversion system determined in step (a); (F) designing a tire based on tire design parameters based on design variables that give optimum values of the tires.
  • step (d) one of the tire design parameters included in the conversion system is selected as a design variable.
  • the value of the design variable to be selected is changed from within the conversion system until the optimal value of the objective function is given, taking into account the constraints.
  • the value of the design variable changes slightly or gradually, and the optimum value of the objective function is given.
  • the tire is designed based on the tire design parameters based on the design variables that give the optimal value of the objective function. In this way, one of the design parameters of the tire included in the conversion system is selected as a design variable, and the value of the design variable selected from the conversion system is changed while giving the optimal value of the objective function while considering the constraint conditions.
  • step (b) it is possible to define a constraint that restricts the allowable range of at least one of the tire performance and the tire design parameter other than the determined objective function.
  • the tire performance differs from the objective function in the restricted allowable range.
  • Tire performance is used. If the above constraints are not set, tire performance and design parameters other than the objective function will deviate from the desired ranges, making practical application difficult in many cases.
  • step (a) the transformation system in which the non-linear correspondence between the tire design parameters and the tire performance selected as the objective function in step (a) is related by a neural network or the like is shown in the contour lines in Fig. 8, and the contour lines shown in the contour lines
  • the constraint condition defined in step (b) is provided as a fence on the top of the mountain, and in step (d) included in step (c), the tire design parameters in the conversion system determined in step (a) are selected.
  • an optimization method such as a mathematical programming or a genetic algorithm is performed until the objective function obtains an optimal solution so as not to exceed the constraint condition in step (e). With the help of, climb the mountain of the shape.
  • the constraint condition (fence) is effective for an optimization method as a guide for climbing a mountain, in addition to setting a desired range of tire performance and design parameters other than the objective function in advance.
  • the optimal solution is obtained only when the steps (a) to (e) are integrated.
  • step (e) the so-called mathematical programming is applied, and the sensitivity of the objective function, which is the ratio of the change of the objective function to the unit change of the design variable, and the change of the constraint condition with respect to the unit change of the design variable.
  • the change amount of the design variable that gives the optimum value of the objective function is predicted based on the sensitivity of the constraint condition, which is the ratio of the constraint condition, while considering the constraint condition, and the objective when the design variable is changed by an amount corresponding to the predicted amount.
  • Calculate the value of the constraint condition when the value of the function and the design variable are changed by an amount corresponding to the predicted amount and determine the value in step (a) based on the predicted value and the calculated value while considering the constraint condition.
  • the value of the design variable to be selected can be changed until the optimal value of the objective function is given using the transformed system. Wear. In this way, by calculating the value of the objective function when the design variable is changed by the amount corresponding to the predicted amount and the value of the constraint condition when the design variable is changed by the amount corresponding to the predicted amount, the value of the objective function is calculated. The values of the design variables up to the optimum value can be easily obtained.
  • the tire design parameters in the conversion system determined in the step (a) are used as a basic model and a plurality of basic models are used. Is determined, and for each basic model of the selected group, an adaptive function that can be evaluated from the objective function, design variables, constraints, and objective function is determined, and two basic models are selected from the selected group. Then, a new basic model is generated by crossing the design variables of each basic model with a predetermined probability, and a new basic model is generated by partially changing the design variables of at least one of the basic models. At least one of the above is performed, the design variables are changed, and the objective function, constraints and adaptation function of the basic model are obtained using the conversion system defined in step (a).
  • This model and the basic model in which the design variables are not changed are stored and repeated until the number of stored basic models reaches a predetermined number, and whether a new population of the stored predetermined number of basic models satisfies a predetermined convergence condition If the convergence condition is not satisfied, the new group is used as the selection target group until the selection target group satisfies a predetermined convergence condition, and the new group is repeated until the predetermined convergence condition is satisfied.
  • the tire design parameters based on the design variables that give the optimal value of the objective function are obtained and used as the tire design parameters. Design the tire based on it.
  • a conversion system can be configured with data of a multilayer feedforward neural network that has been learned to convert tire design parameters into tire performance.
  • general optimization methods include mathematical programming and genetic algorithms, and finding the optimal solution is like climbing. .
  • the optimal solution corresponds to the top of the mountain
  • step (a) a conversion system that associates the non-linear correspondence between the tire design parameters (design variables) and the performance of the tire is determined using a neural network.
  • Tire performance is determined by the design parameters for tire shape, structure, and pattern. However, even if the design parameters are changed linearly, the tire performance often does not change linearly.
  • neural networks are described in P.11-13, P.162- in “Nonlinear Multivariate Analysis: An Approach by Neural Networks” by Hideki Toyoda (published by Asakura Shoten, 1996). As can be seen in 166, higher accuracy of prediction and discrimination can be expected than multivariate analysis of linear transformation, and it is also possible to learn the relationship between input data and the unit of the intermediate layer.
  • a conversion system in which the correspondence including the non-linear correspondence between the design parameter and the tire performance is determined in advance.
  • This conversion system can be determined using a non-linear prediction technology that engineeringly models a neural network such as a neural network.
  • the formed tire is configured with design parameters having the best performance.
  • the tire designing method includes: a conversion system calculating means for obtaining a non-linear correspondence between tire design parameters and the performance of the tire; and an objective function representing the tire performance.
  • Input means for defining a constraint condition for restricting at least one allowable range of the tire manufacturing conditions, and inputting the same as an optimization item; and an optimum input by the input means using the conversion system calculating means.
  • the present invention can be realized by an optimization analysis device including optimization calculation means for obtaining design parameters of a tire that gives an optimum value of an objective function based on optimization items.
  • the conversion system calculating means can determine a non-linear correspondence between the tire design parameters and application conditions for the tire and the tire performance. These application conditions include the manufacturing conditions, tire weight or overall cost when building the tire. Further, the conversion system calculation means may be configured by a multilayer feedforward neural network that has been learned to convert the tire design variables into the tire performance.
  • the optimization calculation means includes: a selection means for selecting one of the tire design parameters included in the conversion system calculation means as a design variable; and an optimum value of the objective function while considering the constraints.
  • a changing means for changing the value of a design variable selected from within the conversion system calculating means until the optimum value is calculated, and an optimum value calculation for calculating the value of the design variable until the optimum value of the objective function is given using the conversion system calculating means.
  • the optimization calculating means determines a selection target group including a plurality of basic models using the tire design parameters in the correspondence obtained by the conversion system calculation means as a basic model, and selects each basic model of the selection target group.
  • an objective function, design variables, constraints, and an adaptive function that can be evaluated from the objective function are determined, and the selected pair Select two basic models from the elephant group and cross over the design variables of each basic model with a predetermined probability to generate a new basic model, and change at least one of the basic model design variables to create a new basic model.
  • the stored basic models are stored until the number of stored basic models reaches a predetermined number, and it is determined whether or not a new group of the stored predetermined number of basic models satisfies a predetermined convergence condition. If not, the new group is used as the selection target group, and the process is repeated until the selection target group satisfies a predetermined convergence condition. While taking into account the constraints among the predetermined number of basic models stored at that time, the conversion system calculation means is used to obtain tire design parameters based on design variables that give the optimum value of the objective function, and based on the tire design parameters. Tire design.
  • the conversion system calculation means may be configured by a multilayer feedforward type neural network learned to convert the tire design parameters into the tire performance.
  • the above-described tire designing method can provide a storage medium storing an optimization analysis program which can be easily carried by a storage medium including a program according to the following procedure. That is, a recording medium that records a tire optimization analysis program for designing a tire using a computer defines a non-linear correspondence between tire design parameters and performance of the tire as the optimization analysis program. In addition to defining an objective function representing the performance of the tire, defining a constraint condition that restricts at least one of a permissible range of the performance of the tire and a manufacturing condition of the tire, the determined correspondence, the objective function, and the constraint condition.
  • a tire optimization analysis program is characterized in that tire design parameters that give the optimum value of the objective function are obtained based on the tire design parameters, and the tire is designed based on the tire design parameters.
  • one of the tire design parameters included in the determined correspondence is selected as a design variable based on the determined correspondence, the objective function, and the constraint. While taking the above constraints into account. Changing the design variables selected from the above-mentioned correspondences until the optimal value of the objective function is given, and designing the tire based on the tire design parameters based on the design variables giving the optimal value of the objective function. Can be.
  • the constraint condition can restrict an allowable range of at least one of a tire performance other than the determined objective function and a design parameter of the tire.
  • the change in the design variable is determined by the sensitivity of the objective function, which is the ratio of the change of the objective function to the unit change of the design variable, and the sensitivity of the constraint, which is the ratio of the change of the constraint condition to the unit change of the design variable.
  • the amount of change in the design variable that gives the optimal value of the objective function is predicted while considering the constraint conditions based on the constraints. Calculate the value of the constraint condition when the amount is changed by a corresponding amount, and change the value of the design variable to be selected based on the predicted value and the calculated value until the optimal value of the objective function is given while considering the aforementioned constraint condition. It can be done.
  • a selection target group including a plurality of basic models is determined using the tire design parameters in the determined correspondence relationship as a basic model, and the objective function, the design variables, and the constraint conditions are defined for each of the basic models of the selection target group.
  • an adaptive function that can be evaluated from the objective function is determined, two basic models are selected from the selection target group, and a design variable of each basic model is crossed at a predetermined probability to generate a new basic model; and At least one of generating a new basic model by changing a part of the design variables of at least one of the basic models is performed, and the objective function, constraint condition, and adaptation function of the basic model whose design variables are changed are obtained.
  • the basic model without changing the basic model and design variables was saved, and the stored basic model was repeated until the number of the stored basic models reached a predetermined number. It is determined whether a new group consisting of a basic model of constants satisfies a predetermined convergence condition. If the new group does not satisfy the convergence condition, the new group is used as the selection target group until the selection target group satisfies a predetermined convergence condition.
  • the tire can be designed based on the design parameters of the tire by obtaining the parameters.
  • a conversion system that associates the non-linear correspondence between the tire shape, structure, and pattern design parameters and the performance of the tire is determined using a neural network. Therefore, there is an effect that the conversion system can be created with high accuracy and low arbitrariness.
  • FIG. 1 is an external view of an optimization device according to the present embodiment.
  • FIG. 2 is a schematic configuration diagram of the optimization device according to the present embodiment.
  • FIG. 3 is a schematic block diagram for each function of the optimization device according to the present embodiment.
  • Fig. 4 is a conceptual diagram of the neural network.
  • FIG. 5 is a flowchart showing a flow of the operation of the optimization device according to the present embodiment.
  • FIG. 6 is a flowchart showing a flow of a learning process of the neural network.
  • FIG. 7 is a flowchart illustrating a flow of the optimization processing according to the first embodiment.
  • FIG. 8 is a conceptual diagram of an image for explaining the optimization of the present invention.
  • FIG. 9 is another conceptual image diagram for explaining the optimization of the present invention.
  • FIG. 10 is a flowchart illustrating the flow of the optimization process according to the second embodiment.
  • FIG. 11 is a flowchart showing the flow of the crossover process.
  • FIGS. 12A and 12B are diagrams illustrating a mountain-shaped mapping function, wherein FIG. 12A is a diagram illustrating a continuous mountain-shaped mapping function, and FIG. 12B is a diagram illustrating a linear mountain-shaped mapping function.
  • FIG. 13 is a diagram showing a valley-type mapping function
  • (a) is a diagram showing a continuous valley-type mapping function
  • (b) is a diagram showing a linear valley-type mapping function.
  • FIG. 14 is a flowchart showing the flow of the mutation process.
  • FIG. 15 is an image diagram for explaining the design variables of the first test example.
  • FIG. 16 is an image diagram for explaining the design variables of the second test example.
  • FIG. 17 is a cross-sectional view showing a tire element immediately below a load.
  • FIG. 18 is an image diagram for explaining the design variables of the third test example. BEST MODE FOR CARRYING OUT THE INVENTION
  • the present invention is applied to an optimizing device that obtains optimal tire design parameters.
  • the neural network of a higher animal is designed as a conversion system using a neural network after learning, which is a non-linear prediction technology that is modeled by engineering. I am looking for data.
  • FIG. 1 shows an outline of an optimizing device 30 for performing the optimization according to the present invention.
  • the optimizer 30 uses a keyboard 10 for inputting data and the like, and uses a neural network based on a nonlinear prediction method in accordance with a program stored in advance to determine the shape, structure and pattern design parameters of the tire. From the computer body 12 that predicts performance and computes design variables that satisfy the constraints and optimize the objective function (eg, maximum or minimum), and the CRT 14 that displays the computation results of the computer body 12 It is configured.
  • the objective function eg, maximum or minimum
  • the optimization device 30 includes a computer main body 12 including a microcomputer, a data input / output device 28, and a key board 10 for inputting data and commands. , And a monitor.
  • the computer main body 12 has a CPU 16, a ROM 18, a RAM 20, a memory 22 for storing a conversion system (to be described in detail later), and exchanges data between the main body and another device. It comprises an input / output device (hereinafter, referred to as IZO) 26 for receiving data and a bus 24 connected to these so that data and commands can be input and output.
  • the ROM 18 stores a processing program described later.
  • the data input / output device 28 has a numerically expressed tire shape, structure, and pattern design parameters, manufacturing conditions, and tire performance (in the present embodiment, the tire shape, structure, and pattern, etc.). A device for reading from external storage means when stored in storage means. This is not necessary when the key board 10 is used as an input device.
  • FIG. 3 is a block diagram illustrating a schematic configuration of each function of the optimization device 30 according to the present embodiment.
  • the optimization device 30 of the present embodiment optimizes the tire performance to be maximized or minimized (this is called an objective function) and outputs design parameters for the optimized tire performance.
  • the optimization device 30 is classified into a nonlinear operation section 32, an optimization operation section 34, an experiment data input section 40, an optimization item input section 42, and an optimization result output section 44 by function. It is.
  • the non-linear operation unit 32 functions as a calculation unit of a conversion system composed of a neural network (details will be described later).
  • the conversion system referred to here is the conversion system itself that can perform conversion and inverse conversion such that the shape, structure, pattern design parameters, manufacturing conditions, etc. of the tire and the performance correspond one-to-one.
  • the experiment data input unit 40 is for inputting data on tire shape, structure, pattern design parameters and manufacturing conditions, etc., and the performance corresponding to them.
  • the shape, structure, pattern design parameters and manufacturing conditions such as vulcanization temperature, etc.
  • the above-mentioned optimization method is based on optimization methods such as mathematical programming and genetic algorithms. However, in the present embodiment, an optimization method based on mathematical programming is selected.
  • the optimization operation unit 34 is for optimizing the objective function until it converges, and includes an objective function / constraint condition operation unit 36 and an objective function optimization operation unit 38.
  • Objective functionConstraint condition calculation unit 36 predicts tire performance from shape, structure, pattern design parameters and manufacturing conditions using the conversion system by nonlinear calculation unit 32, and optimizes the objective function.
  • the operation section 38 is for optimizing the objective function input in the optimization item input section 42 until the convergence is achieved while satisfying the constraint conditions.
  • the optimization result output unit 44 performs optimization of the tire optimization parameters to satisfy the input optimization items, such as tire shape, structure, and pattern design parameters. This is for outputting manufacturing conditions.
  • the non-linear operation unit 32 is configured using the hardware resources shown in FIG. 2 and the software resources described later, and a conversion function configured as a conceptual neural network as described later. And a learning function to learn it.
  • the non-linear operation unit 32 can be configured to have only a conversion function without a learning function. That is, as will be described later, the non-linear calculation unit 32 obtains a conversion system in which the design parameters of the tire shape, structure, and pattern, and the manufacturing conditions are associated with the tire performance. It is only necessary to be able to convert between the design parameters and manufacturing conditions of the shape, structure, and pattern, and their performance.
  • the relationship between the design parameters of tire shape, structure, and pattern, and the correspondence between manufacturing conditions and its performance is learned in advance by another neural network, and the learned conversion coefficients of the other neural network are input.
  • this conversion coefficient a conversion system in which the design parameters of the tire, the design parameters of the pattern, and the manufacturing conditions are associated with the performance may be obtained.
  • this function is a conversion-only function that converts between the tire shape, structure and pattern design parameters and manufacturing conditions and the tire performance using the conversion coefficient.
  • these correspondences may be stored as a look-up table, and the conversion may be performed by referring to the stored look-up table.
  • the above-mentioned non-linear operation section 32 is used as an input layer to enable the input of each parameter of tire shape, structure, and pattern, and the design parameters of the manufacturing conditions. It has neurons according to the number of data and manufacturing conditions, and as an output layer via an intermediate layer, predicts the performance term related to the objective function and constraints.
  • the neural network has neurons corresponding to the number of eyes, and each neuron constitutes a neural network connected by synapses.
  • the known performance corresponding to the tire's shape, structure, and pattern design parameters and manufacturing conditions is input as a teacher, and the shape of the tire is determined by the size of the error difference between the output performance and the known performance.
  • the values are set so that the design parameters of the structure and pattern and the manufacturing conditions correspond to their performance.
  • An intermediate layer consisting of a number of units M l, M 2,..., M q (q> 1), and a predetermined number of output units U 1, U 2,..., U r (r > 1).
  • the number of units in the input layer and the number of units in the output layer may be set according to the design parameters of the shape, structure, and pattern of the tire, the number of manufacturing conditions, and the number of performances.
  • each unit of the intermediate layer and each unit of the output layer are connected to offset units 46 and 48 for offsetting the output value by a predetermined value.
  • the unit of the input layer for example, the width of the belt of the tire, the angle of the belt, the material of the belt, parameters representing the tire shape, and the cost can be used as input values.
  • the output layer unit for example, rolling resistance, stress / strain, tire panel characteristics, tire contact characteristics, and the like can be used as output values.
  • the unit of the intermediate layer and the unit of the output layer are constituted by neural circuit elements having a sigmoid characteristic whose input / output relationship is represented by a sigmoid function.
  • the relationship is composed of linear neural network elements.
  • the outputs of the unit in the intermediate layer and the unit in the output layer in the nonlinear operation unit 32 can be expressed by the following equations (1) and (2). That is, for a unit, the number of synapses on the input side is P, and the weight corresponding to the strength of each synaptic connection , Where W ji (1 ⁇ j ⁇ N, 1 ⁇ i ⁇ p) and each input signal is, a virtual internal state variable corresponding to the average value of the membrane potential of the neuron u can be expressed by the following equation (1), and the output y can be expressed by the following equation (2) using a nonlinear function f representing the characteristics of the neuron.
  • W ji represents the weight between the i-th and j-th units in different layers. Therefore, by inputting the values of the design parameters and the manufacturing conditions of the tire shape, structure, and pattern to the unit of the input layer, each value according to the number of tire performances is output from the unit of the output layer.
  • each unit in the input layer described above may be characteristics that output the input as it is.
  • the weight (coupling coefficient) of each unit of the non-linear operation unit 32 (neural network) is learned and corrected by a learning process described later so that the error is minimized for known experimental data.
  • data on tire performance is obtained by prototyping / evaluating a tire based on the design parameters of tire shape, structure, pattern, and manufacturing conditions, or by modeling and predicting the tire using a computer.
