CN114046815A - Encoder self-correction method and device based on deep learning - Google Patents
Encoder self-correction method and device based on deep learning Download PDFInfo
- Publication number
- CN114046815A CN114046815A CN202111322204.3A CN202111322204A CN114046815A CN 114046815 A CN114046815 A CN 114046815A CN 202111322204 A CN202111322204 A CN 202111322204A CN 114046815 A CN114046815 A CN 114046815A
- Authority
- CN
- China
- Prior art keywords
- encoder
- physical signal
- correction
- signal
- residual error
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Transmission And Conversion Of Sensor Element Output (AREA)
Abstract
The encoder self-correction method and device based on deep learning are characterized in that an encoder is mounted on target equipment, and displacement change of the encoder is carried out; acquiring a first physical signal and a second physical signal before the displacement of an encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packaging; and respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after signal preprocessing according to the initialization mark to obtain compensation hyper-parameters and forward reasoning decision. The invention can more flexibly, quickly and accurately finish the correction of the encoder without depending on any external physical reference; particularly for the encoder without bearing installation, a user can complete compensation more conveniently without a dismounting process, so that the production efficiency is improved, and the time cost is greatly reduced; the self-correcting super-parameter can be updated along with the change of the physical installation position, and the stability of the position precision of the encoder is kept.
Description
Technical Field
The invention relates to a depth learning-based encoder self-correction method and device, and belongs to the technical field of encoder detection.
Background
An encoder is a device that converts angular displacement or linear displacement into an electric signal, and is an angle detection device widely used in the industrial field. In the process of assembling or using the encoder, because of the influence of shafting precision and sensitive elements on the position deviation, an angle feedback error can be generated, and further the precision of the equipment is lost. In severe cases, error codes of the encoder can be caused, angle feedback errors can be caused, and dangers such as galloping can be caused. It is critical that the encoder be detected and calibrated before measurement applications are performed by the encoder.
In the prior art, an encoder is coaxially or rigidly connected through a connecting piece (a coupler, a transmission shaft and the like), is directly or indirectly compared with a reference position (such as a reference encoder, a high-precision optical machine platform, a grating reading head, a so-called self-correcting reference device and the like), establishes a one-to-one correspondence relationship, and then electronically calibrates the existing encoder in a data table storage mode. The calibration tool in the traditional technical scheme has the disadvantages of complex structure, complex process and high time-consuming and cost-consuming calibration process. The requirement on the precision of a calibration standard is extremely high, and the locking of the physical position relation is strictly ensured in a mechanical mode after the calibration is finished, so that the mechanical position strain caused by the change of an external environment cannot be adapted.
Disclosure of Invention
Therefore, the invention provides a depth learning-based encoder self-correction method and device, and solves the problems that the encoder correction process is complicated, the time consumption and the cost are high, the encoder needs to depend on an external physical reference, and the mechanical position strain caused by external environment change cannot be adapted.
In order to achieve the above purpose, the invention provides the following technical scheme: the encoder self-correction method based on deep learning comprises the following steps:
mounting an encoder to a target device, and performing displacement variation of the encoder;
acquiring a first physical signal and a second physical signal before the displacement of the encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packing;
respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision, and obtaining the encoder position after compensation and correction.
As a preferable scheme of the encoder self-correction method based on the deep learning, unidirectional or bidirectional displacement variation of the encoder is performed manually or electrically, and the displacement variation mode comprises rotation or linear displacement;
the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder.
As a preferred scheme of the encoder self-correction method based on deep learning, when the encoder is judged not to finish initializing a correction mark, residual model modeling and operation of an encoder position signal are carried out, and a first physical signal and a second physical signal before and after the encoder displacement change, which are collected according to a preset frequency, are synthesized into batch data.
As a preferable scheme of the encoder self-correction method based on the deep learning, the residual error model is established according to a mathematical relationship between the first physical signal and the second physical signal before and after the displacement variation of the encoder.
As an optimal scheme of a deep learning-based encoder self-correction method, a hyper-parameter of the residual error model is optimized and updated through a gradient strategy, and a residual error calculation result of the updated residual error model is judged;
when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped;
and the number of the hyper-parameters of the residual error model is adjusted according to the actual error model.
As a preferred scheme of the encoder self-correction method based on the deep learning, after the initial correction of the encoder is completed, the stored initial hyper-parameters are used for directly reasoning and deciding the deep learning network.
As a preferred solution of the encoder self-correction method based on deep learning, the direct reasoning and decision of the deep learning network includes: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; and reasoning and predicting the periodic position and calculating the position fine-tuning decision probability.
As a preferred scheme of an encoder self-correction method based on deep learning, whether the output position of an encoder needs to be corrected or not is determined according to the position fine-tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
The invention also provides a self-correcting device of the encoder based on deep learning, wherein the encoder is arranged on the target equipment to carry out displacement change of the encoder; the method comprises the following steps:
the signal acquisition unit is used for acquiring a first physical signal and a second physical signal before the displacement of the encoder changes;
the signal preprocessing unit is used for performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packaging;
and the compensation correction unit is used for respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision so as to obtain the encoder position after compensation correction.
As a preferable scheme of the encoder self-correcting device based on the deep learning, unidirectional or bidirectional displacement variation of the encoder is performed manually or electrically, and the displacement variation mode comprises rotation or linear displacement;
the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder.
As a preferred scheme of the encoder self-correction device based on deep learning, the compensation correction unit further comprises a residual error model construction subunit, wherein the residual error model construction subunit is used for performing residual error model modeling and operation on an encoder position signal when the encoder is judged not to complete the initialization correction mark, and synthesizing a first physical signal and a second physical signal before and after the encoder displacement change, which are acquired according to a preset frequency, into batch data;
the residual error model is established according to the mathematical relationship between the first physical signal and the second physical signal before and after the displacement change of the encoder.
As a preferred solution of the encoder self-correction device based on deep learning, the compensation correction unit further includes a hyper-parameter updating subunit, and the hyper-parameter updating subunit is configured to optimize and update a hyper-parameter of the residual error model through a gradient strategy; judging the residual error calculation result of the updated residual error model;
and when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped.
As a preferred scheme of the encoder self-correction device based on deep learning, the number of the hyper-parameters of the residual error model is adjusted according to an actual error model;
and after the initial correction of the encoder is finished, the stored initial hyper-parameters are utilized to carry out direct reasoning and decision of the deep learning network.
As a preferred solution of the encoder self-correction device based on deep learning, the direct reasoning and decision of the deep learning network includes: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; reasoning and predicting the periodic position and calculating the position fine tuning decision probability;
determining whether the output position of the encoder needs to be corrected or not according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
The invention has the following advantages: mounting an encoder to a target device, and performing displacement change of the encoder; acquiring a first physical signal and a second physical signal before the displacement of an encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packaging; respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision, and obtaining the encoder position after compensation and correction. The invention can more flexibly, quickly and accurately finish the correction of the encoder without depending on any external physical reference; particularly for the encoder without bearing installation, a user can complete compensation more conveniently without a dismounting process, so that the production efficiency is improved, and the time cost is greatly reduced; when the environmental temperature changes or the relative position of the structure changes, the self-correcting super-parameter can be updated along with the change of the physical installation position, and the stability of the position precision of the encoder is maintained.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It should be apparent that the drawings in the following description are merely exemplary, and that other embodiments can be derived from the drawings provided by those of ordinary skill in the art without inventive effort.
The structures, ratios, sizes, and the like shown in the present specification are only used for matching with the contents disclosed in the specification, so that those skilled in the art can understand and read the present invention, and do not limit the conditions for implementing the present invention, so that the present invention has no technical significance, and any structural modifications, changes in the ratio relationship, or adjustments of the sizes, without affecting the functions and purposes of the present invention, should still fall within the scope of the present invention.