  • the correspondence between each value of the design parameters and manufacturing conditions of the tire shape, structure, and pattern, and each value representing its performance is used as learning data.
  • a predetermined number (for example, 90% of the total) of the plurality of data is set as learning data, and the other data (for example, the remaining 10%) is set as test data. This is because the experimental data is used for the data used for learning the neural network and for confirming whether the learned neural network has been optimally learned.
  • design parameters for the shape, structure and pattern of these tires The values of the meter and the manufacturing conditions are used as input data, and the values representing tire performance are used as output teacher data.
  • step 200 learning data and test data obtained in advance are read.
  • step 202 initialization is performed by setting the coupling coefficient (weight) and the offset value of each unit in the neural network to predetermined values.
  • step 204 in order to train the neural network using a plurality of learning data with known tire shape, structure, pattern design parameters and manufacturing conditions, the error of each unit in the intermediate layer and the output layer Ask for.
  • the difference between the learning data and the tire performance can be used as the error.
  • the output layer error that is, the unit error, can be minimized by slightly changing at least one of the coupling coefficient and the offset value.
  • the error in the intermediate layer can be obtained by using the error in the output layer and performing an inverse calculation such as an error back propagation method.
  • each of the coupling coefficients and the offset value obtained above is updated (rewritten), and in the next step 208, a test is performed by the neural network using the updated coupling coefficients and the offset values. Each of the data is tested to obtain data representing the performance of the tire as a value of the test result.
  • each coupling coefficient and offset value are determined so that the error of each unit of the intermediate layer and the output layer is minimized.
  • the neural network is trained using a plurality of experimental data with known tire shape, structure, pattern design parameters and manufacturing conditions. That is, learning is performed so that the error of the output value from the output layer of the neural network with respect to the teacher signal is minimized.
  • the nonlinear operation unit 32 outputs a value representing the performance of the tire.
  • the operation of the optimization device 30 of the present embodiment will be further described with reference to the flowchart of FIG.
  • next step 106 it is determined whether or not past experimental data can be used for the tire shape, structure, pattern design parameters Xi, and tire performance set in step 100, and used in the affirmative determination. If possible, proceed to step 108, and if a new determination is required with a negative judgment, proceed to step 120.
  • step 120 the tire shape, the tire shape, and the design parameter Xi are changed by using an orthogonal table or an optimal experiment plan to determine the tire shape.
  • the design parameters for the shape, structure, and pattern of this tire are determined in Box and Draper; "Empirical Mode l Builing and Response Surfaces", John Wiley & Sons, New York ".
  • a prototype or numerical model of the tire is created based on the design parameters of the tire shape, structure, and pattern according to the experimental plan determined in step 120, and an experiment or numerical analysis is performed to evaluate its performance. Get the data. Incidentally, the number of all experiments or numeric analysis at the time of this and n e.
  • the neural network is trained as described above. That is, the neural network learns the values input to the input layer as the values of the design parameters of the shape, structure, and pattern of the tire, and the values output from the output layer as the values of the performance of the tire.
  • the sensitivity indicating the tendency of the tire performance of the output layer to change when the design parameter X i of the shape, structure, and pattern of the tire input to at least one unit of the input layer is slightly changed, and Calculate the degree of decrease in the prediction accuracy of the tire performance of the output layer when the output from at least one unit is reduced to zero, and determine the design parameters of the shape, structure, and pattern of the tire that contribute little.
  • the tire shape, structure, and pattern design parameters which have low sensitivity and do not reduce the prediction accuracy even if the input is ignored, are considered to have little contribution.
  • step 126 If there are tire shape, structure, and pattern design parameters that make little contribution, an affirmative decision is made in step 126, and in the next step 128, the tire shape, structure, and pattern design parameters that make little contribution X i is deleted, and the tire is re-learned by the design parameters of the shape, structure, and pattern of the deleted tire (step 124).
  • step 124 On the other hand, if there is no design parameter for the shape, structure, or pattern of the tire that makes little contribution, a negative decision is made in step 126, and in the next step 130, the input layer (tire The relationship between the shape, structure, and pattern design parameters) and the output layer (tire performance) is stored. That is, each coupling coefficient and offset value are stored.
  • the stored input layer (tire shape, structure, pattern
  • the optimal tire shape, structure, and pattern design parameters X i are obtained by optimizing the objective function as described later using the relationship between the design parameters of the tire and the output layer (tire performance) (Fig. 7). ).
  • step 138 the variable e is incremented.
  • step 140 the allowable range of the tire shape, structure, and pattern design parameters is reset as shown in the following equations (4) to (6). Return to step 1 20.
  • the shape of the optimal tire structure can it to improve the accuracy of the design parameters Ichita X i 0 ft patterns.
  • the allowable range is reset by narrowing the allowable range of the design parameters of the tire shape, structure, and pattern determined in step 102, and in step 120, the re-experiment point of the narrowed region is determined. Make a plan.
  • NN is a coefficient for determining the degree to which the allowable range of the design parameters of the tire shape, structure and pattern is narrowed, and it is desirable to set a value of about 1.5 to about 5.
  • step 136 determines whether or not there is a similar experiment or numerical analysis in the past experimental data.
  • step 144 it is determined whether or not there is a similar experiment or numerical analysis in the past experimental data.
  • the performance of the optimal tire design is stored in the memory 22 or in the next step 146. External storage via the input / output device 28 Register in the database of the device. The performance of the tire may be obtained by an experiment or a numerical analysis again.
  • step 108 a past tire shape and structure related to each item set in step 100 are prepared from a database prepared in advance. Then, the pattern design parameters and tire performance are read, and in the next step 110, the read data is converted using Equations (7) to (10) below so that the kurtosis and skewness are reduced. I do.
  • next step 112 the neural network is learned in the same manner as in the above step 124, and in the next step 114, the learning result is stored in the same manner as in the above step 130.
  • next step 1 16 in order to return to the experimental data, the inverse conversion of the conversion in step 1 10 is performed.
  • the next step 1 18, the total number of experiments ne is reset ( 0), and the procedure goes to step 1 32. move on.
  • step 300 of FIG. 7 the objective function representing the tire performance to be improved, the constraints that restrict the tire performance that must not be degraded when a certain tire performance is improved, and the design parameters of the tire shape, structure, and pattern are specified.
  • design parameters of tire shape, structure, and pattern used as initial values for optimization are set.
  • the problem of optimizing the design of tire shape, structure and pattern is based on a 3D image obtained by plotting input values (eg, belt width and accuracy) on a 2D plane and plotting the value of the objective function in the height direction. From a perspective, it is necessary to know the solution space of the optimal solution by optimizing from different initial values because the design space related to tire performance has multi-modality.
  • the initial value for example, the following equation (11) can be used.
  • Mu nit Number of allowable divisions of design parameters for tire shape, structure, and pattern
  • the neural network is used with the initial tire shape, structure, and pattern design parameters set in step 304 as input. , And predicts the performance of the tire corresponding to the design parameters of the input tire shape, structure and pattern.
  • the initial values of the objective function and constraints are calculated using the results.
  • step 308 the tire shape, structure, and pattern design parameters X i set in step 304 to change the tire shape, structure, and pattern design parameters are changed by Axi.
  • step 310 the value of the objective function OB Ji and the value of the constraint condition Gi after changing the design variables by Axi are calculated, and in step 312, according to the following equations (1 2) and (1 3) ,
  • the sensitivity of the objective function which is the ratio of the change in the objective function to the unit change of the design variable, d 0 BJZ d X i
  • the sensitivity of the constraint which is the ratio of the change in the constraint to the unit change of the design variable d G / d X i is calculated for each design variable.
  • step 14 is to design parameters for all tire shapes, structures, and patterns.
  • steps 308 to 312 are repeated. .
  • Step 3 16 using the sensitivity of the objective function and the constraints on the design variables, the change of the design variables that minimizes (or maximizes) the objective function while satisfying the constraints is predicted by mathematical programming. . Step 3 1
  • step 8 the design parameters of the shape, structure, and pattern of each tire are corrected, and the objective function value based on the design parameters of the corrected shape, structure, and pattern of each tire is calculated.
  • step 320 the difference between the objective function value OB J calculated in step 3 18 and the initial value OB J o of the objective function calculated in step 306 is compared with a threshold value input in advance. Determine whether the value of the objective function has converged. If the value of the objective function has not converged, repeat steps 306 to 320 with the design variable values obtained in step 3 16 as the initial values. Execute.
  • the value of the design variable at this time is used as the design variable value that optimizes the objective function while satisfying the constraints, and the value of the design variable is used in step 322 by using the value of the design variable.
  • Design parameters for tire shape, structure and pattern are used in the next step 324, and the variable j is incremented, and the process proceeds to step 326.
  • step 32 it is determined whether or not j exceeds the allowable number of design parameters of the initial tire shape, structure, and pattern: (1 + M unit) p. If not, step 30 is executed. Returning to step 4, the values of the initial tire shape, structure, and pattern design parameters are changed, and steps 304 to 326 are repeated. On the other hand, if a positive determination is made in step 326, the optimal tire design is determined in the next step 328, and this routine ends. The determination of the optimum tire design in step 328 of the present embodiment is determined in consideration of the following two conditions, and the one that has a high degree of coincidence with the conditions is determined as the optimum tire design.
  • the objective function O B J has a small value.
  • the conversion system in order to determine the conversion system, the correspondence between the tire shape, structure, pattern design parameters, manufacturing conditions, and tire performance is determined in the non-linear operation unit using the neural network. Since learning is performed based on data obtained through experiments or numerical analysis, it is not necessary to assume a functional type as a means for calculating the conversion system, and the design parameters of tire shape, structure, pattern, and manufacturing conditions It is possible to create a conversion system that can find a mutual relationship in which the correspondence with the performance of the tire is associated with high accuracy and low arbitrariness. In addition, by combining the conversion system and the optimization calculation unit, it is possible to output an effective design proposal of the effective shape, structure, and pattern of the tire.
  • FIG. 10 shows a processing routine of the optimization processing program according to the present embodiment.
  • step 132 in FIG. 5 the processing routine shown in FIG. 10 is executed.
  • Modeling N means generating N inputs I1 to IP to be input to the input layer of the neural network shown in FIG. 4 based on random numbers. Note that N is input by the user in advance.
  • an objective function and constraints are determined.
  • an objective function that represents the tire performance that is desired to be improved or newly desired, and design variables that determine constraints that restrict tire performance that must not be degraded when a certain tire performance is improved are determined.
  • design variables that determine constraints that restrict tire performance that must not be degraded when a certain tire performance is improved are determined.
  • OB J and the constraint G are determined.
  • the adaptive function Fj of each of the N models is calculated according to the following equation (14) using the objective function OBJj and the constraint conditions of each of the N models obtained in step 404. .
  • the value of the adaptation function (fitness) increases as the value of the re-constraint condition Gj, in which the value of the objective function OB Jj is large, decreases.
  • m in mi ⁇ ( ⁇ i, ⁇ 2 , ⁇ ⁇ )
  • two models to be crossed are selected from the N models.
  • a selection method a generally known fitness proportional strategy is used, and the probability P L that an individual having N models is selected by selection is represented by the following equation.
  • the fitness proportional strategy is used as the selection method.
  • the expected value strategy, the rank strategy, the elite preservation strategy, the tournament selection, and the like as described in the genetic algorithm (edited by Hiroaki Kitano) A strategy or a GENITOR algorithm may be used.
  • next step 410 it is determined whether or not the two selected models are crossed according to the probability T input by the user in advance.
  • crossing refers to exchanging some of the elements of the two models, as described later. If no crossover is made in the negative judgment, the current two models are scanned as they are in Steps 4 and 12. Proceed to step 4 16. On the other hand, in the case of crossover with an affirmative determination, two models are crossed in step 414 as described later.
  • the two models selected in step 408 of FIG. 10 are referred to as a model a and a model b, and the design variables of the respective models a and b are represented by design variable vectors including a list.
  • the variable vector V r a (ri a, r 2 a, ⁇ ⁇ -, ri a, ⁇ ⁇ ⁇ , r ft - t a)
  • V r b (ri, r 2 b , ⁇ ri b , ⁇ r n —! b ).
  • step 450 of FIG. 11 a predetermined random number is generated, and a crossover position i of the design variable vectors of the models a and b is determined according to the random number.
  • the distance d is calculated according to the following equation for the design variables ri a and ri b of the models a and b determined to cross.
  • ri a using the possible range minimum value B L and the maximum value B u of ri b, determine the normalized distance d 'according to the following equation.
  • the design variable vector V r 'a, V r' is a sequence of new design variables in Step 460 b is prompted as follows.
  • V r '(r, rz • ⁇ , ⁇ n - 1 3) The minimum value B L and the maximum value B u of possible range of the user previously inputted.
  • the mapping function Z (X) may be a valley-shaped function as shown in Figs. 13 (a) and (b).
  • the number of the crossover points i is one.
  • a multipoint crossover or a uniform crossover as shown in a genetic algorithm may be used.
  • step 416 of FIG. 10 it is determined whether or not to mutate at the probability S input by the user in advance. As will be described later, this mutation involves changing a part of the design variables minutely, and is intended to increase the accuracy of including a population that can be an optimal design variable. If it is determined in step 416 that the mutation is not to be performed in the negative judgment, in step 426, the process proceeds to the next step 422 with the current two models. When the mutation is made in the affirmative judgment, the mutation processing is performed in the following step 420 as follows.
  • step 462 a random number is generated, and the location i of the mutation is determined by the random number.
  • step 464 the distance d '
  • a new design variable ri ′ is found according to the following equation.
  • the design variable vector Vr' which is a sequence of new design variables, obtained in step 470 is as follows.
  • V r ' (rj, r 2 , ri, r i + i,', r n _i)
  • step 422 The value of the objective function and the value of the constraint condition are calculated in step 422 in FIG. 10 for the two newly generated models.
  • step 424 an adaptive function is calculated from the obtained value of the objective function and the value of the constraint condition by using the equation (14) in the same manner as described above.
  • next step 426 the above two models are saved.
  • next step 428 it is determined whether or not the number of models saved in step 426 has reached N, and if not, steps 408 to 428 are repeated until the number of models has reached N I do. On the other hand, if the number of models has reached N, the convergence is determined in step 430. If the number of models has not converged, the N models are updated to the models saved in step 426. Repeat steps 408 to 430. On the other hand, if it is determined in step 430 that convergence has occurred, the N models In Step 432, the value of the design variable that maximizes the objective function while substantially satisfying the constraints is defined as the value of the design variable of the model that maximizes the value of the objective function while substantially satisfying the constraints. Then, the optimal tire design is determined in the same manner as in the above embodiment, and the routine ends.
  • the convergence determination in step 430 is regarded as convergence if any of the following conditions is satisfied.
  • the user inputs M, q, and s in advance.
  • a first test example will be described.
  • the present invention was applied to a tire crown shape design.
  • a tire (size: 205-55R16) was designed under the following conditions (1) to (6), the designed tire was actually created, and various tests were performed on the created tire.
  • the objective function is calculated by the following equation.
  • Pi N Contact pressure distribution when load is applied (area immediately below load)
  • Pi L Contact pressure distribution when lateral force is applied (area immediately below load)
  • ct Weighting factor for each objective function
  • the design variables are, r 2 , r 3 , ⁇ 1 , ⁇ 2 .
  • the above ground pressures Pi N and Pi L are assumed to be areas where the tire near the ground contact surface is divided into 20 (assuming that the element immediately below the load is 20). For the contact pressure distribution.
  • the neural network consisted of 5 units for the input layer, 40 units for the middle layer, and 20 units for the output layer when a load was applied and a lateral force was applied.
  • Table 6 shows the design variables to be input
  • Table 7 shows that the output is the contact pressure distribution under load and lateral load.
  • the output layer of the neural network corresponds to the contact pressure distribution in each area (element) immediately below the load (see Fig. 17).
  • Design variable ⁇ ! ⁇ 3, ⁇ ,, ⁇ 2 is set as follows. In other words, for the shape of the crown portion, which is a design variable, the range of the previously specified crown portion is approximated by a plurality of circular arcs. For example, as shown in Fig.
  • the range of the approximable crown from the node q7 at the center of the tire to the node q8 near the belt end in this case, the ranges CR1, CR2, CR
  • the crown shape is represented by three arcs in three regions.
  • the radius can be approximated by r1 and the angle ⁇ 1.
  • the radius can be approximated by r2 and the angle ⁇ 2.
  • the range CR3, half The diameter can be approximated by r 3 and the angle ⁇ 3.
  • the crown shape is represented by three arcs in the range from the node q7 at the center of the tire to the node q8 near the end of the belt, as well as the force of the arc.
  • the number may be 1-2 or 4 or more.
  • each arc need not necessarily be smoothly continuous, and the coordinate value of the center of each arc may be treated as an independent variable.
  • Table 8 shows a prototype of a tire designed under the conditions of this test example, and the load expressed by OBJl and OBJ2 and the degree of uniformity under lateral force load were determined by actual measurement. The results of comparison between the prototype tire and the tire according to this test example are shown. Table 8 shows that the smaller the value, the more uniform the contact pressure.
  • Table 9 shows a trial production of a tire designed under the conditions of this test example, testing of steering stability and abrasion, and comparison results between a tire prototyped by the conventional design method and the tire of this test example. . Table 5 shows that the larger the value, the higher the performance.
  • a second test example will be described.
  • the present invention was applied to a tire side shape design.
  • a tire (size: 205 / 55R16) was designed under the following conditions 1 to ⁇ , the designed tire was actually created, and various tests were performed on the created tire. .
  • the objective function is the same as the crown shape design (described above).
  • the input layer has 5 units and the hidden layer has 40 units.
  • the output layer consisted of 20 units.
  • Table 1 shows orthogonal table used in the 3-level orthogonal experimental design of L 27 to (L 27 3 13).
  • Table 2 shows the design variables to be input, and
  • Table 3 shows that the output is the ground pressure distribution under lateral load.
  • the design variable ⁇ r 5 is set as follows.
  • a tire cross-sectional shape in a natural equilibrium state is modeled as a reference shape.
  • Fig. 15 shows the modeled tire cross-sectional shape model, where CL is the force line, 0L is the line representing the outer shape of the tire, PL is the folded ply line, and Bl and B2 are the lines representing the belt. Is shown. Also, the tire cross-sectional shape model, the force - Kasurain plurality of normal NLi of CL, NL 2, NL 3, is divided into a plurality of elements by ... '.
  • the tire cross-sectional shape model was divided into a plurality of elements by a plurality of normals of force-cass lines, but a plurality of normals of a line representing the tire outer shape and a plurality of folded ply lines were described. It may be divided into a plurality of elements by a normal line, or may be divided into an arbitrary shape such as a triangle depending on the design purpose.
  • the shape of the carcass line is used as a design variable and is determined by Lagrange interpolation that approximates a curve.
  • a reference point P is set in advance in the tire, and a range from a node near the belt end to a node q 2 near the rim is specified as a range for changing the tire shape.
  • the normals n 1 j, n 1 2 , n 13, ... that are closest to the imaginary line are selected, and the selected normal n 1 j, n 12, n 13,-...
  • Table 4 shows a trial production of a tire designed under the conditions of this test example, a test of uniformity under a lateral force load, and a comparison between the tire prototyped by the conventional design method and the tire of this test example. The comparison results are shown. Table 4 shows that the smaller the value, the higher the performance.
  • Table 5 shows a trial production of a tire designed under the conditions of this test example, and a test of steering stability. The results of comparison between a tire prototyped by the conventional design method and a tire according to this test example are shown. Table 5 shows that the larger the value, the higher the performance.
  • a third test example will be described.
  • the present invention was applied to a tire belt structure design.
  • a tire (size: 11/70 R22.5) was designed under the following conditions 1 to 6, the designed tire was actually created, and various tests were performed on the created tire.
  • the design variables are, 12 , 13, 14, ⁇ ⁇ , ⁇ 2 , ⁇ 3 , ⁇ 4 .
  • Table 10 shows the design variables to be input, and Table 11 shows the output. Note that
  • the neural network consisted of 8 units for the input layer, 20 units for the hidden layer, and 4 units for the output layer.
  • the design variable ⁇ r 5 is set as follows.
  • the design variables are the lines B1, B2, B3
  • Table 1 2 below, the prototype of a tire designed under the conditions of this test example, speed 60 KmZh with indoor drum test machine, pressure 8. 00 kgf ZCM 2, from the load 2725 kg slip angle of one degree The load is increased by 10% every 30 minutes, and the load at the time of the occurrence of a belt failure is indicated as an index with the control as 100 as the belt durability, and the tires prototyped by the conventional design method and the tires of this test example The results of comparison with are shown. In Table 12, the larger the value, the higher the performance. [Table 10] [Table 11] Input
  • the tire design method, the optimization analysis device, and the storage medium storing the optimization analysis program according to the present invention are useful for designing the shape, structure, and pattern of a tire for manufacturing a tire.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Automation & Control Theory (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Mechanical Engineering (AREA)
  • Tires In General (AREA)
  • Feedback Control In General (AREA)

Description

明細書 タィャの設計方法、 最適化解析装置及び最適化解析プログラムを記憶した記憶媒 体 技術分野
本発明は、 タイヤの設計方法、 最適化解析装置及び最適化解析プログラムを記 憶した記憶媒体にかかり、 特に、 タイヤの構造、 形状、 パターン設計、 例えばタ ィャのサイドウオールの形状や、 クラウン部の形状設計に用いることができるタ ィャの設計方法、 最適化解析装置及び最適化解析プログラムを記憶した記憶媒体 に関する。 背景技術
タイヤの構造、 形状、 パターン設計は、 目的とするタイヤの性能を得るための 構造、 形状、 パターンや製造条件等を求めることである。 このタイヤの性能とは 計算や実験によって求められる物理量または実車フィ一リング評価結果である。 従来のタイヤの構造、 形状、 及びパターン設計等のタイヤ設計方法は、 実験及び 計算機を用いた数値実験の繰リ返しによる試行錯誤的な経験則から成り立つてい た。 このため、 開発に必要な試作 '試験の件数が膨大なものとなり、 開発コスト がアップし、 開発期間もなかなか短縮できなかった。
これを解決する手段として数理計画法や、 遺伝的アルゴリズムを用いた最適化 法等のように、 最適解を求める技術が提案されてきた。 この数理計画法に関連す るものとして、 本出願人も既に出願済の国際公開番号: W 094/ 16877に記載され た設計方法を提案している。
最適解を求めることは山登りにたとえられる。 この時、 山の高度は性能に関係 しているので、 最適解は山の頂上に相当する。 目的関数が単純な場合にはその設 計空間 (山の形) は図 8に示すようなピークが一つの山型であるので、 数理計画 法をベースにした最適化手法で最適解を求めることができる。
しかしながら、 目的関数が複雑になると設計空間は図 9に示すように数多くの ピークを有することになるので、 数理計画法をベースにした最適化手法では最適 解を求めることができない。 これは、 数理計画法をべ一スにした最適化法では、 たまたま最初に到達したピークを最適解と認識してしまうためである。
この問題点を解決するために遺伝的ァルゴリズムが提案されてきたが、 遺伝的 アルゴリズムでは膨大な実験回数や計算時間を要し、 場合によっては計算が収束 しないこともあり、 実用上問題があった。 つまり、 目的関数が複雑になった場合 に、 限られた期間内に最適解を得ることが難しかった。 しかしながら、 従来の設計方法を用いたタイヤの設計開発はある性能について 目標値を定め、 この目標値をクリアすれば一応終了とされ、 与えられた資源でそ の最良の性能を得るものではなかった。 また、 二律背反する性能を設計するもの でなく、 そのべストな形状、 構造を決定するものでもなかった。 さらに、 従来の 設計方法では開発が試作 ·試験の試行錯誤の繰返しで行われるため、 非常に非効 率でコスト 'パフォーマンスは低かった。
本発明は、 上記事実を考慮して、 与えられた条件でタイヤのベストモードを設 計することができると共に、 タイヤの設計 · 開発を高効率化できるタイヤの設計 方法、 最適化解析装置及び最適化解析プログラムを記憶した記憶媒体を得ること が目的である。 発明の開示
上記目的を達成するために本発明者は種々検討を加えた結果、 異分野に利用さ れている 「高等動物の神経回路網を工学的にモデル化された非線形予測技術、 例 えばニューラル ·ネットワーク」 及び 「最適化設計手法」 をタイヤ設計と言う特 殊分野に応用することに着目し、 検討を試み、 具体的にタイヤの設計方法として 確立したものである。
詳細には、 本発明のタイヤの設計方法は、 ( a ) 内部構造を含むタイヤ断面形 状またはタイヤ構造を表すタイヤの設計パラメ一タと該タイヤの性能との非線形 な対応を関係付ける変換系を定めるステップ、 (b ) 前記タイヤの性能を表す目 的関数を定めると共に、 前記タイヤの性能及び前記タイヤの製造条件の少なくと も一方の許容範囲を制約する制約条件を定めるステップ、 (c ) 前記ステップ (a) で定めた変換系を用いて、 前記目的関数及び前記制約条件に基づいて目的関数の 最適値を与えるタイヤの設計パラメータを求めて該タイヤの設計パラメータに基 づいてタイヤを設計するステツプの各ステップを含んでいる。
タィャ性能、 例えば操縦安定性やベルト耐久力等の値はタイヤの設計パラメ一 タ、 例えば内部構造を含むタイヤ断面形状やタイヤの構造で定まる。 しかし、 タ ィャ断面形状やタイヤ構造の値を線形的に変化させてもタイヤ性能が線形に変化 しない場合が多い。 そこで、 本発明のステップ (a ) では内部構造を含むタイヤ 断面形状またはタイヤ構造を表すタイヤの設計パラメータと該タイヤの性能との 非線形な対応を関係付ける変換系を予め定めている。 この変換系は、 ニューラル ネットワーク等の神経回路網を工学的にモデル化した非線形予測技術を用いて定 めることができる。
次のステップ (b ) では、 タイヤ性能を表す目的関数を定めると共に、 タイヤ 性能及びタイヤの製造条件の少なくとも一方の許容範囲を制約する制約条件を定 める。 タイヤ性能を表す目的関数としては、 例えば操縦安定性を向上させるため の空気充填時のタイヤ周方向ベルト張力や横ばね定数、 直進もしくは横力時の接 地面内の接地特性等のタィャ性能の優劣を支配する物理量を使用することができ る。 タイヤ性能及びタイヤの設計パラメータの少なくとも一方の許容範囲を制約 する制約条件としては、 例えばタイヤ断面形状やタイヤ構造を制約する制約条件 としては、 力一カスラインのペリフェリ値の制約、 上下一次固有振動数の制約、 ベルト層の角度の制約、 ベルト層の幅、 タイヤ寸度、 パネ定数、 タイヤ変形量、 タイヤ重量、 応力、 歪、 歪エネルギー、 転がり抵抗の制約等がある。 なお、 目的 関数、 及び制約条件は、 上記の例に限られるものではなく、 タイヤ設計目的に応 じて種々のものを定めることができる。
そして、 次のステップ (c ) で、 ステップ (a)で定めた変換系を用いて、 目的関 数及び制約条件に基づいて目的関数の最適値を与えるタイヤの設計パラメータを 求めて該設計パラメ一タに基づいてタイヤを設計する。 これにより、 タイヤの設 計パラメータとタイヤの性能との非線形な対応を関係付ける変換系が定められ、 変換系によリ複数のタイヤの設計パラメータとその性能との対応が関連付けられ る相互の関係を見出すことができる。 従って、 目的関数の最適値を与えるタイヤ の設計パラメ一タを求めて該設計パラメ一タに基づいてタイヤを設計することで 、 高性能なタイヤの設計が可能となる。 このステップ (C ) では、 制約条件を考 慮しながら目的関数の最適値を与える設計変数の値を求めることができる。 このステップ (C ) でタイヤを設計する場合、 タイヤの設計パラメータを設計 変数と定め、 制約条件を考慮しながらステップ (a)で定めた変換系を用いて目的関 数の最適値を与える設計変数の値を求め、 目的関数の最適値を与える設計変数に 基づいてタイヤを設計することができる。 このように、 制約条件を考慮すること によリ、 タイヤ性能及びタイヤの設計パラメータの少なくとも一方の許容範囲を 考慮することができ、 設計範囲を予め特定したり、 所望の範囲を設定できる。 また、 ステップ (c ) で設計変数の値を求める場合には、 設計変数の単位変化 量に対する目的関数の変化量の割合である目的関数の感度及び設計変数の単位変 化量に対する制約条件の変化量の割合である制約条件の感度に基づいて制約条件 を考慮しながら目的関数の最適値を与える設計変数の変化量を予測すると共に、 設計変数を予測量に相当する量変化させたときの目的関数の値及び設計変数を予 測量に相当する量変化させたときの制約条件の値を演算し、 予測値と演算値とに 基づいて、 制約条件を考慮しながらステップ (a)で定めた変換系を用いて目的関数 の最適値を与える設計変数の値を求めると効果的である。 これによつて、 制約条 件を考慮し目的関数の値が最適になるときの設計変数の値が求められる。 そして 、 目的関数の最適値を与える設計変数に基づいてタイヤの設計パラメータ等を変 更することによりタイヤを設計できる。
ここで、 上記で説明したように、 一般的な最適化手法において最適解を求める ことは、 山登りにたとえられることが知られている。 このとき、 山の高度が性能 等に関係するとすると、 最適解は山の頂上に相当する。 従って、 目的関数が単純 な場合にはその設計空間 (山の形) は図 8に示すような山型になるので、 数理計 画法をベースにした最適化手法で最適解を求めることができる。 そこで、 最適化 を山登りに喩えた模式的な図 8をモデル (変換系) としてタイヤの最適設計を概 略説明すると、 変換系はタイヤの設計パラメータと該タイヤの性能との非線形な 対応が関係付けられるものである。 この変換系は、 設計空間 (山の形) において 同一レベル (等高線) により示される。 すなわち、 タイヤ性能には種々の設計パ ラメータが関係し、 タイヤ性能が最適になるに従って、 等高線のように設計パラ メータの範囲が狭範囲化することが一般的である。 また、 タイヤの設計パラメ一 タの範囲には、 設計上の制約や実際に取り得る範囲があることが一般的であり、 目的関数であるタイヤ性能とタイヤの設計パラメータとの関係は、 図 8に示すよ うに山の稜線に沿った柵 (フェンス) によって制限できる。 これを制約条件とす れば、 等高線に示されたような形の山の上を、 変換系内の設計変数を変化させる ことにより、 制約条件のフェンスを越えないように、 目的関数が最適解を得る頂 上まで数理計画法等の最適化手法の助けを借りて、 図 8に示すような形の山を登 つて行くことに相当する。
そこで、 本発明では、 上記ステップ (a ) 〜 (c ) によって最適解を求めると き、 以下の (d ) 〜 ( f ) の各ステップが実行されて初めて最適解を得ることが できる。 詳細には、 ステップ (c ) は、 前記ステップ (a)で定めた変換系に含まれ るタイヤの設計パラメータの 1つを設計変数として選択するステップ (d ) と、 前記制約条件を考慮しながら前記ステップ (a)で定めた変換系を用いて目的関数の 最適値を与えるまで前記ステップ (a)で定めた変換系内から選択する設計変数の値 を変化させるステップ (e ) と、 目的関数の最適値を与える設計変数によるタイ ャの設計パラメータに基づいてタイヤを設計するステップ (f ) とを含んで構成 することができる。 ステップ (d ) では、 変換系に含まれるタイヤの設計パラメ —タの 1つを設計変数として選択する。 次のステップ (e ) では、 制約条件を考 慮しながら目的関数の最適値を与えるまで変換系内から、 選択する設計変数の値 を変化させる。 これにより、 設計変数の値は微妙に変化または徐々に変化し、 目 的関数の最適値が与えられる。 そして、 次のステップ ( f ) において目的関数の 最適値を与える設計変数によるタイヤの設計パラメータに基づいてタイヤを設計 する。 このように、 変換系に含まれるタイヤの設計パラメータの 1つを設計変数 として選択して制約条件を考慮しながら目的関数の最適値を与えるまで変換系内 から選択する設計変数の値を変化させているので、 目的関数の最適値を与える設 計変数の値そのものを準備することなく、 近傍の値を変換系内から選択すればよ く、 より高性能のタイヤの設計が可能となる。 この場合、 ステップ (b ) においては、 定めた目的関数以外のタイヤ性能及び タイヤの設計パラメータの少なくとも一方の許容範囲を制約する制約条件を定め ることができる。 このように、 定めた目的関数以外のタイヤ性能及びタイヤの設 計パラメータの少なくとも一方の許容範囲を制約する制約条件を定めることによ リ、 制約する許容範囲のうちタイャ性能としては目的関数と異なるタイャ性能が 用いられる。 前記制約条件を設定しない場合には、 目的関数以外のタイヤ性能や 設計パラメータが所望の範囲を逸脱し、 実用上の適用が困難となる場合が多い。 つまり、 ステップ ( a ) でタイヤ設計パラメータと目的関数として選んだタイ ャ性能との非線形な対応をニューラルネットワーク等によって関係付けた変換系 を図 8の等高線に示したが、 その等高線に示した形の山の上を、 ステップ (b ) において定めた制約条件をフェンスとして設け、 ステップ (c ) に含まれるステ ップ (d ) で前記ステップ (a)で定めた変換系内のタイヤ設計パラメータである選 択された設計変数を変更することにより、 ステップ (e ) で前記制約条件のフエ ンスを越えないように、 前記目的関数が最適解を得る頂上まで数理計画法や遺伝 的アルゴリズム等の最適化手法の助けを借りて、 前記形の山を登って行くのであ る。 加えて、 前記制約条件 (フェンス) は、 目的関数以外のタイヤ性能や設計パ ラメータの所望の範囲を予め設定する以外にも、 山登りの際のガイドとしても最 適化手法に有効である。 つまり、 制約条件なしでは、 計算時間の増加のみならず 、 計算自体が収束しない場合も発生する。 従って、 前記 ( a ) 〜 (e ) の各ステ ップが一体となり初めて最適解が得られる。
また、 ステップ (e ) では、 いわゆる数理計画法を応用し、 設計変数の単位変 化量に対する目的関数の変化量の割合である目的関数の感度及び設計変数の単位 変化量に対する制約条件の変化量の割合である制約条件の感度に基づいて制約条 件を考慮しながら目的関数の最適値を与える設計変数の変化量を予測すると共に 、 設計変数を予測量に相当する量変化させたときの目的関数の値及び設計変数を 予測量に相当する量変化させたときの制約条件の値を演算し、 予測値と演算値と に基づいて、 前記制約条件を考慮しながら前記ステップ (a)で定めた変換系を用い て目的関数の最適値を与えるまで、 選択する設計変数の値を変化させることがで きる。 このように、 設計変数を予測量に相当する量変化させたときの目的関数の 値及び設計変数を予測量に相当する量変化させたときの制約条件の値を演算する ことで、 目的関数の最適値を与えるまでの設計変数の値を容易に求めることがで きる。
また、 本発明者等は種々検討を加えた結果、 異分野に利用されている 「遺伝的 アルゴリズム手法」 をタイヤと言う特殊分野に応用することに着目し、 あらゆる 検討を試み、 具体的にそれをタイヤ設計方法として確立した。
具体的には、 本発明のタイヤの設計方法における、 前記ステップ (C ) で、 前 記ステップ (a)において定めた変換系におけるタイヤの設計パラメ一タを基本モデ ルとして複数個の基本モデルからなる選択対象集団を定め、 該選択対象集団の各 基本モデルについて、 前記目的関数、 設計変数、 制約条件、 及び目的関数から評 価できる適応関数を定め、 前記選択対象集団から 2つの基本モデルを選択し、 所 定の確率で各基本モデルの設計変数を交叉させて新規の基本モデルを生成するこ と及び少なくとも一方の基本モデルの設計変数の一部を変更させて新規の基本モ デルを生成することの少なくとも一方を行い、 設計変数を変化させて前記ステツ プ (a)で定めた変換系を用いて基本モデルの目的関数、 制約条件及び適応関数を求 めて該基本モデル及び設計変数を変化させなかった基本モデルを保存しかつ保存 した基本モデルが所定数になるまで繰リ返し、 保存した所定数の基本モデルから なる新規集団が所定の収束条件を満たすか否かを判断し、 収束条件を満たさない ときには該新規集団を前記選択対象集団として該選択対象集団が所定の収束条件 を満たすまで繰り返すと共に、 該所定の収束条件を満たしたときに保存した所定 数の基本モデルのなかで制約条件を考慮しながら前記ステップ (a)で定めた変換系 を用いて目的関数の最適値を与える設計変数によるタイヤの設計パラメータを求 めて該タイヤの設計パラメ一タに基づいてタイヤを設計する。
このステップ ( a ) においては、 タイヤの設計パラメータをタイヤの性能に変 換するように学習された多層フィードフォワード型ニューラルネットワークのデ —タで変換系を構成することができる。
すなわち、 上記で説明したように、 一般的な最適化手法としては、 数理計画法 や、 遺伝的アルゴリズム等があり、 最適解を求めることは山登りにたとえられる 。 この時、 山の高度は性能等に関係しているので、 最適解は山の頂上に相当する
。 目的関数が単純な場合にはその設計空間 (山の形) は図 1のような富士山型 ( ピークは一つ) であるので、 数理計画法をベースにした最適化手法で最適解を求 めることができる。 しかし、 目的関数が複雑になってくると設計空間は図 9に示 すように数多くのピークを有するので、 数理計画法をべ一スにした最適化手法で 最適解を求めることができない。 なぜなら、 数理計画法をベースにした最適化手 法では、 たまたま最初に到達したピークを最適解と認識するからである。 この問 題点を解決するために遺伝的アルゴリズムが提案されてきたが、 遺伝的アルゴリ ズムでは膨大な実験回数や計算時間を要し、 場合によっては計算が収束しないこ ともあった。
そこで、 ステップ ( a ) で、 タイヤの設計パラメ一タ (設計変数) と該タイヤ の性能との非線形な対応を関係付けた変換系をニューラルネットワークを用いて 定める。 タイヤの性能はタイヤの形状 ·構造 ·パターンに関する設計パラメータ をどう設計するかで定まる。 しかし、 設計パラメータを線形的に変化させてもタ ィャ性能が線形に変化しない場合が多い。 また、 ニューラルネットワークは, 豊 田秀樹著の 「非線形多変量解析—ニュ一ラルネットワークによるアプローチ—」 (朝倉書店 · 1 9 9 6発行) の P . 1 1〜 1 3、 P . 1 6 2〜 1 6 6にもあるよ うに、 線形変換の多変量解析より予測や判別の高い精度が期待でき、 入力したデ —タ一相互間の関係付けも学習することができるために、 中間層のュニットの数 さえ増やせば任意の関数を任意の精度で近似変換でき、 併せて外揷性に優れると いう利点がある。 そこで、 本発明ではその設計パラメータとタイヤ性能との非線 形な対応を含む対応を関連付けた変換系を予め定めている。 この変換系は、 ニュ —ラルネットワーク等の神経回路網を工学的にモデル化した非線形予測技術を用 いて定めることができる。 そこで、 ニューラルネットワークを応用し、 前記最適 化手法を組み合わせることによって、 目的関数が複雑になつた場合においても、 限られた期間内に最適解を得ることが可能となった。
本発明の設計方法に基づき設計 ·開発した場合、 従来の試行錯誤を基本とした 設計 · 開発と異なり、 コンピュータ一計算を主体にしてタイヤの性能が最良のタ ィャの設計から性能評価までが可能となり、 著しい効率化を達成でき、 開発にか かる費用が削減可能となる。
なお、 上記のタイヤの設計方法によリ設計されたタイヤの設計パラメータによ る構造等によってタイヤを形成することにより、 形成されたタイヤは、 性能が最 良の設計パラメ一タで構成されることになリ、 製造条件やコスト等の適用条件に よリ最適な設計パラメータの内容を直接的に決定することができる。
上記タイヤの設計方法は、 タィャの設計パラメ一タと該タイヤの性能との非線 形な対応関係を求める変換系計算手段と、 前記タイヤ性能を表す目的関数を定め ると共に、 前記タイヤ性能及び前記タイヤの製造条件の少なくとも一方の許容範 囲を制約する制約条件を定めて、 最適化項目として入力する入力手段と、 前記変 換系計算手段を用いて前記入力手段によリ入力された最適化項目に基づいて目的 関数の最適値を与えるタイャの設計パラメ一タを求める最適化計算手段とを備え た最適化解析装置によリ実現できる。