Fig. 1 is a schematic flow chart of a deep learning-based encoder self-correction method according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating a processing flow of a residual error model in a deep learning-based encoder self-correction method according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an inference network in a deep learning-based encoder self-correction method according to an embodiment of the present invention;
fig. 4 is a schematic diagram illustrating speed feedback convergence in a self-calibration process in a deep learning-based encoder self-calibration method according to an embodiment of the present invention;
fig. 5 is a schematic diagram of an encoder self-calibration apparatus based on deep learning according to an embodiment of the present invention.
Detailed Description
The present invention is described in terms of particular embodiments, other advantages and features of the invention will become apparent to those skilled in the art from the following disclosure, and it is to be understood that the described embodiments are merely exemplary of the invention and that it is not intended to limit the invention to the particular embodiments disclosed. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
Referring to fig. 1, there is provided a depth learning-based encoder self-correction method, including the steps of:
s1, mounting the encoder on a target device, and performing displacement change of the encoder;
s2, acquiring a first physical signal and a second physical signal before the displacement of the encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packing;
and S3, respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation hyperparameter and forward reasoning decision, and obtaining the encoder position after compensation and correction.
In the embodiment, the encoder is subjected to unidirectional or bidirectional displacement variation in a manual or electric mode, and the displacement variation mode comprises rotation or linear displacement; the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder. And when the encoder is judged not to complete the initialization correction mark, residual model modeling and operation of the encoder position signal are carried out, and the first physical signal and the second physical signal before and after the encoder displacement change, which are acquired according to the preset frequency, are synthesized into batch data.
Specifically, the encoder only needs to be manually or electrically operated to complete unidirectional or bidirectional rotation or linear displacement of the encoder after the encoder is installed on the target equipment. Collecting a first physical signal Yc and a second physical signal Y before the displacement of an encoder changess, performing signal preprocessing on the first physical signal Yc and the second physical signal Ys, wherein the signal preprocessing unit comprises a normalization operation and signal data packing, and the normalization operation Yc, k is Yc/2N,Ys,k＝Ys/2NAnd N is the number of sampling circuit bits. After the signal data is packed and processed, the signal data is packed,
when the initial correction mark is judged not to be completed, residual model modeling and operation are required to be performed firstly, and batch data are synthesized according to the first physical signal Yc and the second physical signal Ys which are acquired at the preset frequency, so that the learning process is accelerated.
Referring to fig. 2, in this embodiment, the residual error model is established according to a mathematical relationship between the first physical signal and the second physical signal before and after the displacement variation of the encoder, or according to a current physical motion state of the encoder. Optimizing and updating the hyper-parameters of the residual error model through a gradient strategy, and judging the residual error calculation result of the updated residual error model; when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped; and the number of the hyper-parameters of the residual error model is adjusted according to the actual error model.
For example, residual model basis
The establishment of the connection between the mobile terminal and the mobile terminal,
wherein
Θ＝(θ1,θ2,θ3,θ4) ', k is the number of sampling points and theta is the hyperparameter.
Specifically, the hyper-parameters are optimized and updated through a gradient strategy,
wherein,
in the formula, mkA first moment estimate being a gradient parameter; vkSecond moment estimates which are gradient parameters; γ is learning rate, and is recommended to be 1e-8, θ5＝0.98,θ6＝0.9995。
Specifically, whether the updated residual calculation result is smaller than a set threshold is judged, and when the updated residual calculation result is smaller than the preset threshold, updating of the hyper-parameters can be stopped, and the hyper-parameters at the moment are stored for later use. And when the set value does not meet the judgment condition, continuously repeating the updating of the super-parameters until the set value meets the requirement, and stopping learning.
Specifically, the number of θ parameters can be adjusted according to the actual error model (e.g., nonlinear error, harmonic error, etc.), and is not limited to 4, and may be many. The residual model may also be established by other means, such as a deep enhanced network (DQN), a multi-layer perceptron (MLP), or a Deep Neural Network (DNN).