この変換系計算手段は、 前記タイヤの設計パラメ一タ及びタイヤに対する適用 条件と、 前記タイヤ性能との非線形な対応関係を求めることができる。 この適用 条件にはタイヤを形成するときの製造条件やタイヤ重量または総合的なコストが ある。 また、 変換系計算手段は、 前記タイヤの設計変数を、 前記タイヤの性能に 変換するように学習された多層フィードフォワード型ニューラルネットワークで 構成することができる。
また、 前記最適化計算手段は、 前記変換系計算手段に含まれるタイヤの設計パ ラメ一タの 1つを設計変数として選択する選択手段と、 前記制約条件を考慮しな がら目的関数の最適値を与えるまで前記変換系計算手段内から選択する設計変数 の値を変化させる変化手段と、 前記変換系計算手段を用いて目的関数の最適値が 与えられるまで設計変数の値を計算する最適値計算手段と、 目的関数の最適値を 与える設計変数によるタイヤの設計パラメータに基づいてタイヤを設計する設計 手段とから構成することができる。
また、 前記最適化計算手段は、 前記変換系計算手段において求めた対応関係に おけるタイヤの設計パラメータを基本モデルとして複数個の基本モデルからなる 選択対象集団を定め、 該選択対象集団の各基本モデルについて、 前記目的関数、 設計変数、 制約条件、 及び目的関数から評価できる適応関数を定め、 前記選択対 象集団から 2つの基本モデルを選択し、 所定の確率で各基本モデルの設計変数を 交叉させて新規の基本モデルを生成すること及び少なくとも一方の基本モデルの 設計変数の一部を変更させて新規の基本モデルを生成することの少なくとも一方 を行い、 設計変数を変化させて前記変換系計算手段を用いて基本モデルの目的関 数、 制約条件及び適応関数を求めて該基本モデル及び設計変数を変化させなかつ た基本モデルを保存しかつ保存した基本モデルが所定数になるまで繰り返し、 保 存した所定数の基本モデルからなる新規集団が所定の収束条件を満たすか否かを 判断し、 収束条件を満たさないときには該新規集団を前記選択対象集団として該 選択対象集団が所定の収束条件を満たすまで繰り返すと共に、 該所定の収束条件 を満たしたときに保存した所定数の基本モデルのなかで制約条件を考慮しながら 前記変換系計算手段を用いて目的関数の最適値を与える設計変数によるタイヤの 設計パラメータを求めて該タイヤの設計パラメータに基づいてタイヤを設計する ことができる。
また、 前記変換系計算手段は、 前記タイヤの設計パラメ一タを、 前記タイヤ性 能に変換するように学習された多層フィ一ドフォワード型ニューラルネットヮー クで構成することができる。
上記タイヤの設計方法は、 以下の手順によるプログラムを含んだ記憶媒体によ リ容易に持ち運びが可能な最適化解析プログラムを記憶した記憶媒体を提供でき る。 すなわち、 コンピュータによってタイヤを設計するためのタイヤの最適化解 析プログラムを記録した記録媒体は、 その最適化解析プログラムとして、 タイヤ の設計パラメ一タと該タイヤの性能との非線形な対応関係を定め、 前記タイヤの 性能を表す目的関数を定めると共に、 前記タイヤの性能及び前記タイヤの製造条 件の少なくとも一方の許容範囲を制約する制約条件を定め、 前記定めた対応関係 、 前記目的関数及び前記制約条件に基づいて目的関数の最適値を与えるタイヤの 設計パラメ一タを求めて該タイヤの設計パラメータに基づいてタイヤを設計する ことを特徴とするタイヤの最適化解析プログラムを記録している。
このタイヤの設計パラメータに基づくタィャの設計は、 前記定めた対応関係、 前記目的関数及び前記制約条件に基づいて、 前記定めた対応関係に含まれるタイ ャの設計パラメータの 1つを設計変数として選択し、 前記制約条件を考慮しなが ら目的関数の最適値を与えるまで前記定めた対応関係内から選択する設計変数の 値を変化させ、 目的関数の最適値を与える設計変数によるタイヤの設計パラメ一 タに基づいてタイヤを設計することができる。
また、 制約条件は、 前記定めた目的関数以外のタイヤ性能及び前記タイヤの設 計パラメータの少なくとも一方の許容範囲を制約することができる。
また、 設計変数の変化は、 設計変数の単位変化量に対する目的関数の変化量の 割合である目的関数の感度及び設計変数の単位変化量に対する制約条件の変化量 の割合である制約条件の感度に基づいて制約条件を考慮しながら目的関数の最適 値を与える設計変数の変化量を予測すると共に、 設計変数を予測量に相当する量 変化させたときの目的関数の値及び設計変数を予測量に相当する量変化させたと きの制約条件の値を演算し、 予測値と演算値とに基づいて、 前記制約条件を考慮 しながら目的関数の最適値を与えるまで、 選択する設計変数の値を変化させるこ とができる。
また、 前記定めた対応関係におけるタイヤの設計パラメータを基本モデルとし て複数個の基本モデルからなる選択対象集団を定め、 該選択対象集団の各基本モ デルについて、 前記目的関数、 設計変数、 制約条件、 及び目的関数から評価でき る適応関数を定め、 前記選択対象集団から 2つの基本モデルを選択し、 所定の確 率で各基本モデルの設計変数を交叉させて新規の基本モデルを生成すること及び 少なくとも一方の基本モデルの設計変数の一部を変更させて新規の基本モデルを 生成することの少なくとも一方を行い、 設計変数を変化させた基本モデルの目的 関数、 制約条件及び適応関数を求めて該基本モデル及び設計変数を変化させなか つた基本モデルを保存しかつ保存した基本モデルが所定数になるまで繰リ返し、 保存した所定数の基本モデルからなる新規集団が所定の収束条件を満たすか否か を判断し、 収束条件を満たさないときには該新規集団を前記選択対象集団として 該選択対象集団が所定の収束条件を満たすまで繰り返すと共に、 該所定の収束条 件を満たしたときに保存した所定数の基本モデルのなかで制約条件を考慮しなが ら前記対応関係を用いて目的関数の最適値を与える設計変数によるタイヤの設計 パラメータを求めて該タイヤの設計パラメ一タに基づいてタイヤを設計すること ができる。 以上説明したように本発明によれば、 タイヤの形状 ·構造 ·パターンの設計パ ラメ一タと該タイヤの性能との非線形な対応を関係付けた変換系をニューラルネ ットワークを用いて定めているので、 変換系を精度がかつ高く任意性の少なく作 成することができる、 という効果がある。
また、 変換系を用いて目的関数の最適値を与える形状 ·構造 ·パターンの設計 パラメ一タを求めているので、 どんな複雑な目的関数に関する最適解をも短時間 で求めることができる、 という効果がある。 図面の簡単な説明
図 1は、 本実施の形態にかかる最適化装置の外観図である。
図 2は、 本実施の形態にかかる最適化装置の概略構成図である。
図 3は、 本実施の形態にかかる最適化装置の機能別概略プロック図である。 図 4は、 ニューラルネットワークの概念構成図である。
図 5は、 本実施 形態にかかる最適化装置の作動の流れを示すフローチャート である。
図 6は、 ニューラルネットワークの学習処理の流れを示すフローチャートであ る。
図 7は、 第 1実施の形態の最適化処理の流れを示すフローチャートである。 図 8は、 本発明の最適化を説明するためのィメージ概念図である。
図 9は、 本発明の最適化を説明するための他のィメ一ジ概念図である。
図 1 0は、 第 2実施の形態の最適化処理の流れを示すフローチャートである。 図 1 1は、 交叉処理の流れを示すフローチャートである。
図 1 2は、 山型写像関数を示す線図であり、 (a ) は連続的な山型写像関数を 示す線図、 (b ) は線型的な山型写像関数を示す線図である。
図 1 3は、 谷型写像関数を示す線図であり、 (a ) は連続的な谷型写像関数を 示す線図、 (b ) は線型的な谷型写像関数を示す線図である。
図 1 4は、 突然変異処理の流れを示すフローチャートである。 図 1 5は、 第 1試験例の設計変数を説明するためのイメージ図である。
図 1 6は、 第 2試験例の設計変数を説明するためのイメージ図である。
図 1 7は、 荷重直下のタイヤ要素を示す断面図である。
図 1 8は、 第 3試験例の設計変数を説明するためのイメージ図である。 発明を実施するための最良の形態
以下、 図面を参照して本発明の実施の形態を詳細に説明する。 本実施の形態は 、 最適なタイヤの設計パラメータを求める最適化装置に本発明を適用したもので ある。 第 1実施の形態の最適化装置では、 高等動物の神経回路網が工学的にモデ ル化された非線形予測技術である学習後のニューラル ·ネットワークを変換系と して最適化計算によって設計パラメ一タを求めている。
図 1には、 本発明である最適化の実施をするための最適化装置 3 0の概略を示 した。 この最適化装置 3 0は、 データ等を入力するためのキ一ボード 1 0、 予め 記憶されたプログラムに従って非線形化予測手法によるニューラルネットワーク を用いてタイヤの形状、 構造、 パターンの設計パラメータ等からその性能を予測 すると共に制約条件を満たしかつ目的関数を最適 (例えば、 最大または最小) に する設計変数を演算するコンピュータ本体 1 2、 及びコンピュータ本体 1 2の演 算結果等を表示する C R T 1 4から構成されている。
詳細には図 2に示すように、 最適化装置 3 0は、 マイクロコンピュータを含ん で構成されたコンピュータ本体 1 2、 データ入出力装置 2 8、 データやコマンド を入力するためのキ一ボード 1 0、 及びモニタ一 1 4から構成されている。 コン ピュータ本体 1 2は、 C P U 1 6、 R O M 1 8、 R AM 2 0、 変換系等 (詳細は 後述) を記憶するためのメモリ 2 2、 本体と他の装置との間でデータ等をやりと りするための入出力装置 (以下、 I Z Oという) 2 6及びこれらをデータやコマ ンドが入出力可能なように接続されたバス 2 4から構成されている。 なお、 R O M l 8には、 後述する処理プログラムが記憶されている。 なお、 データ入出力装 置 2 8は、 数値表現されたタイヤの形状、 構造、 パターンの設計パラメータ、 製 造条件、 タイヤ性能 (本実施の形態ではタイヤの形状、 構造、 パターン等) が外 部記憶手段に記憶されている場合に、 外部記憶手段から読み込むための装置であ リ、 キ一ボード 1 0を入力装置として用いる場合には不要である。
図 3は本実施の形態の最適化装置 3 0の機能別概略構成を示すプロック図であ る。 本実施の形態の最適化装置 3 0では、 最大化もしくは最小化すべきタイヤ性 能 (これを目的関数という) を最適化してその最適化したタイヤ性能に対する設 計パラメータを出力する。
この最適化装置 3 0は、 機能別に、 非線形演算部 3 2、 最適化演算部 3 4、 実 験データ入力部 4 0、 最適化項目入力部 4 2及び最適化結果出力部 4 4に分類さ れる。 非線形演算部 3 2は、 ニューラルネットワークで構成された (詳細後述) 変換系の計算部として機能し、 実験デ一タ入力部 4 0から入力されたデータに基 づいて、 タイヤの形状、 構造、 パターン、 製造条件とその性能とが関連付けられ た変換系を求めるためのものである。 なお、 ここでいう変換系とは、 タイヤの形 状、 構造、 パターンの設計パラメータ及び製造条件等とその性能とが 1対 1に対 応するように変換及び逆変換が可能な変換系そのものをいい、 学習後の二ユーラ ルネットワークを数式で表現するときは数式及びその係数を含めたものをいう。 実験データ入力部 4 0は、 タイヤの形状、 構造、 パターンの設計パラメータ及び 製造条件等と、 それらに対応する性能との各データを入力するためのものである 最適化項目入力部 4 2は、 ①最大化もしくは最小化すべきタイヤの予測もしく は計測される物理量等タイヤ性能 (後述の目的関数) 、 ②最大化、 もしくは最小 化する時に制約を設けるタイヤの予測もしくは計測される物理量、 及びタイヤの 形状 ·構造 ·パターンの設計パラメータ並びに加硫温度や等の製造条件、 ③タイ ャの形状 ·構造 ·パターンの設計パラメータ及び製造条件のとりうる範囲、 ④最 適化に関する方法の選択及びその時のパラメ一タ等を入力するためのものである なお、 上記の最適化に関する方法は、 数理計画法及び遺伝的アルゴリズム等の 最適化手法があるが、 本実施の形態では数理計画法による最適化手法を選択する ものとする。
最適化演算部 3 4は、 目的関数を収束するまで最適化するためのものであり、 目的関数 ·制約条件演算部 3 6及び目的関数最適化演算部 3 8から構成される。 目的関数 ·制約条件演算部 3 6は非線形演算部 3 2による変換系を用い の形状、 構造、 パターンの設計パラメータ及び製造条件からそのタイヤ性能を予 測するためのものであり、 目的関数最適化演算部 3 8は最適化項目入力部 4 2で 入力した目的関数を制約条件を満たしつつ収束するまで最適化するためのもので ある。
最適化結果出力部 4 4は、 最適化演算部 3 4による最適化の結果として、 入力 された最適化項目を満足するように最適化された、 タイヤの形状、 構造、 パター ンの設計パラメータと製造条件を出力するためのものである。
なお、 本実施の形態では、 非線形演算部 3 2は、 図 2に示すハードウェア資源 及び後述するソフトウェア資源を用いて構成され、 後述するように概念的なニュ —ラルネットワークで構成された変換機能を有すると共に、 それを学習する学習 機能を有している。 また、 非線形演算部 3 2は、 学習機能を有さない変換機能の みを有する構成とすることも可能である。 すなわち、 後述するように、 非線形演 算部 3 2は、 タイヤの形状、 構造、 パターンの設計パラメ一タ及び製造条件とそ のタイヤ性能とが関連付けられた変換系を求めるものであるが、 タイヤの形状、 構造、 パターンの設計パラメ一タ及び製造条件とその性能との間で変換できれば よい。 従って、 タイヤの形状、 構造、 パターンの設計パラメ一タ及び製造条件と その性能との対応を予め他のニューラルネットワークで学習し、 学習された他の ニューラルネットワークの変換係数を入力するようにして、 この変換係数を用い てタイヤの形状、 構造、 パターンの設計パラメ一タ及び製造条件とその性能とが 関連付けられた変換系を求めてもよい。 つまり、 変換係数が入力される構成であ れば、 変換係数を用いてタイヤの形状、 構造、 パターンの設計パラメ一タ及び製 造条件とそのタイヤ性能との間で変換する変換のみの機能でよい。 また、 これら の対応をルックアツプテ一ブルとして記憶して、 記憶されたルックアップテ一ブ ルを参照することによって、 変換してもよい。
上記の非線形演算部 3 2は、 タイヤの形状 ·構造 ·パターンの設計パラメ一タ 及び製造条件の各値毎の入力を可能とするために入力層としてタイヤの形状 ·構 造 ·パターンの設計パラメ一タ及び製造条件の数に応じたニューロンを有し、 中 間層を介して出力層として目的関数や制約条件に関係した予測するタィャ性能項 目の数に応じたニューロンを有して各々のニューロンがシナプスによって結合さ れたニューラルネットワークを構成している。 この非線形演算部 32は、 後述する 学習後に、 タイヤの形状 ·構造 ·パターンの設計パラメータ及び製造条件の各値 が入力されると、 それに対応する性能が出力される。 学習時には、 タイヤの形状 -構造 ·パターンの設計パラメータ及び製造条件に対応する既知の性能が教師と して入力され、 出力の性能と既知の性能との誤差差分等の大小により、 タイヤの 形状 ·構造 ·パターンの設計パラメ一タ及び製造条件の各値と、 その性能とが対 応されるように設定される。
この非線形演算部 3 2に用いられているニューラルネットワークの一例として は、 図 4に示すように、 ニューロンに対応する所定数のユニット I 1 、 1 2、 · • · 、 I p ( p > 1 ) から成る入力層、 多数のユニット M l 、 M 2、 . . . 、 M q ( q > 1 ) から成る中間層、 及び所定数の出力ユニット U 1 、 U 2、 · · - 、 U r ( r > 1 ) から成る出力層から構成されている。 なお、 入力層のユニット数 、 及び出力層のュニット数はタイヤの形状 ·構造 ·パターンの設計パラメ一タ及 び製造条件個数、 性能の個数に応じて設定すればよい。 また、 中間層の各ュニッ ト及び出力層の各ュニットには出力値を所定値だけオフセットさせるためのオフ セットユニット 4 6、 4 8に接続されている。 上記入力層のユニットには例えば 、 タイヤのベルトの幅、 ベルトの角度、 ベルトの材質、 タイヤ形状をあらわすパ ラメ一タ、 そしてコストを入力値として用いることができる。 出力層のユニット には例えば、 転がり抵抗、 応力 ·歪み、 タイヤのパネ特性、 タイヤの接地特性等 を出力値として用いることができる。
なお、 本実施の形態では、 中間層のユニット及び出力層のユニットは入出力関 係がシグモイ ド関数によって表されるシグモイド特性を有する神経回路素子によ リ構成され、 入力層のュニットは入出力関係が線形の神経回路素子で構成されて いる。 このシグモイド特性を有するように構成することによって、 出力値は実値 (正の数) となる。
非線形演算部 3 2における、 中間層のュニット及び出力層のュニットの各々の 出力は、 次の ( 1 ) 、 (2 ) 式で表すことができる。 すなわち、 或るユニットに ついて、 入力側のシナプスの個数を Pとし、 各シナプス結合の強さに相当する重 み (ュニットの結合係数) を W j i ( 1≤ j ≤ N , 1≤ i≤p ) とし、 各入力信号 を とするとき、 ニューロンの膜電位の平均値に相当する仮想的な内部状態変 数 uは次の ( 1 ) で表すことができ、 出力 yはニューロンの特性を表す非線形関 数 f により次の (2 ) 式で表すことができる。 p
U =∑ w ji - X {+ b y · · · ( 1 )
y尸 f ( uゾ) · ' ·(2 ) 但し、 はオフセットユニットから供給されるオフセット値を表す。
W j iは異なる層の i番目と j番目のュニット間の重みを表す。 従って、 入力層のユニットへタイヤの形状、 構造、 パターンの設計パラメータ 及び製造条件の各値を入力することによって、 出力層のユニットから、 タイヤ性 能の個数に応じた各値が出力される。
なお、 上記の入力層の各ュニットの特性は入力をそのまま出力する特性でよい 。 また、 非線形演算部 3 2 (ニューラルネットワーク) の各ユニットの重み (結 合係数) は、 後述する学習処理により、 既知である実験データについて誤差が最 小となるように学習 ·修正される。
次に、 非線形演算部 3 2におけるニューラルネットワークの学習の処理の詳細 を図 6を参照して説明する。 本実施の形態では、 タイヤの形状、 構造、 パターン の設計パラメータ及び製造条件の各値によってタイヤを試作 ·評価、 またはタイ ャをモデル化しコンピュータで予測することによってタイヤの性能に関するデ一 タを得る。 次に、 タイヤの形状、 構造、 パターンの設計パラメータ及び製造条件 の各値と、 その性能を表す各値との対応をデータとして学習に用いる。 なお、 複 数のデータのうち所定数 (例えば、 全体の 9 0 % ) のデータを学習データとする と共に、 それ以外 (例えば、 残り 1 0 % ) のデータをテストデータとしている。 これは実験データを、 ニューラルネットワークの学習時に用いるデータと、 学習 が終了したニューラルネットワークが最適に学習がなされたかを確認するデ一タ とに用いるためである。 また、 これらタイヤの形状、 構造、 パターンの設計パラ メータ及び製造条件の各値を入力データとすると共に、 タイヤの性能を表す各値 を出力教師データとしている。
まず、 ステップ 2 0 0では、 予め求めた、 学習データ及びテストデータを読み 取る。 次のステップ 2 0 2では、 ニューラルネットワークにおける各ユニットの の結合係数 (重み) 及びオフセット値を予め定めた値に設定することによって初 期化する。 次のステップ 2 0 4では、 タイヤの形状、 構造、 パターンの設計パラ メータ及び製造条件が既知の複数の学習データを用いてニューラルネットワーク を学習させるため、 中間層及び出力層の各々のュニットの誤差を求める。
出力層の誤差は学習データのタイヤ性能に対する差を誤差とすることができる 。 各結合係数及びオフセット値の少なくとも 1つを僅かづつ変化させることによ つて出力層の誤差、 すなわちュニットの誤差が最小になるようにすることができ る。 また、 中間層の誤差は、 出力層の誤差を用いて誤差逆伝搬法等の逆計算によ リ求めることができる。
次のステップ 2 0 6では、 上記求めた各結合係数及びオフセット値を更新 (書 換え) して、 次のステップ 2 0 8においてその更新された各結合係数及びオフセ ット値によるニューラルネットワークによってテストデータの各々をテストして テスト結果の値としてタイヤの性能を表すデータを得る。 次のステップ 2 1 0で は、 上記ステップ 2 0 8で求めたテスト結果の値が収束判定の基準である所定範 囲内の値か否かを判別することによリ収束したか否かを判断するか、 または上記 の処理を所定回数繰り返ししたか否かを判断し、 肯定判断の場合には本ルーチン を終了する。 一方、 否定判断の場合にはステップ 2 0 4へ戻り、 上記処理を繰り 返す。 これによつて、 学習データを入力した場合に、 中間層及び出力層の各々の ュニットの誤差が最小になるように各結合係数及びオフセット値が定まる。 このようにして、 タイヤの形状、 構造、 パターンの設計パタメ一タ及び製造条 件が既知の複数の実験データを用いてニューラルネットワークを学習させる。 す なわち、 教師信号に対するニューラルネットワークの出力層からの出力値の誤差 が最小となるように学習される。 このように、 学習することによって非線形演算 部 3 2では、 タイヤの形状、 構造、 パターンの設計パラメータ及び製造条件の値 が入力されると、 タイヤの性能を表す値を出力することになる。 なお、 以上の処理が終了し、 ニューラルネットワークの学習が十分に行われた 後に、 ネットワークの構造、 すなわち結合係数やオフセット値をメモリ 1 8に記 憶し、 変換系を構築するようにしてもよい。
上記では非線形演算部 3 2としてニューラルネットワークを用いた場合を説明 したが、 次の (3) 式に示すように、 多項式による応答曲面法を利用した変換系 を用いることもできる。
P P P
y = a。+ aiXi十 b XiX · · · ( 3 )
ι=1 i=l =1 次に、 本実施の形態の最適化装置 30の作動を図 5のフローチャートを参照し てさらに説明する。 最適化装置 30の電源が投入または実行開始の指示がキ一ボ —ドよりなされると、 図 5のステップ 1 00へ進み、 タイヤの形状、 構造、 パタ ーンの設計パラメータ X i ( i = l〜p ) 、 目的関数、 最大実験回数を設定する 。 すなわち、 何れの性能を改良するか、 またその場合、 何回程度の実験回数まで に最適なタイヤの形状、 構造、 パターンの設計パラメータを決定したいかを設定 する。
次のステップ 1 02では、 ステップ 1 00で設定したタイヤの形状、 構造、 パ ターンの設計パラメータ X i の許容範囲を設定し (X i L ≤ X i ≤ X i ϋX i L は下限値、 X i u は上限値) 、 次のステップ 1 04では実験または数値計算に よる解析回数 M及びタイヤの形状、 構造、 パターンの設計パラメータの位置を表 す変数 eを初期化する (M= 0、 e = l ) 。
次のステップ 1 06では、 ステップ 1 00で設定したタイヤの形状、 構造、 パ ターンの設計パラメータ X i 、 タイヤ性能に関して、 過去の実験データを利用で きるか否かを判定し、 肯定判定で利用できるときはステップ 1 08へ進み、 否定 判定で新規に求めなければならないときはステップ 1 20へ進む。
ステップ 1 20では、 直交表または最適実験計画等を用いて、 何れのタイヤの 形状、 構造、 パターンの設計パラメ一タ X i を変化させて実験を行うかを決定す ることによってタイヤの形状、 構造、 パターンの設計パラメ一タパラメ一タを決 定する。 このタイヤの形状、 構造、 パターンの設計パラメータの決定は、 「Box and Draper ; " Emp i ri cal Mode l Bu i l i ng and Response Surfaces" , John Wi l ey & Sons, New York 」 に記載の方法を利用することができる。
次のステップ 1 2 2では、 ステップ 1 2 0で決定した実験計画に従ったタイヤ の形状、 構造、 パターンの設計パラメータによりタイヤを試作または数値モデル 化し、 その性能を評価する実験または数値解析を行い、 データを得る。 なお、 こ のときの全実験数または数値解析の回数を n e とする。
次のステップ 1 2 4では、 上記で説明したようにしてニューラルネットワーク を学習する。 すなわち、 入力層へ入力する値をタイヤの形状、 構造、 パターンの 設計パラメータの各値、 出力層から出力される値をタイヤの性能の各値として二 ュ一ラルネットヮ一クを学習する。
次のステップ 1 2 6では、 目標物性 ·特性に対して寄与が少ないタイヤの形状 、 構造、 パターンの設計パラメータの有無を判断する。 例えば、 入力層の少なく とも 1つのユニットへ入力したタイヤの形状、 構造、 パターンの設計パラメ一タ X i を僅か変化させたときに対する出力層のタイヤ性能の変化傾向を示す感度、 及び入力層の少なくとも 1つのュニットからの出力を零にしたときに対する、 出 力層のタイヤ性能の予測精度の低下度合を計算し、 寄与が少ないタイヤの形状、 構造、 パターンの設計パラメータを決定する。 これは感度が小さくその入力を無 視しても予測精度が低下しないタイヤの形状、 構造、 パターンの設計パラメータ は寄与が少ないと考えられるためである。
寄与が少ないタイヤの形状、 構造、 パターンの設計パラメータが有るときは、 ステップ 1 2 6で肯定判断され、 次のステップ 1 2 8において寄与が少ないタイ ャの形状、 構造、 パターンの設計パラメ一タ X i を削除し、 その削除された後の タイヤの形状、 構造、 パターンの設計パラメ一タによって再度学習する (ステツ プ 1 2 4 ) 。 一方、 寄与が少ないタイヤの形状、 構造、 パターンの設計パラメ一 タが無のときはステップ 1 2 6で否定判断され、 次のステップ 1 3 0において上 記学習されたニューラルネットワークの入力層 (タイヤの形状、 構造、 パターン の設計パラメ一タ) と出力層 (タイヤ性能) の関係を記憶する。 すなわち、 各結 合係数及びオフセット値を記憶する。
次のステップ 1 3 2では、 記憶された入力層 (タイヤの形状、 構造、 パターン の設計パラメータ) と出力層 (タイヤ性能) の関係を用いて後述するようにして 目的関数を最適化することによって最良のタイヤの形状、 構造、 パターンの設計 パラメ一タ X i を求める (図 7) 。
最適化が終了すると、 次のステップ 1 34で実験回数または解析回数 Mが増加 され (M = M+ne ) 、 次のステップ 1 36において、 M< (設定された最大の 実験回数または解析回数) か否かが判断され、 小さい場合には、 ステップ 1 38 へ進む。
ステップ 1 38では変数 eをインクリメントし、 次のステップ 140で、 以下 の (4) 〜 (6) 式に示すようにタイヤの形状、 構造、 パターンの設計パラメ一 タの許容範囲を再設定してステップ 1 20へ戻る。 この処理を繰り返すことで、 最適なタイヤの形状、 構造、 パターンの設計パラメ一タ X i 0FT の精度を向上す ることができる。 なお、 ステップ 140の許容範囲の再設定は、 ステップ 1 02 で定めたタイヤの形状、 構造、 パターンの設計パラメータの許容範囲を狭め設定 を行い、 ステップ 1 20ではこの狭めた領域について再実験点の計画を行う。
X Α Lne ≤ χ , ≤ χ i Unew . . , ( 4 )
x . U — x . L
X i Lnew = M i n X i L , i 0PT ··· (5)
1 NN J
X . U _ χ . L
X i Uftew = M a x χ ά u , X i 0PT ··· (6)
1 NN J ここで、 NNは、 タイヤの形状、 構造、 パターンの設計パラメータの許容範囲 を狭める程度を定めるための係数であり、 1. 5から 5程度の値を設定すること が望ましい。
一方、 ステップ 1 36で否定判断、 すなわち予め定めた最大の実験回数または 解析回数より多く実験または数値解析した場合には、 次のステップ 142で最後 に得られたタイヤの形状、 構造、 パターンの設計パラメ一タを最適タイヤ設計と して出力する。 次のステップ 144では、 過去の実験データ内に同様な実験また は数値解析があるか否かを判断し、 否定判断の場合には次のステップ 146で最 適タイヤ設計の性能をメモリ 22またはデ一タ入出力装置 28を介して外部記憶 装置等のデータベースへ登録する。 なお、 再度実験または数値解析してタイヤの 性能を求めてもよい。
なお、 最大の実験回数または解析回数は、 実験または数値解析にかかる費用及 び最適タイヤ設計を求めるのに用する時間等によって定められた定数である。 次に、 上記ステップ 1 06で肯定判断された場合には、 ステップ 1 08におい て、 予め用意されたデータべ一スからステップ 1 00で設定した各項目に関連し た過去のタイヤの形状、 構造、 パターンの設計パラメータ、 タイヤ性能を読み取 リ、 次のステップ 1 1 0において、 その読み取ったデータを以下の (7) 〜 ( 1 0) 式を用いて尖度、 歪度が小さくなるように変換する。
C i—β _
∑ 3 (7)
Figure imgf000024_0001
1 p
• (9)
Figure imgf000024_0002
次のステップ 1 1 2では、 上記ステップ 1 24と同様にニューラルネットヮ一 クを学習し、 次のステップ 1 1 4で上記ステップ 1 30と同様に学習結果を記憶 する。 次のステップ 1 1 6では実験データに戻すために、 ステップ 1 1 0による 変換の逆変換を行い、 次のステップ 1 1 8で全実験数 ne をリセットし (= 0) 、 ステップ 1 3 2へ進む。
次に、 図 5のステップ 1 3 2の最適化処理の詳細を説明する。 図 7のステップ 300では、 改良したいタイヤ性能を表す目的関数、 或るタイヤ性能を改良する ときに悪化してはならないタイヤ性能等を制約する制約条件及びタイヤの形状、 構造、 パターンの設計パラメータを決定する設計変数を定め、 次のステップ 3 0 2でタイヤの形状、 構造、 パターンの設計パラメータの数を表す変数 j をリセッ ト (= 0) する。 次のステップ 304では、 最適化するときの初期値として用いるタイヤの形状 、 構造、 パターンの設計パラメータを設定する。 タイヤの形状、 構造、 パターン の設計の最適化問題は、 入力値 (例えばベルトの幅と確度) を 2次元平面にプロ ットして目的関数の値を高さ方向にプロットしたイメージによる 3次元的に捉え るとタイヤの性能に関する設計空間が多峰性を有するために、 異なった初期値か ら最適化を行って最適解の解空間を知る必要がある。 初期値としては、 例えば、 以下に示す ( 1 1 ) 式を用いることができる。 x X
X i ; tart = X + k · · · ( 1 1 )
M u n l t 但し、 X i ( i = 1〜p ) : タイヤの形状、 構造、 パターンの設計パラメータ X i L ≤ X i ≤ X i u : タイヤの形状、 構造、 パターンの設計パラメ一タ の取りうる範囲
k = 0〜M u n i t
Mu n i t : タイヤの形状、 構造、 パターンの設計パラメータ の許容範囲の分割数 次のステップ 306では、 ステップ 304で設定した初期のタイヤの形状、 構 造、 パターンの設計パラメ一タを入力としてニューラルネットワークによる出力 を実行し、 入力したタイヤの形状、 構造、 パターンの設計パラメータに対応した タイヤの性能を予測する。 その結果を用いて、 目的関数、 制約条件の初期値を演 算する。
次のステップ 308では、 タイヤの形状、 構造、 パターンの設計パラメ一タを 変化させるためにステップ 304で設定されたタイヤの形状、 構造、 パターンの 設計パラメータ X i を各々 Axi づっ変化させて、 次のステップ 3 1 0で、 設計 変数を Axi 変化させた後の目的関数の値 OB Ji 及び制約条件の値 Gi を演算 し、 ステップ 3 1 2で以下の式 ( 1 2) 、 ( 1 3) に従って、 設計変数の単位変 化量に対する目的関数の変化量の割合である目的関数の感度 d 0 B J Z d X i 及 び設計変数の単位変化量に対する制約条件の変化量の割合である制約条件の感度 d G/d X i を各設計変数毎に演算する。 d 0 B J O B J £ 一 OB J o
d X i Δ Α
OB J ( X i + Δ X i ) -OB J ( X i )
( i +Δ Α ) - (χ )
一一 ( 1 2) d G Gi — G o
d X i △ X i ( 1 3) この感度によって、 設計変数を Δχι 変化させたときに目的関数の値がどの程 度変化するか予測することができる。 この予測、 すなわち、 最適化の過程は、 登 山にたとえることができ、 目的関数の値の変化を予測することは登山の方向を指 示することに相当する。
次のステップ 3 14では、 全てのタイヤの形状、 構造、 パターンの設計パラメ
—タについて演算が終了したか否かを判断し、 全てのタイヤの形状、 構造、 バタ ーンの設計パラメータについて演算が終了していない場合には、 ステップ 308 からステップ 3 1 2を繰り返し実行する。
次のステップ 3 1 6では、 目的関数、 制約条件の設計変数に関する感度を用い て、 数理計画法により制約条件を満たしながら目的関数を最小 (又は最大) にす る設計変数の変化量を予測する。 この設計変数の予測値を用いて、 ステップ 3 1
8で各タイヤの形状、 構造、 パターンの設計パラメータを修正すると共に、 修正 された各タイヤの形状、 構造、 パターンの設計パラメ一タによる目的関数値を演 算する。 次のステップ 320では、 ステップ 3 1 8で演算した目的関数値 OB J とステップ 306で演算した目的関数の初期値 OB J oとの差と、 予め入力され たしきい値とを比較することで目的関数の値が収束したか否かを判断し、 目的関 数の値が収束していない場合にはステップ 3 1 6で求められた設計変数値を初期 値として、 ステップ 306からステップ 320を繰り返し実行する。 目的関数の 値が収束したと判断されたときには、 このときの設計変数の値をもって制約条件 を満たしながら目的関数を最良にする設計変数の値とし、 ステップ 322におい てこの設計変数の値を用いてタイヤの形状、 構造、 パターンの設計パラメ一タを 決定し、 次のステップ 3 2 4で変数 j をインクリメントしてステップ 3 2 6へ進 む。
ステップ 3 2 6では jが初期のタイヤの形状、 構造、 パターンの設計パラメ一 タの許容数: ( 1 + M u n i t ) p を越えるか否かを判断し、 越えない場合には 、 ステップ 3 0 4へ戻り初期のタイヤの形状、 構造、 パターンの設計パラメータ の値を変更して上記ステップ 3 0 4からステップ 3 2 6を繰り返し実行する。 一方、 ステップ 3 2 6で肯定判断の場合には次のステップ 3 2 8で最適タイヤ 設計を決定し、 本ルーチンを終了する。 本実施の形態のステップ 3 2 8における 最適タイヤ設計の決定は、 次の 2つの条件を考慮して求めるものであり、 条件に 対する一致度が大きいものを最適タイヤ設計とする。
[条件]
①目的関数 O B Jが小さい値を有する。
(目的関数に選んだタイヤ性能が小さい方が良いように設定する。
大きい方が良い場合にはマイナス符号を付与して対応する。 )
②求められた最適解の周りでタイヤの形状、 構造、 パターンの設計パラメ一タを 少し変更しても目的関数、 制約条件が余リ変化しない。 以上説明したように、 本実施の形態では、 変換系を定めるために、 ニューラル ネットワークによる非線形演算部において、 タイヤの形状、 構造、 パターンの設 計パラメータ、 製造条件とタイャの性能との対応関係を実験または数値解析によ るデータによリ学習しているので、 変換系を計算する手段として関数型を仮定す る必要がなく、 タイヤの形状、 構造、 パターンの設計パラメ一タ及び製造条件と 、 タイヤの性能との対応が関連付けられた相互の関係を見出すことのできる変換 系を、 精度が高く任意性の少なく作成することができる。 また、 その変換系と最 適化演算部を組み合わせることによって、 有効性のあるタイヤの形状、 構造、 パ ターンの最適設計案を出力することができる。
次に、 第 2実施の形態を説明する。 本実施の形態は、 上記実施形態における感 度解析 (図 7 ) に代えて遺伝的アルゴリズムの手法によって、 最適化するもので ある。 なお、 本実施例は、 上記実施例と略同様の構成であるため、 同一部分には 同一符号を付して詳細な説明を省略する。
図 1 0は、 本実施の形態の最適化処理プログラムの処理ルーチンを示すもので ある。 上記図 5のステップ 1 32が実行されると、 図 1 0に示す処理ル一チンが 実行され、 ステップ 400では、 N個のタイヤについてモデル化がなされる。 つ まリタイヤの形状、 構造、 パターンの設計パラメータ Xij ( i = 1〜p , j = 1 〜N) と、 タイヤの性能との対応関係に基づいてモデル化する。 N個のモデル化 は、 図 4に示したニューラルネットワークの入力層へ入力されるべきィンプット I 1〜 I Pを乱数に基づいて N個生成することを言う。 なお、 Nは予め使用者が ィンプットする。
次のステップ 402では、 目的関数、 制約条件を決定する。 すなわち、 改良し たいまたは新規に望まれるタイヤ性能を表す目的関数、 或るタイヤ性能を改良す るときに悪化してはならないタイヤ性能等を制約する制約条件を決定する設計変 数を定める (目的関数 OB J、 制約条件 Gを決定) 。 次のステップ 404では、 N個のモデルの各々の設計変数 ri%1の各々の目的関数 OB J j 及び制約条件 G j を演算する。
次のステップ 406では、 ステップ 404で求めた N個のモデルの各々の目的 関数 OB J j 及び制約条件 を用いて、 N個のモデルの各々の適応関数 Fj を 以下の式 ( 14) に従って演算する。 本実施の形態では、 例えばタイヤ性能ゃコ ストを最適にするため、 適応関数による値 (適応度) は、 目的関数 OB Jj の値 が大きくなリ制約条件 Gj の値が小さくなると大きくなる。
Φ j = - 0 B J j + γ - ma (G j 、 0)
( 14)
または、
Figure imgf000028_0001
または、
F j =- a · + b
3ν9 ( c - 1 )
但し、 a
Figure imgf000028_0002
Φ a vg ( C— Φ,
b =
(Φ a vg min )
N
J=l
N c :定数
Ύ :ペナルティ係数
min =m i η (Φ i 、 Φ2 、 · · · ΦΝ )
j : Ν個のモデルの J番目のモデルのペナルティ関数