Referring to fig. 3 and 4, in the present embodiment, after the initial correction of the encoder is completed, the stored initial hyper-parameters are used to perform direct reasoning and decision making for the deep learning network. The direct reasoning and decision making of the deep learning network comprises the following steps: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; and reasoning and predicting the periodic position and calculating the position fine-tuning decision probability. Determining whether the output position of the encoder needs to be corrected or not according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
Specifically, the deep learning network forward reasoning decision comprises the compensation and correction of the collected position signal,
obtaining an ideal value corrected by an encoder; inferential prediction of periodic position Pk＝N’(θ1,θ2,θ3,θ4…), calculation bitsSetting and fine-tuning decision probability Qk＝N”(θ1,θ2,θ3,θ4,…|Qk-1,Qk-2,Qk-3,Qk-4…), where N, N', N "are different partial units of the designed deep learning network.
According to QkDetermining whether the output position needs to be corrected, if so, repeating the updating of the hyper-parameter, if not, directly delivering PkAs a high-precision ideal position value of the encoder obtained after compensation.
In summary, the encoder is mounted to the target device, and displacement variation of the encoder is performed; acquiring a first physical signal and a second physical signal before the displacement of an encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packaging; respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision, and obtaining the encoder position after compensation and correction. When the encoder is judged not to complete the initialization correction mark, residual model modeling and operation of the encoder position signal are carried out, and a first physical signal and a second physical signal which are acquired according to the preset frequency and before and after the encoder displacement changes are combined into batch data. The residual error model is established according to the mathematical relation between the first physical signal and the second physical signal before and after the displacement change of the encoder. Optimizing and updating the hyper-parameters of the residual error model through a gradient strategy, and judging the residual error calculation result of the updated residual error model; when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped; and adjusting the number of the hyper-parameters of the residual error model according to the actual error model. And after the initial correction of the encoder is finished, the stored initial hyper-parameters are utilized to carry out direct reasoning and decision of the deep learning network. The direct reasoning and decision making of the deep learning network comprises the following steps: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; and reasoning and predicting the periodic position and calculating the position fine-tuning decision probability. Determining whether the output position of the encoder needs to be corrected or not according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value. The invention can more flexibly, quickly and accurately finish the correction of the encoder without depending on any external physical reference; particularly for the encoder without bearing installation, a user can complete compensation more conveniently without a dismounting process, so that the production efficiency is improved, and the time cost is greatly reduced; when the environmental temperature changes or the relative position of the structure changes, the self-correcting super-parameter can be updated along with the change of the physical installation position, and the stability of the position precision of the encoder is maintained.
Example 2
Referring to fig. 5, embodiment 2 of the present invention further provides an encoder self-calibration apparatus based on deep learning, in which an encoder is mounted to a target device, and displacement variation of the encoder is performed; the method comprises the following steps:
the signal acquisition unit 1 is used for acquiring a first physical signal and a second physical signal before the displacement of the encoder changes;
the signal preprocessing unit 2 is used for performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packing;
and the compensation correction unit 3 is configured to perform reward and punishment training on the first physical signal and the second physical signal data after the signal preprocessing respectively according to the initialization flag to obtain a compensation superparameter and a forward inference decision, and obtain a compensated and corrected encoder position.
In the embodiment, the encoder is subjected to unidirectional or bidirectional displacement variation in a manual or electric mode, and the displacement variation mode comprises rotation or linear displacement;
the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder.
In this embodiment, the compensation correction unit 3 further includes a residual error model constructing subunit 31, where the residual error model constructing subunit 31 is configured to perform residual error model modeling and operation on an encoder position signal when it is determined that the encoder does not complete the initialization correction flag, and synthesize a first physical signal and a second physical signal, which are acquired according to a preset frequency, before and after the encoder displacement changes into batch data;
the residual error model is established according to the mathematical relationship between the first physical signal and the second physical signal before and after the displacement change of the encoder.