( J = 1、 2、 3、 , · · N)
なお、 c及び γは使用者が予め入力する。
次のステップ 408では、 N個のモデルの中から交叉させるモデルを 2個選択 する。 選択方法としては、 一般に知られている適応度比例戦略を用い、 N個のモ デルのある個体 が各々選択で選ばれる確率 PL は以下の式で表わされる。
FL
PL =
N
2: F
J=l
但し、 FL : N個のモデルの中のある個体^の適応関数
F j : N個のモデルの J番目の適応関数
J = l、 2、 3、 · · · Ν
本実施の形態では、 選択方法として適応度比例戦略を用いたが、 この他、 遺伝 的アルゴリズム (北野宏明 編) に示されている様な、 期待値戦略、 ランク戦略 、 エリート保存戦略、 トーナメント選択戦略、 あるいは GEN I TORアルゴリ ズム等を用いてもよい。
次のステップ 4 1 0では、 選択された 2個のモデルを、 使用者が予め入力した 確率 Tによって交叉させるか否かを決定する。 ここでいう、 交叉とは、 後述する ように、 2個のモデルの要素の一部を交換することをいう。 否定判定で交叉させ ない場合は、 ステップ 4 1 2において現在の 2個のモデルをそのままの状態でス テツプ 4 1 6へ進む。 一方、 肯定判定で交叉させる場合には、 ステップ 4 1 4に おいて後述するように 2個のモデルを交叉させる。
2個のモデルの交叉は、 図 1 1に示す交叉ル一チンによって行われる。 先ず、 図 1 0のステップ 408において選択された 2個のモデルをモデル a及びモデル bとすると共に、 各々のモデル a, bの設計変数について並びを含む設計変数べ クトルで表し、 モデル aの設計変数ベクトルを V r a = ( r i a 、 r2 a 、 · · -、 ri a 、 · · · 、 r ft— t a ) 、 モデル bの設計変数べクトルを V r b = ( r i , r2 b 、 · · · r i b 、 · · · rn— ! b ) とする。 図 1 1のステップ 45 0では、 予め定めた乱数を生成し、 この乱数に応じてモデル a, bの設計変数べ クトルに関する交叉場所 i を決定する。
次のステップ 452では、 交叉すると決定されたモデル a, bの設計変数 ri ar i b に対して、 以下の式に従って距離 dを求める。
d = I r a - r i b I
次のステップ 454では、 ri ar i b の取り得る範囲の最小値 B L 及び最 大値 Bu を用いて、 以下の式に従って正規化距離 d' を求める。
d
d, =
Bu 一 BL ステップ 456では、 正規化距離 d' の値を適度に分散させるために、 図 1 2 ( a) , (b) に示すような山型の写像関数 Z (x) (0≤x≤ 1 , 0≤Ζ (χ ) ≤ 0. 5) を用いて、 以下の式に従って関数値 Zabを求める。
Z ab Z (d' ) このようにして、 関数値 Zabを求めた後、 ステップ 458において新しい設計 変数 ri ' a 、 Γ ' b を次の式に従って求める。 m l n ( I 一 BL B u | ) r i r i Z ab
0. 5
m i n ( BL r i a - B u I ) r i = r + Z ab
0. 5
または、
m i n ( I r i BL r i B u i )
r i r i + Z ab
0. 5
m i n ( I r i b - BL |、 | r i a - B u | )
= r Z ab
0. 5
このようにして、 ri ' a 、 r i ' b を求めた後、 ステップ 460で新しい設 計変数の並びである設計変数ベクトル V r' a 、 V r' b は以下のように求めら れる。
V r ' ( r , r2 、 · · · r i , r i + j - ·、 rn.j b )
V r ' ( r , r z • · 、 Γ n- 1 3 ) なお、 の取り得る範囲の最小値 BL 及び最大値 B uは、 使用者が予め入力 しておく。 また、 写像関数 Z (X) は図 1 3 ( a) , ( b ) に示すような、 谷型 の関数でもよい。 また、 上記では交叉場所 iは 1ケ所であるが、 この他に遺伝的 アルゴリズム (北野 宏明 編) に示されているような、 複数点交叉または一様 交叉等を用いてもよい。
このような交叉によって新規な 2個のモデルを生成した後、 図 1 0のステップ 4 1 6では、 使用者が予め入力した確率 Sで、 突然変異させるか否かを決定する 。 この突然変異は、 後述するように、 設計変数の一部を微小に変更することをい い、 最適な設計変数となりうる母集団を含む確度を高くするためである。 ステツ プ 4 1 6で、 否定判定で突然変異させない場合には、 ステップ 426では現在の 2個のモデルのまま、 次のステップ 422へ進む。 肯定判定で突然変異させる場 合には、 次のステップ 420で以下のようにして突然変異処理を行う。
この突然変異は、 図 1 4に示す突然変異ルーチンによって行われる。 先ず、 ス テツプ 4 6 2では乱数を生成し、 乱数によって突然変異の場所 i を決定する。 次 のステップ 464では、 距離 d' を
0≤ d' ≤ 1
の範囲で乱数により決定する。
次のステップ 46 6では、 図 1 2 ( a) , (b) に示すような山型の写像関数 Z (x) (0≤x lで、 0≤Z (x) ≤ 0. 5) あるいは図 1 3 (a) , (b ) に示すような谷型の写像関数 Z ( X ) を用いて、 以下の式に従って、 関数値 Z dを求める。
Z d - Z ( d' )
このようにして、 関数値 Z dを求めた後、 ステップ 468において新しい設計 変数 ri ' を以下の式に従って求める。
m i n ( I r i — B L |、 r -B u I )
r i = r i Z d
0. 5
または、
m l n BL |、 I B u | )
r i = r j + Z d
0. 5
このようにして、 設計変数 ri ' を求めた後、 ステップ 470で求められる、 新しい設計変数の並びである設計変数べクトル V r ' は以下のようになる。
V r ' = ( r j , r2 、 · · · r i 、 ri + i 、 ' · ·、 rn_i )
このようにして、 新たに生成された 2個のモデルについて、 目的関数の値と制 約条件の値を図 1 0のステップ 422で演算する。 次のステップ 424では、 得 られた目的関数の値と制約条件の値から上記と同様に式 ( 1 4) を用いて適応関 数を演算する。
次のステップ 42 6では、 上記 2個のモデルを保存する。 次のステップ 428 では、 ステップ 426で保存したモデルの数が、 N個に達したか否かを判断し、 N個に達していない場合は、 N個になるまでステップ 408からステップ 428 を繰り返し実行する。 一方、 モデルの数が N個に達した場合には、 ステップ 43 0で収束判定をし、 収束していない場合には、 N個のモデルをステップ 42 6で 保存されたモデルに更新し、 ステップ 408からステップ 430を繰り返し実行 する。 一方、 ステップ 430で収束したと判断された場合には、 N個のモデルの 中で制約条件を略満たしながら目的関数の値が最大となるモデルの設計変数の値 をもって制約条件を略満たしながら目的関数を最大にする設計変数の値とし、 ス テツプ 432においてこの設計変数の値を用いて上記実施の形態と同様にして最 適タイヤ設計を決定し、 本ルーチンを終了する。
なお、 ステップ 430の収束判定は以下の条件のいずれかを満足したら収束と みなす。
1 ) 世代数が M個に達した
2 ) 一番目的関数の値が大き 、線列の数が全体の q %以上になつた
3) 最大の目的関数の値が、 続く s回の世代で更新されない。
なお、 M、 q、 sは使用者が予め入力しておく。
〔試験例〕
次に、 本発明の実施の形態による設計方法を用いて設計したタイヤの試験例を 説明する。
まず、 第 1試験例を説明する。 本試験例はタイヤのクラウン形状設計に本発明 を適用したものである。 本試験例は、 次の条件の①〜⑤によりタイヤ (サイズ: 205ノ 55 R 1 6) を設計し、 設計したタイヤを実際に作成し、 作成したタイ ャについて各種試験を行った。
〔条件〕
①過去のデータを利用しない。
② L 27の 3水準直交実験計画法を利用する。
(表 1の直交表— L 27313参照)
③実験のセット回数は 1回にする。
④目的関数 OB Jを最小化する。
(横力負荷時の接地圧を均一化する)
なお、 目的関数は次式で求める。
OB J =a - OB J l +^ - OB J 2 20
OBJ 1=∑ Pt -P
j=l 但し、 Pi N >0のときにのみ加算
20
◦ B J2 =∑ PiL— P 但し、 Pi L 〉0のときにのみ加算 但し、 P w :荷重を負荷したときの平均 ¾¾E
P L: を負荷したときの平均 ¾¾E
Pi N : 荷重を負荷したときの接地圧分布 (荷重直下の領域) Pi L : 横力を負荷したときの接地圧分布 (荷重直下の領域) ct、 β :各目的関数に対する重み係数
⑤設計変数は、 , r2 , r3 , θ 1 , θ2 とする。
( rはタイャの基準点とタイャ最内側の節点との距離)
なお、 上記接地圧 Pi N 、 Pi L は、 図 1 7に示すように、 接地面付近のタイ ャを 20個に分割した領域を想定し (荷重直下の要素を 20個として) 、 各々の 領域について接地圧分布求めるものとする。
また、 ニューラルネットワークは、 荷重付加時及び横力付加時のそれぞれにつ いて入力層が 5ユニット、 中間層が 40ユニット、 及び出力層が 20ユニットで 構成した。 以下の表 6には入力する設計変数を示し、 表 7には出力が荷重及び横力負荷時 の接地圧分布であることを示した。 なお、 ニューラルネットワークの出力層が荷 重直下の各領域 (要素) の接地圧分布に対応することになる (図 1 7参照) 。 設計変数 〜! ·3 , θ , , θ2 は、 以下のようにして設定する。 すなわち、 設計変数であるクラゥン部形状については、 予め指定したクラゥン部の範囲を複 数個の円弧で近似する。 例えば、 図 1 6に示すように、 タイヤ中心の節点 q 7か らベルト端部付近の節点 q 8までの範囲で、 近似可能なクラゥン部の範囲、 この 場合、 範囲 CR 1, C R 2 , CR 3の 3領域についてクラウン部形状を 3個の円 弧で表わす。 範囲 CR 1については、 半径が r 1で角度 θ 1で近似でき、 範囲 C R 2については、 半径が r 2で角度 Θ 2で近似でき、 範囲 CR 3については、 半 径が r 3で角度 θ 3で近似できる。 これらの半径 r l , r 2 , r 3及び角度 Θ 1 , θ 2を設計変数と設定する。
上記の図 1 6の例では、 タイヤ中心の節点 q 7からベルト端部付近の節点 q 8 までの範囲で、 クラウン部形状を 3個の円弧で表わすことを示している力 もち ろん円弧の数は 1〜 2個あるいは 4個以上でも構わない。 また、 各円弧は必ずし も滑らかに連続している必要はなく、 各円弧の中心の座標値を独立した変数とし て扱ってもよい。
なお、 上記ラグランジェ補間および円弧補間を用いたが、 スプライン曲線、 B スプライン曲線、 ベジエ曲線、 N U R B S等を用いてもよい。 以下の表 8には、 本試験例の条件で設計したタイヤの試作し、 O B J l , O B J 2で表される荷重及び横力負荷時の均一化度を実測により求め、 従来の設計方 法によって試作したタイヤと本試験例によるタイヤとの比較結果を示した。 なお 、 表 8では数値が小さい程接地圧が均一であることを示している。 また、 表 9に は、 本試験例の条件で設計したタイヤの試作し、 操縦安定性及び磨耗を試験し、 従来の設計方法によって試作したタイヤと本試験例によるタイヤとの比較結果を 示した。 なお、 表 5では数値が大きい程性能が高いことを示している。
:表 6〕 〔表 7〕 ィンプット
設計変数
1 r 1
2 r z
3 r 3
Figure imgf000035_0001
4 θ 1
5 θ 2 〔表 8〕
Figure imgf000036_0001
〔表 9〕 実車フィ一リング評価
Figure imgf000036_0002
次に、 第 2試験例を説明する。 第 2試験例はタイヤのサイド形状設計に本発明 を適用したものである。 本試験例は、 次の条件の①〜⑤によりタイヤ (サイズ: 2 0 5/ 5 5 R 1 6 ) を設計し、 設計したタイヤを実際に作成し、 作成したタイ ャについて各種試験を行った。
〔条件〕
①過去のデータを利用しない。
② L 2 7の 3水準直交実験計画法を利用する。
(表 1の直交表一 L 273 13参照)
③実験のセット回数は 1回にする。
④目的関数はクラウン形状設計 (前述) と同一にする。
(横力負荷時の接地圧を均一化する)
⑤設計変数は、 r , , r 2 , r 3 , r 4 , r 5 とする。
( rはタイヤの基準点とタイャ最内側の節点との距離)
なお、 ニューラルネットワークは、 入力層が 5ユニット、 中間層が 4 0ュニッ ト、 及び出力層が 20ユニットで構成した。 以下の表 1には、 L 27の 3水準直交実験計画法に用いる直交表 (L 27313) を示した。 表 2には入力する設計変数を示し、 表 3には出力が横力負荷時の接地 圧分布であることを示した。
なお、 設計変数 〜r5 は、 以下のようにして設定する。
まず、 自然平衡状態のタイヤ断面形状を基準形状としてモデル化する。 図 1 5 はモデル化したタイヤ断面形状モデルを示すもので、 CLは力一カスライン、 0 Lはタイヤ外面形状を表すライン、 PLは折り返しプライライン、 B l, B 2は ベルトを表すラインを各々示している。 また、 このタイヤ断面形状モデルは、 力 —カスライン C Lの複数の法線 NLi, NL2, NL3, · · ' によって複数の要素に 分割されている。 なお、 上記では、 タイヤ断面形状モデルを力一カスラインの複 数の法線によって複数の要素に分割した例について説明したが、 タイヤ外面形状 を表すラインの複数の法線や折り返しプライラインの複数の法線によって複数の 要素に分割してもよく、 また設計目的によって 3角形等の任意の形状に分割して もよい。
カーカスラインの形状は、 設計変数として用いられ、 曲線を近似するラグラン ジェ補間によって決定される。 このラグランジェ補間処理は、 タイヤ内部に予め 基準点 Pを設定し、 ベルト端付近の節点 からリムに拘束される付近の節点 q 2 までの範囲をタイヤ形状を変化させる範囲として指定する。 この節点 と基 準点 Pとを結ぶ直線を基準線として、 この基準線と、 節点 q2 と基準点 Pとを結 ぶ直線との成す角である見込み角 0を演算し、 角度増分 (= ΘΖラグランジ ェ補間の次数) を演算する。 次に、 基準線を基準として角度増分 d Θ毎に仮想線 を想定して仮想線に最も近い法線 n 1 j, n 12, n 13, · · · を選択し、 選択した 法線 n 1 j, n 12, n 13, - · ·上の最内側の節点 Q!, Q2, Q3, · · · と基準点 P との距離 rし r 2, r3, · · · (以下一般式で r i と表す。 ただし、 i = l, 2, • - ' ラグランジェ補間の次数一 1 ) 、 節点 Q^ Qz, Q3, · · ·の見込み角 0 θ 2, θ 3, ■ ■ ■ (ただし、 i = l, 2, · · ' ラグランジェ補間の次数一 1 ) と を演算し、 距離 ri を設計変数として設定する。 本試験例では設計変数 r , 〜 r 5 を設定した。
以下の表 4には、 本試験例の条件で設計したタイヤの試作し、 横力負荷時の均 一化度を試験し、 従来の設計方法によって試作したタイヤと本試験例によるタイ ャとの比較結果を示した。 なお、 表 4では数値が小さい程性能が高いことを示し ている。 また、 表 5には、 本試験例の条件で設計したタイヤの試作し、 操縦安定 性を試験し、 従来の設計方法によって試作したタイヤと本試験例によるタイヤと の比較結果を示した。 なお、 表 5では数値が大きい程性能が高いことを示してい る。
〔表 1〕 直交表 (L 27 3 1 3) l nul 変数
1 2 3 4 —— · 1 1 1 2 1 3
1 1 1 1 1 —— · 1 1 1
2 1 1 1 1 —— · 2 2 2
3 1 1 1 1 · · · 3 3 3
4 1 2 2 2 - - - 3 3 3
2 4 3 2 1 3 · · · 2 1 3
2 5 3 3 2 1 —— · 2 1 3
2 6 3 3 2 1 —— · 3 2 1
2 7 3 3 2 1 · ' · 1 3 2
〔表 2〕 〔表 3〕
Figure imgf000039_0002
Figure imgf000039_0001
〔表 4〕
Figure imgf000039_0003
次に、 第 3試験例を説明する。 本試験例はタイヤのベルト構造設計に本発明を 適用したものである。 本試験例は、 次の条件の①〜⑥によりタイヤ (サイズ: 1 1 /7 0 R22. 5) を設計し、 設計したタイヤを実際に作成し、 作成したタイ ャについて各種試験を行った。
〔条件〕
①過去のデータを利用しない。
② D最適実験計画 (実験点は 3 0) を利用する。 (" Optimum Experimental Designs" A. A. Atkinson and A. N. Donev oxford science Publications. ppl06 参照)
③実験のセット回数は 1回にする。
④目的関数は補強層間の主歪の最大値を最小にする。
⑤設計変数は、 , 12 , 13 , 14 , θ ι , θ 2 , θ 3 , θ 4 とする。
( 1 i :ベルト幅、 0i :周方向から計測した補強層のコード角度)
⑥制約条件は次の式によるものとする。
4
L min ― ^i = L max
i=l
Qmin < Q. <^ ff ax
以下の表 1 0には、 入力する設計変数を示し、 表 1 1には出力を示した。 なお
、 ニューラルネットワークは、 入力層が 8ユニット、 中間層が 20ユニット、 及 び出力層が 4ュニットで構成した。
なお、 設計変数 〜r5 は、 以下のようにして設定する。
図 1 8に示すように設計変数は、 各ベルト層を表わすライン B 1, B 2, B 3
, B 4の各々のベルト幅 l i 、 周方向から計測した補強層のコード角度 0 i を設 定した。
以下の表 1 2には、 本試験例の条件で設計したタイヤの試作し、 室内ドラム試 験機を用いて速度 60 kmZh, 内圧 8. 00 k g f Zcm2 , スリップ角 1度 で荷重 2725 k gから 30分おきに荷重を 1 0 %づつ増加させて、 ベルト部の 故障発生時の荷重をベルト耐久力としてコントロールを 1 00として指数表示し 、 従来の設計方法によって試作したタイヤと本試験例によるタイヤとの比較結果 を示した。 なお、 表 1 2では数値が大きい程性能が高いことを示している。 〔表 1 0〕 〔表 1 1〕 インプット
設計変数
1
2 £ 2
3 £ 3
4 £ A
Figure imgf000041_0001
5 θ 1
6 θ ζ
7
8 θ
〔表 1 2〕
Figure imgf000041_0002
産業上の利用可能性
以上のように本発明にかかるタイヤの設計方法、 最適化解析装置及び最適化解 析プログラムを記憶した記憶媒体は、 タイヤを製作するためのタイヤの形状、 構 造、 パターンの設計に有用である。

Claims

請求の範囲
1 . 次の各ステップを含むタイヤの設計方法。
( a ) 内部構造を含むタイヤ断面形状またはタイヤ構造を表すタイヤの設計パラ メータと該タイャの性能との非線形な対応を関係付ける変換系を定めるステップ
( b ) 前記タイヤの性能を表す目的関数を定めると共に、 前記タイヤの性能及び 前記タイヤの製造条件の少なくとも一方の許容範囲を制約する制約条件を定める ステツプ、
( c ) 前記ステップ (a)で定めた変換系を用いて、 前記目的関数及び前記制約条件 に基づいて目的関数の最適値を与えるタイヤの設計パラメータを求めて該タイヤ の設計パラメ一タに基づいてタイヤを設計するステツプ。
2 . 前記ステップ (c ) では、 前記タイヤの設計パラメ一タを設計変数と定め 、 制約条件を考慮しながら前記ステップ (a)で定めた変換系を用いて目的関数の最 適値を与える設計変数の値を求め、 目的関数の最適値を与える設計変数に基づい てタイヤを設計することを特徴とする請求項 1に記載のタイヤの設計方法。
3 . 前記ステップ (c ) では、 設計変数の単位変化量に対する目的関数の変化 量の割合である目的関数の感度及び設計変数の単位変化量に対する制約条件の変 化量の割合である制約条件の感度に基づいて制約条件を考慮しながら目的関数の 最適値を与える設計変数の変化量を予測すると共に、 設計変数を予測量に相当す る量変化させたときの目的関数の値及び設計変数を予測量に相当する量変化させ たときの制約条件の値を演算し、 予測値と演算値とに基づいて、 制約条件を考慮 しながら前記ステップ (a)で定めた変換系を用いて目的関数の最適値を与える設計 変数の値を求める請求項 2に記載のタイヤの設計方法。
4 . 前記ステップ (c ) は、 前記ステップ (a)で定めた変換系に含まれるタイヤ の設計パラメータの 1つを設計変数として選択するステップ (d ) と、 前記制約 条件を考慮しながら前記ステップ (a)で定めた変換系を用いて目的関数の最適値を 与えるまで前記ステップ (a)で定めた変換系内から選択する設計変数の値を変化さ せるステップ (e ) と、 目的関数の最適値を与える設計変数によるタイヤの設計 パラメータに基づいてタイヤを設計するステップ ( f ) とを含むことを特徴とす る請求項 1に記載のタイヤの設計方法。
5 . 前記ステップ (b ) は、 前記定めた目的関数以外のタイヤ性能及び前記タ ィャの設計パラメータの少なくとも一方の許容範囲を制約する制約条件を定める ことを特徴とする請求項 4に記載のタイヤの設計方法。
6 . 前記ステップ (e ) は、 設計変数の単位変化量に対する目的関数の変化量 の割合である目的関数の感度及び設計変数の単位変化量に対する制約条件の変化 量の割合である制約条件の感度に基づいて制約条件を考慮しながら目的関数の最 適値を与える設計変数の変化量を予測すると共に、 設計変数を予測量に相当する 量変化させたときの目的関数の値及び設計変数を予測量に相当する量変化させた ときの制約条件の値を演算し、 予測値と演算値とに基づいて、 前記制約条件を考 慮しながら前記ステップ (a)で定めた変換系を用いて目的関数の最適値を与えるま で、 選択する設計変数の値を変化させることを特徴とする請求項 4または 5に記 載のタイヤの設計方法。
7 . 前記ステップ (c ) では、 前記ステップ (a)において定めた変換系における タィャの設計パラメ一タを基本モデルとして複数個の基本モデルからなる選択対 象集団を定め、 該選択対象集団の各基本モデルについて、 前記目的関数、 設計変 数、 制約条件、 及び目的関数から評価できる適応関数を定め、 前記選択対象集団 から 2つの基本モデルを選択し、 所定の確率で各基本モデルの設計変数を交叉さ せて新規の基本モデルを生成すること及び少なくとも一方の基本モデルの設計変 数の一部を変更させて新規の基本モデルを生成することの少なくとも一方を行い 、 設計変数を変化させて前記ステップ (a)で定めた変換系を用いて基本モデルの目 的関数、 制約条件及び適応関数を求めて該基本モデル及び設計変数を変化させな かつた基本モデルを保存しかつ保存した基本モデルが所定数になるまで繰リ返し 、 保存した所定数の基本モデルからなる新規集団が所定の収束条件を満たすか否 かを判断し、 収束条件を満たさないときには該新規集団を前記選択対象集団とし て該選択対象集団が所定の収束条件を満たすまで繰り返すと共に、 該所定の収束 条件を満たしたときに保存した所定数の基本モデルのなかで制約条件を考慮しな がら前記ステップ (a)で定めた変換系を用いて目的関数の最適値を与える設計変数 によるタイヤの設計パラメータを求めて該タイヤの設計パラメ一タに基づいてタ ィャを設計することを特徴とすることを特徴とする請求項 1に記載のタイヤの設 計方法。
8 . 前記ステップ ( a ) において、 前記タイヤの設計パラメータを前記タイヤ 性能に変換するように学習された多層フィードフォワード型二ユーラルネットヮ ークのデータで前記変換系を構成したことを特徴とする請求項 1乃至請求項 7の 何れか 1項に記載のタイヤの設計方法。
9 . 請求項 1乃至請求項 8の何れか 1項に記載のタイヤの設計方法によリ設計 された設計パラメ一タによって形成されたタイヤ。
1 0 . タイヤの設計パラメータと該タイヤの性能との非線形な対応関係を求め る変換系計算手段と、
前記タイヤ性能を表す目的関数を定めると共に、 前記タイヤ性能及び前記タイ ャの製造条件の少なくとも一方の許容範囲を制約する制約条件を定めて、 最適化 項目として入力する入力手段と、
前記変換系計算手段を用いて前記入力手段により入力された最適化項目に基づ いて目的関数の最適値を与えるタイヤの設計パラメ一タを求める最適化計算手段 と、
を備えた最適化解析装置。
1 1 . 前記変換系計算手段は、 前記タイヤの設計パラメータ及びタイヤに対す る適用条件と、 前記タイヤ性能との非線形な対応関係を求めることを特徴とする 請求項 1 0に記載の最適化解析装置。
1 2 . 前記最適化計算手段は、 前記変換系計算手段に含まれるタイヤの設計パ ラメータの 1つを設計変数として選択する選択手段と、 前記制約条件を考慮しな がら目的関数の最適値を与えるまで前記変換系計算手段内から選択する設計変数 の値を変化させる変化手段と、 前記変換系計算手段を用いて目的関数の最適値が 与えられるまで設計変数の値を計算する最適値計算手段と、 目的関数の最適値を 与える設計変数によるタイヤの設計パラメータに基づいてタイヤを設計する設計 手段とから構成されたことを特徴とする請求項 1 0または 1 1に記載の最適化解 析装置。
1 3 . 前記最適化計算手段は、 前記変換系計算手段において求めた対応関係に おけるタイャの設計パラメ一タを基本モデルとして複数個の基本モデルからなる 選択対象集団を定め、 該選択対象集団の各基本モデルについて、 前記目的関数、 設計変数、 制約条件、 及び目的関数から評価できる適応関数を定め、 前記選択対 象集団から 2つの基本モデルを選択し、 所定の確率で各基本モデルの設計変数を 交叉させて新規の基本モデルを生成すること及び少なくとも一方の基本モデルの 設計変数の一部を変更させて新規の基本モデルを生成することの少なくとも一方 を行い、 設計変数を変化させて前記変換系計算手段を用いて基本モデルの目的関 数、 制約条件及び適応関数を求めて該基本モデル及び設計変数を変化させなかつ た基本モデルを保存しかつ保存した基本モデルが所定数になるまで繰リ返し、 保 存じた所定数の基本モデルからなる新規集団が所定の収束条件を満たすか否かを 判断し、 収束条件を満たさないときには該新規集団を前記選択対象集団として該 選択対象集団が所定の収束条件を満たすまで繰り返すと共に、 該所定の収束条件 を満たしたときに保存した所定数の基本モデルのなかで制約条件を考慮しながら 前記変換系計算手段を用いて目的関数の最適値を与える設計変数によるタイヤの 設計パラメ一タを求めて該タイヤの設計パラメータに基づいてタイヤを設計する ことを特徴とすることを特徴とする請求項 1 0または 1 1に記載の最適化解析装
1 4 . 前記変換系計算手段は、 前記タイヤの設計パラメータを、 前記タイヤ性 能に変換するように学習された多層フィードフォワード型ニューラルネットヮー クであることを特徴とする請求項 1 0乃至請求項 1 3の何れか 1項に記載の最適 化解析装置。
1 5 . コンピュータによってタイヤを設計するためのタイヤの最適化解析プロ グラムを記憶した記憶媒体であって、
最適化解析プログラムは、
タイヤの設計パラメータと該タイヤの性能との非線形な対応関係を定め、 前記タイヤの性能を表す目的関数を定めると共に、 前記タイヤの性能及び前記 タイヤの製造条件の少なくとも一方の許容範囲を制約する制約条件を定め、 前記定めた対応関係、 前記目的関数及び前記制約条件に基づいて目的関数の最 適値を与えるタイヤの設計パラメータを求めて該タイヤの設計パラメータに基づ いてタイヤを設計することを特徴とするタイヤの最適化解析プログラムを記録し た記録媒体。
1 6 . 前記タイヤの設計パラメータに基づくタイヤの設計は、
前記定めた対応関係、 前記目的関数及び前記制約条件に基づいて、 前記定めた 対応関係に含まれるタイヤの設計パラメータの 1つを設計変数として選択し、 前 記制約条件を考慮しながら目的関数の最適値を与えるまで前記定めた対応関係内 から選択する設計変数の値を変化させ、 目的関数の最適値を与える設計変数によ るタイヤの設計パラメータに基づいてタイヤを設計することを特徴とする請求項 1 5に記載のタイヤの最適化解析プログラムを記録した記録媒体。
1 7 . 前記制約条件は、 前記定めた目的関数以外のタイヤ性能及び前記タイヤ の設計パラメータの少なくとも一方の許容範囲を制約することを特徴とする請求 項 1 6に記載のタイヤの最適化解析プログラムを記録した記録媒体。
1 8 . 前記設計変数の変化は、 設計変数の単位変化量に対する目的関数の変化 量の割合である目的関数の感度及び設計変数の単位変化量に対する制約条件の変 化量の割合である制約条件の感度に基づいて制約条件を考慮しながら目的関数の 最適値を与える設計変数の変化量を予測すると共に、 設計変数を予測量に相当す る量変化させたときの目的関数の値及び設計変数を予測量に相当する量変化させ たときの制約条件の値を演算し、 予測値と演算値とに基づいて、 前記制約条件を 考慮しながら目的関数の最適値を与えるまで、 選択する設計変数の値を変化させ ることを特徴とする請求項 1 6または 1 7に記載のタイヤの最適化解析プロダラ ムを記録した記録媒体。
1 9 . 前記タイヤの設計パラメータに基づくタイヤの設計は、 前記定めた対応 関係におけるタイヤの設計パラメータを基本モデルとして複数個の基本モデルか らなる選択対象集団を定め、 該選択対象集団の各基本モデルについて、 前記目的 関数、 設計変数、 制約条件、 及び目的関数から評価できる適応関数を定め、 前記 選択対象集団から 2つの基本モデルを選択し、 所定の確率で各基本モデルの設計 変数を交叉させて新規の基本モデルを生成すること及び少なくとも一方の基本モ デルの設計変数の一部を変更させて新規の基本モデルを生成することの少なくと も一方を行い、 設計変数を変化させた基本モデルの目的関数、 制約条件及び適応 関数を求めて該基本モデル及び設計変数を変化させなかった基本モデルを保存し かつ保存した基本モデルが所定数になるまで繰り返し、 保存した所定数の基本モ デルからなる新規集団が所定の収束条件を満たすか否かを判断し、 収束条件を満 たさないときには該新規集団を前記選択対象集団として該選択対象集団が所定の 収束条件を満たすまで繰り返すと共に、 該所定の収束条件を満たしたときに保存 した所定数の基本モデルのなかで制約条件を考慮しながら前記対応関係を用いて 目的関数の最適値を与える設計変数によるタイヤの設計パラメ一タを求めて該タ ィャの設計パラメータに基づいてタイヤを設計することを特徴とすることを特徴 とする請求項 1 6乃至請求項 1 8の何れか 1項に記載のタイヤの最適化解析プロ グラムを記録した記録媒体。
PCT/JP1997/002783 1997-08-08 1997-08-08 Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation WO1999007543A1 (fr)