In this embodiment, the compensation correction unit 3 further includes a hyper-parameter updating subunit 32, where the hyper-parameter updating subunit 32 is configured to optimize and update a hyper-parameter of the residual error model through a gradient strategy; judging the residual error calculation result of the updated residual error model;
and when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped.
In this embodiment, the number of hyper-parameters of the residual error model is adjusted according to the actual error model;
and after the initial correction of the encoder is finished, the stored initial hyper-parameters are utilized to carry out direct reasoning and decision of the deep learning network.
In this embodiment, the direct reasoning and decision of the deep learning network includes: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; reasoning and predicting the periodic position and calculating the position fine tuning decision probability;
determining whether the output position of the encoder needs to be corrected or not according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
It should be noted that, because the contents of information interaction, execution process, and the like between the modules/units of the apparatus are based on the same concept as the method embodiment in embodiment 1 of the present application, the technical effect brought by the contents is the same as the method embodiment of the present application, and specific contents may refer to the description in the foregoing method embodiment of the present application, and are not described herein again.
Example 3
The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
Example 4
An embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;
the processor and the memory are communicated with each other through a bus; the memory stores program instructions executable by the processor, the processor invoking the program instructions capable of performing the deep learning based encoder self-correction method of embodiment 1 or any possible implementation thereof.
Specifically, the processor may be implemented by hardware or software, and when implemented by hardware, the processor may be a logic circuit, an integrated circuit, or the like; when implemented in software, the processor may be a general-purpose processor implemented by reading software code stored in a memory, which may be integrated in the processor, located external to the processor, or stand-alone.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.).
It will be apparent to those skilled in the art that the modules or steps of the present invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different than that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
Although the invention has been described in detail above with reference to a general description and specific examples, it will be apparent to one skilled in the art that modifications or improvements may be made thereto based on the invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.
Claims (10)
1. The encoder self-correction method based on deep learning is characterized by comprising the following steps of:
mounting an encoder to a target device, and performing displacement variation of the encoder;
acquiring a first physical signal and a second physical signal before the displacement of the encoder changes, and performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packing;
respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision, and obtaining the encoder position after compensation and correction.
2. The encoder self-correction method based on deep learning of claim 1, characterized in that the encoder unidirectional or bidirectional displacement variation is performed manually or electrically, and the displacement variation comprises rotation or linear displacement;
the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder.
3. The encoder self-correction method based on deep learning of claim 1, wherein when it is determined that the encoder has not completed initializing the correction flag, residual model modeling and operation of encoder position signals are performed, and a first physical signal and a second physical signal before and after the encoder displacement variation, which are collected according to a preset frequency, are synthesized into batch data.
4. The depth-learning-based encoder self-correction method according to claim 3, wherein the residual error model is established according to a mathematical relationship between the first physical signal and the second physical signal before and after the displacement variation of the encoder.
5. The encoder self-correction method based on deep learning of claim 4, wherein the residual calculation result of the updated residual model is judged by optimizing and updating the hyper-parameters of the residual model through a gradient strategy;
when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped;
and the number of the hyper-parameters of the residual error model is adjusted according to the actual error model.
6. The deep learning based encoder self-correction method according to claim 5, wherein after the initial correction of the encoder is completed, the stored initial hyper-parameters are used for direct reasoning and decision making of the deep learning network.
7. The deep learning based encoder self-correction method according to claim 6, wherein the direct reasoning and decision making of the deep learning network comprises: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; and reasoning and predicting the periodic position and calculating the position fine-tuning decision probability.