Priority Applications (6)

Application Number Priority Date Filing Date Title
PCT/JP1997/002783 WO1999007543A1 (fr) 1997-08-08 1997-08-08 Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation
DE69731222T DE69731222T2 (de) 1997-08-08 1997-08-08 Verfahren zum entwerfen von reifen, optimierungsanalysereinrichtung und speichermedium zur aufnahme des optimierungsanalyseprogramms
US09/269,972 US7369976B1 (en) 1997-08-08 1997-08-08 Method of designing tire, optimization analyzer and storage medium on which optimization analysis program is recorded
EP97934746A EP0937570B1 (en) 1997-08-08 1997-08-08 Method of designing tire, optimization analyzer and storage medium on which optimization analysis program is recorded
JP51194999A JP4393595B2 (ja) 1997-08-08 1997-08-08 タイヤの設計方法、最適化解析装置及び最適化解析プログラムを記憶した記憶媒体
ES97934746T ES2229377T3 (es) 1997-08-08 1997-08-08 Procedimiento para el diseño de neumaticos, analizador de optimizacion y medios de almacenamiento en los que se registra el programa de analisis de optimizacion.

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP1997/002783 WO1999007543A1 (fr) 1997-08-08 1997-08-08 Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation

Publications (1)

Publication Number Publication Date
WO1999007543A1 true WO1999007543A1 (fr) 1999-02-18

Family

ID=14180952

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP1997/002783 WO1999007543A1 (fr) 1997-08-08 1997-08-08 Methode de conception de pneumatique, module analyseur d'optimisation et support de memorisation sur lequel est enregistre le programme d'analyse d'optimisation

Country Status (6)

Country Link
US (1) US7369976B1 (ja)
EP (1) EP0937570B1 (ja)
JP (1) JP4393595B2 (ja)
DE (1) DE69731222T2 (ja)
ES (1) ES2229377T3 (ja)
WO (1) WO1999007543A1 (ja)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001052041A (ja) * 1999-07-30 2001-02-23 Parametric Technology Corp 多重制約を満たし多重目標機能を最適化する設計パラメータを取得する方法およびシステム
JP2001121540A (ja) * 1999-10-25 2001-05-08 Bridgestone Corp タイヤ設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、タイヤ設計プログラムを記録した記録媒体
JP2001124667A (ja) * 1999-10-25 2001-05-11 Bridgestone Corp 空気入りタイヤの設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、最適化解析装置及びタイヤの最適化解析プログラムを記憶した記憶媒体
JP2001287516A (ja) * 2000-04-04 2001-10-16 Bridgestone Corp タイヤの設計方法、タイヤ用加硫金型の設計方法、タイヤ用加硫金型の製造方法、タイヤの製造方法、タイヤの最適化解析装置及びタイヤの最適化解析プログラムを記録した記憶媒体
JP2002144817A (ja) * 2000-11-06 2002-05-22 Sumitomo Rubber Ind Ltd タイヤトレッドプロファイル展開方法、及びそれによって決定された空気入りタイヤ
JP2002144816A (ja) * 2000-11-06 2002-05-22 Sumitomo Rubber Ind Ltd タイヤトレッドプロファイル展開方法、及びそれによって決定された空気入りタイヤ
US6868716B2 (en) 2000-06-29 2005-03-22 The Yokohama Rubber Co., Ltd. Shape design process of engineering products and pneumatic tire designed using the present design process
JP2008280029A (ja) * 2007-04-13 2008-11-20 Toyo Tire & Rubber Co Ltd タイヤの設計方法
JP2021111312A (ja) * 2020-01-08 2021-08-02 株式会社科学計算総合研究所 情報処理システム、情報処理方法及びプログラム
JP2021176131A (ja) * 2020-05-01 2021-11-04 ダイキン工業株式会社 学習モデル生成方法、プログラム、記憶媒体、学習済みモデル
CN114253157A (zh) * 2021-12-21 2022-03-29 华中科技大学 一种基于二阶灵敏度分析的电机多参数优化方法和系统
CN114930338A (zh) * 2020-01-08 2022-08-19 株式会社科学计算综合研究所 信息处理系统、信息处理方法及程序
CN116484655A (zh) * 2023-06-21 2023-07-25 宁波力劲科技有限公司 一种挤压铸设备合模机构的多目标优化设计方法

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2237140T3 (es) * 1998-09-07 2005-07-16 Bridgestone Corporation Procedimiento de prediccion del comportamiento de un neumatico.
US7133736B2 (en) 2000-07-11 2006-11-07 Pirelli Pneumatici S.P.A. Method for designing a tire and method for producing at least one tire
ATE360856T1 (de) * 2000-07-11 2007-05-15 Pirelli Verfahren zur reifenherstellung
DE10216558A1 (de) * 2002-04-15 2003-10-30 Bayer Ag Verfahren und Computersystem zur Planung von Versuchen
DE102005053296B4 (de) * 2005-11-08 2009-08-13 Kuka Innotec Gmbh Verfahren und Vorrichtung zum automatisierten Stapeln von Reifen auf einem Träger
CA2898187C (en) 2013-01-15 2017-07-11 Wynright Corporation Automatic tire loader/unloader for stacking/unstacking tires in a trailer
US9582616B2 (en) * 2015-01-23 2017-02-28 Siemens Product Lifecycle Management Software Inc. Method for representing and generating a flat pattern for a composite ply that folds over itself
CN107608838A (zh) * 2017-09-26 2018-01-19 郑州云海信息技术有限公司 一种存储设备测试方法、系统、设备及计算机存储介质
CN109033595B (zh) * 2018-07-16 2024-03-29 清华大学 建筑绿色性能模拟分析新方法及装置
KR102576560B1 (ko) * 2019-11-28 2023-09-11 연세대학교 산학협력단 딥 러닝 기반 설계 최적화 방법 및 시스템
US11574093B2 (en) * 2019-12-20 2023-02-07 Google Llc Neural reparameterization for optimization of physical designs
CN111177854B (zh) * 2019-12-31 2023-05-30 西北工业大学 基于直接搜索法的起落架收放机构的优化设计方法
CN113642160A (zh) * 2021-07-26 2021-11-12 南京工业大学 一种基于bp神经网络和鱼群算法的铝合金发动机缸体铸造工艺设计优化方法
CN114896953B (zh) * 2022-04-26 2024-06-18 东风汽车集团股份有限公司 车身侧倾刚度性能指标正向分解方法及分解装置
CN114693172A (zh) * 2022-04-29 2022-07-01 上海钧正网络科技有限公司 共享车辆综合性能的量化分析方法、系统、介质及计算机设备
CN116859725A (zh) * 2023-06-25 2023-10-10 盐城工学院 一种基于遗传算法的汽车底盘控制系统优化方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1994016877A1 (en) * 1993-01-27 1994-08-04 Bridgestone Corporation Design method for a pneumatic tire
JPH07156238A (ja) * 1993-12-03 1995-06-20 Mitsubishi Heavy Ind Ltd 加工機械の設定条件最適化方法
JPH08235242A (ja) * 1995-02-24 1996-09-13 Jasco Corp カーブフィッティング最適化方法
JPH0916654A (ja) * 1995-07-03 1997-01-17 Hitachi Ltd 加工装置および最適加工条件の決定方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5226092A (en) * 1991-06-28 1993-07-06 Digital Equipment Corporation Method and apparatus for learning in a neural network
JP2643065B2 (ja) * 1992-10-19 1997-08-20 住友ゴム工業株式会社 空気入りタイヤ
JP3432573B2 (ja) * 1993-06-10 2003-08-04 株式会社ブリヂストン タイヤのピッチ配列決定方法
JP3686107B2 (ja) * 1993-10-06 2005-08-24 株式会社ブリヂストン 空気入りタイヤの設計方法
US6061673A (en) * 1996-11-06 2000-05-09 Sowa Institute Of Technology Co., Ltd. Learning methods in binary systems

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1994016877A1 (en) * 1993-01-27 1994-08-04 Bridgestone Corporation Design method for a pneumatic tire
JPH07156238A (ja) * 1993-12-03 1995-06-20 Mitsubishi Heavy Ind Ltd 加工機械の設定条件最適化方法
JPH08235242A (ja) * 1995-02-24 1996-09-13 Jasco Corp カーブフィッティング最適化方法
JPH0916654A (ja) * 1995-07-03 1997-01-17 Hitachi Ltd 加工装置および最適加工条件の決定方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP0937570A4 *

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001052041A (ja) * 1999-07-30 2001-02-23 Parametric Technology Corp 多重制約を満たし多重目標機能を最適化する設計パラメータを取得する方法およびシステム
JP2001121540A (ja) * 1999-10-25 2001-05-08 Bridgestone Corp タイヤ設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、タイヤ設計プログラムを記録した記録媒体
JP2001124667A (ja) * 1999-10-25 2001-05-11 Bridgestone Corp 空気入りタイヤの設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、最適化解析装置及びタイヤの最適化解析プログラムを記憶した記憶媒体
JP2001287516A (ja) * 2000-04-04 2001-10-16 Bridgestone Corp タイヤの設計方法、タイヤ用加硫金型の設計方法、タイヤ用加硫金型の製造方法、タイヤの製造方法、タイヤの最適化解析装置及びタイヤの最適化解析プログラムを記録した記憶媒体
US6868716B2 (en) 2000-06-29 2005-03-22 The Yokohama Rubber Co., Ltd. Shape design process of engineering products and pneumatic tire designed using the present design process
JP4580086B2 (ja) * 2000-11-06 2010-11-10 住友ゴム工業株式会社 タイヤトレッドプロファイル展開方法
JP2002144816A (ja) * 2000-11-06 2002-05-22 Sumitomo Rubber Ind Ltd タイヤトレッドプロファイル展開方法、及びそれによって決定された空気入りタイヤ
JP2002144817A (ja) * 2000-11-06 2002-05-22 Sumitomo Rubber Ind Ltd タイヤトレッドプロファイル展開方法、及びそれによって決定された空気入りタイヤ
JP4580087B2 (ja) * 2000-11-06 2010-11-10 住友ゴム工業株式会社 タイヤトレッドプロファイル展開方法
JP2008280029A (ja) * 2007-04-13 2008-11-20 Toyo Tire & Rubber Co Ltd タイヤの設計方法
JP2021111312A (ja) * 2020-01-08 2021-08-02 株式会社科学計算総合研究所 情報処理システム、情報処理方法及びプログラム
CN114930338A (zh) * 2020-01-08 2022-08-19 株式会社科学计算综合研究所 信息处理系统、信息处理方法及程序
CN114930338B (zh) * 2020-01-08 2024-03-01 株式会社科学计算综合研究所 信息处理系统、信息处理方法及记录介质
JP2021176131A (ja) * 2020-05-01 2021-11-04 ダイキン工業株式会社 学習モデル生成方法、プログラム、記憶媒体、学習済みモデル
CN114253157A (zh) * 2021-12-21 2022-03-29 华中科技大学 一种基于二阶灵敏度分析的电机多参数优化方法和系统
CN114253157B (zh) * 2021-12-21 2024-05-14 华中科技大学 一种基于二阶灵敏度分析的电机多参数优化方法和系统
CN116484655A (zh) * 2023-06-21 2023-07-25 宁波力劲科技有限公司 一种挤压铸设备合模机构的多目标优化设计方法
CN116484655B (zh) * 2023-06-21 2023-08-25 宁波力劲科技有限公司 一种挤压铸设备合模机构的多目标优化设计方法

Also Published As

Publication number Publication date
ES2229377T3 (es) 2005-04-16
US7369976B1 (en) 2008-05-06
JP4393595B2 (ja) 2010-01-06
EP0937570B1 (en) 2004-10-13
EP0937570A4 (en) 2001-12-05
EP0937570A1 (en) 1999-08-25
DE69731222T2 (de) 2005-03-10
DE69731222D1 (de) 2004-11-18

Similar Documents

Publication Publication Date Title
WO1999007543A1 (fr) Methode de conception de pneumatique, module analyseur d&#39;optimisation et support de memorisation sur lequel est enregistre le programme d&#39;analyse d&#39;optimisation
JP4393586B2 (ja) 多成分系材料の設計方法、最適化解析装置及び多成分系材料の最適化解析プログラムを記録した記録媒体
JP3686107B2 (ja) 空気入りタイヤの設計方法
JP4142094B2 (ja) 空気入りタイヤの設計方法
CN111709097B (zh) 一种基于零亏格网格曲面连续变形的柔顺机构生成方法
US6430993B1 (en) Method of estimating tire performance
JP2001050848A (ja) 空気入りタイヤの設計方法、最適化解析装置及びタイヤの最適化解析プログラムを記憶した記憶媒体
Giannakoglou et al. Low-cost genetic optimization based on inexact pre-evaluations and the sensitivity analysis of design parameters
US8103488B2 (en) Tire design method
JP2001287516A (ja) タイヤの設計方法、タイヤ用加硫金型の設計方法、タイヤ用加硫金型の製造方法、タイヤの製造方法、タイヤの最適化解析装置及びタイヤの最適化解析プログラムを記録した記憶媒体
WO1998029269A1 (fr) Procedes de conception de pneumatiques
Golmohammadi et al. Supplier selection based on a neural network model using genetic algorithm
Sieger et al. A comprehensive comparison of shape deformation methods in evolutionary design optimization
JP4234857B2 (ja) 空気入りタイヤの設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、最適化解析装置及びタイヤの最適化解析プログラムを記憶した記憶媒体
CN118886281A (zh) 模具结构的优化设计方法及装置
CN118884899A (zh) 智能模具的生产控制方法、装置、设备及存储介质
Karen et al. Intelligent die design optimization using enhanced differential evolution and response surface methodology
JP3673323B2 (ja) 空気入りタイヤの設計方法
JP2001124667A (ja) 空気入りタイヤの設計方法、タイヤ用加硫金型設計方法、タイヤ用加硫金型製造方法、空気入りタイヤの製造方法、最適化解析装置及びタイヤの最適化解析プログラムを記憶した記憶媒体
CN114648247A (zh) 一种工艺规划与调度集成的再制造决策方法
WO2022004039A1 (ja) 予測モデルの学習方法、予測モデルの学習装置、及び、プラント制御システム
JP4776275B2 (ja) タイヤ設計装置及びタイヤ設計方法
JP4755015B2 (ja) タイヤの設計方法
Glänzel et al. Parameterization of environmental influences by automated characteristic diagrams for the decoupled fluid and structural-mechanical simulations
Renner Genetic algorithms in computer-aided design

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): JP US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FI FR GB GR IE IT LU MC NL PT SE

WWE Wipo information: entry into national phase

Ref document number: 09269972

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 1997934746

Country of ref document: EP

121 Ep: the epo has been informed by wipo that ep was designated in this application
WWP Wipo information: published in national office

Ref document number: 1997934746

Country of ref document: EP

WWG Wipo information: grant in national office

Ref document number: 1997934746

Country of ref document: EP

点击 这是indexloc提供的php浏览器服务,不要输入任何密码和下载