8. The deep learning based encoder self-correction method according to claim 7, wherein it is determined whether the output position of the encoder needs to be corrected according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
9. The encoder self-correction device based on deep learning is used for mounting an encoder to target equipment and carrying out displacement change of the encoder; it is characterized by comprising:
the signal acquisition unit is used for acquiring a first physical signal and a second physical signal before the displacement of the encoder changes;
the signal preprocessing unit is used for performing signal preprocessing on the first physical signal and the second physical signal, wherein the signal preprocessing comprises normalization operation and signal data packaging;
and the compensation correction unit is used for respectively carrying out reward and punishment training on the first physical signal data and the second physical signal data after the signal preprocessing according to the initialization mark to obtain compensation superparameter and forward reasoning decision so as to obtain the encoder position after compensation correction.
10. The encoder self-correction device based on deep learning of claim 9, characterized in that the encoder unidirectional or bidirectional displacement variation is performed manually or electrically, and the displacement variation comprises rotation or linear displacement;
the encoder is a sensor with a position sensing function, and is a photoelectric encoder, a magnetoelectric encoder, a capacitive encoder or an inductive encoder;
the compensation correction unit further comprises a residual error model building subunit, wherein the residual error model building subunit is used for carrying out residual error model modeling and operation on the position signal of the encoder when the encoder is judged not to complete the initialization correction mark, and synthesizing a first physical signal and a second physical signal which are acquired according to preset frequency before and after the displacement change of the encoder into batch data;
the residual error model is established according to the mathematical relation between the first physical signal and the second physical signal before and after the displacement change of the encoder;
the compensation correction unit further comprises a hyper-parameter updating subunit, and the hyper-parameter updating subunit is used for optimizing and updating the hyper-parameters of the residual error model through a gradient strategy; judging the residual error calculation result of the updated residual error model;
when the residual error calculation result of the residual error model is smaller than a preset threshold value, stopping updating the hyper-parameters of the residual error model, and storing and using the hyper-parameters after updating is stopped;
the number of the hyper-parameters of the residual error model is adjusted according to the actual error model;
after the initial correction of the encoder is completed, the stored initial hyper-parameters are utilized to carry out direct reasoning and decision of the deep learning network;
the direct reasoning and decision making of the deep learning network comprises the following steps: acquiring compensation correction of the position signal to obtain an ideal value after the position of the encoder is corrected; reasoning and predicting the periodic position and calculating the position fine tuning decision probability;
determining whether the output position of the encoder needs to be corrected or not according to the position fine tuning decision probability; when the correction is not needed, the predicted cycle position is directly output as the compensated ideal position value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111322204.3A CN114046815B (en) | 2021-11-09 | 2021-11-09 | Encoder self-correction method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111322204.3A CN114046815B (en) | 2021-11-09 | 2021-11-09 | Encoder self-correction method and device based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114046815A true CN114046815A (en) | 2022-02-15 |
| CN114046815B CN114046815B (en) | 2024-02-09 |
Family
ID=80207794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111322204.3A Active CN114046815B (en) | 2021-11-09 | 2021-11-09 | Encoder self-correction method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114046815B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117705164A (en) * | 2024-02-06 | 2024-03-15 | 科瑞工业自动化系统（苏州）有限公司 | Error compensation method, system and device for magneto-optical encoder |
| CN119737993A (en) * | 2025-01-16 | 2025-04-01 | 浙江禾川科技股份有限公司 | Signal adjustment method, device, equipment and storage medium |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090010310A (en) * | 2007-07-23 | 2009-01-30 | 성균관대학교산학협력단 | Output signal compensation device and compensation method of magnetic encoder |
| CN102680016A (en) * | 2012-05-14 | 2012-09-19 | 北京理工大学 | Error compensating method of photoelectric encoder |
| CN103837169A (en) * | 2014-02-28 | 2014-06-04 | 哈尔滨工业大学 | Self-correcting device and method for magneto-electricity encoder and magneto-electricity encoder |
| CN108180933A (en) * | 2018-02-02 | 2018-06-19 | 哈尔滨理工大学 | It is a kind of based on magnetism encoder automatic correcting method of the permanent magnet synchronous motor without sensor speed control |
| CN108416086A (en) * | 2018-01-25 | 2018-08-17 | 大连理工大学 | A kind of aero-engine whole envelope model adaptation modification method based on deep learning algorithm |
| CN108519768A (en) * | 2018-03-26 | 2018-09-11 | 华中科技大学 | A Fault Diagnosis Method Based on Deep Learning and Signal Analysis |
| CN109000702A (en) * | 2018-05-16 | 2018-12-14 | 苏州汇川技术有限公司 | Encoder corrects system and method |
| CN109543824A (en) * | 2018-11-30 | 2019-03-29 | 腾讯科技（深圳）有限公司 | A kind for the treatment of method and apparatus of series model |
| CN110108312A (en) * | 2018-02-01 | 2019-08-09 | 莱纳林德有限公司 | Method, computer program, equipment and encoder for signal errors correction |
| CN111504360A (en) * | 2020-05-19 | 2020-08-07 | 哈尔滨理工大学 | Method and device for precise angle division of magnetoelectric encoder based on time coordinate |
| CN111521212A (en) * | 2020-04-22 | 2020-08-11 | 上海英威腾工业技术有限公司 | Magnetic encoder correction system and method, control terminal and readable storage medium |
| CN111835366A (en) * | 2020-07-14 | 2020-10-27 | 上海交通大学 | Parallel signal processing device and method based on convolutional circular autoencoder |
| WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
| CN112631434A (en) * | 2021-01-11 | 2021-04-09 | 福州大学 | Deep learning-based vibrotactile coding and decoding method |
| CN113408722A (en) * | 2021-05-13 | 2021-09-17 | 桂林电子科技大学 | Situation assessment factor extraction method based on layer-by-layer loss compensation depth self-encoder |
| US20230124431A1 (en) * | 2020-04-06 | 2023-04-20 | Nova Ltd. | Machine and deep learning methods for spectra-based metrology and process control |
-
2021
- 2021-11-09 CN CN202111322204.3A patent/CN114046815B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090010310A (en) * | 2007-07-23 | 2009-01-30 | 성균관대학교산학협력단 | Output signal compensation device and compensation method of magnetic encoder |
| CN102680016A (en) * | 2012-05-14 | 2012-09-19 | 北京理工大学 | Error compensating method of photoelectric encoder |
| CN103837169A (en) * | 2014-02-28 | 2014-06-04 | 哈尔滨工业大学 | Self-correcting device and method for magneto-electricity encoder and magneto-electricity encoder |
| CN108416086A (en) * | 2018-01-25 | 2018-08-17 | 大连理工大学 | A kind of aero-engine whole envelope model adaptation modification method based on deep learning algorithm |
| CN110108312A (en) * | 2018-02-01 | 2019-08-09 | 莱纳林德有限公司 | Method, computer program, equipment and encoder for signal errors correction |
| CN108180933A (en) * | 2018-02-02 | 2018-06-19 | 哈尔滨理工大学 | It is a kind of based on magnetism encoder automatic correcting method of the permanent magnet synchronous motor without sensor speed control |
| CN108519768A (en) * | 2018-03-26 | 2018-09-11 | 华中科技大学 | A Fault Diagnosis Method Based on Deep Learning and Signal Analysis |
| CN109000702A (en) * | 2018-05-16 | 2018-12-14 | 苏州汇川技术有限公司 | Encoder corrects system and method |
| CN109543824A (en) * | 2018-11-30 | 2019-03-29 | 腾讯科技（深圳）有限公司 | A kind for the treatment of method and apparatus of series model |
| WO2021007812A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳大学 | Deep neural network hyperparameter optimization method, electronic device and storage medium |
| US20230124431A1 (en) * | 2020-04-06 | 2023-04-20 | Nova Ltd. | Machine and deep learning methods for spectra-based metrology and process control |
| CN111521212A (en) * | 2020-04-22 | 2020-08-11 | 上海英威腾工业技术有限公司 | Magnetic encoder correction system and method, control terminal and readable storage medium |
| CN111504360A (en) * | 2020-05-19 | 2020-08-07 | 哈尔滨理工大学 | Method and device for precise angle division of magnetoelectric encoder based on time coordinate |
| CN111835366A (en) * | 2020-07-14 | 2020-10-27 | 上海交通大学 | Parallel signal processing device and method based on convolutional circular autoencoder |
| CN112631434A (en) * | 2021-01-11 | 2021-04-09 | 福州大学 | Deep learning-based vibrotactile coding and decoding method |
| CN113408722A (en) * | 2021-05-13 | 2021-09-17 | 桂林电子科技大学 | Situation assessment factor extraction method based on layer-by-layer loss compensation depth self-encoder |
Non-Patent Citations (2)
| Title |
|---|
| 于海;梁立辉;王树洁;卢新然;万秋华;: "基于径向基函数神经网络的高精度基准编码器误差补偿", 红外与激光工程, no. 12 * |
| 王义文;焦环宇;刘献礼;李博;张帆;权超健;林长友;梅恒;: "绝对式编码器输出信号的误差自动补偿", 光电工程, no. 01 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117705164A (en) * | 2024-02-06 | 2024-03-15 | 科瑞工业自动化系统（苏州）有限公司 | Error compensation method, system and device for magneto-optical encoder |
| CN117705164B (en) * | 2024-02-06 | 2024-05-07 | 科瑞工业自动化系统（苏州）有限公司 | Error compensation method, system and device for magneto-optical encoder |
| CN119737993A (en) * | 2025-01-16 | 2025-04-01 | 浙江禾川科技股份有限公司 | Signal adjustment method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114046815B (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115950557B (en) | Intelligent temperature compensation method based on pressure transmitter | |
| CN102680016A (en) | Error compensating method of photoelectric encoder | |
| CN114462594A (en) | Neural network training method and device, electronic equipment and storage medium | |
| CN109696187B (en) | Eccentric correcting device of rotary encoder | |
| CN112649017B (en) | A method and device for calibrating odometer system error | |
| CN115704697B (en) | Temperature calibration compensation method, device, equipment and medium for gyroscope | |
| CN114046815A (en) | Encoder self-correction method and device based on deep learning | |
| CN111413651A (en) | Compensation method, device and system for total magnetic field and storage medium | |
| CN117007097A (en) | Sensor temperature calibration method, device, system, electronic equipment and medium | |
| CN108871382B (en) | An error correction method and system for a photoelectric encoder based on error fitting | |
| CN117110362A (en) | Pipeline heat loss monitoring method, device, equipment and medium based on Internet of things | |
| CN116638518A (en) | Positioning accuracy compensation method and device, electronic equipment and storage medium | |
| CN112013892B (en) | Encoder ranging correction method and device, electronic equipment and storage medium | |
| CN109571898B (en) | Precision compensation system and method for manipulator of injection molding machine | |
| CN119704090A (en) | Intelligent electric batch head intelligent calibration and identification method, device, equipment and storage medium | |
| CN114298098A (en) | Data processing method and device for error correction | |
| CN119085931A (en) | MEMS pressure sensor nonlinear correction method, system, equipment and medium | |
| CN117705164B (en) | Error compensation method, system and device for magneto-optical encoder | |
| CN114440866A (en) | Sensor data calibration method and calibration system | |
| CN111721210B (en) | Initialization method, device, equipment and medium after conversion of logical raster resolution | |
| CN116361741B (en) | High-precision positioning system and method based on multi-sensor fusion | |
| CN115689061B (en) | Wind power ultra-short term power prediction method and related equipment | |
| CN116482405A (en) | Accelerometer calibration method and system based on ADMM algorithm | |
| CN115752456B (en) | A method and system for correcting temperature compensation parameters of inertial group products | |
| CN113704990A (en) | Single-parameter online correction method for electromagnetic transient model of direct-current measurement data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |