CN116186910A - A Method for Establishing a Drill Tool Wear Prediction Model and a Drill Tool Wear Prediction System - Google Patents

Abstract

The invention discloses a method for establishing a drilling tool wear prediction model and a drilling tool wear prediction system. The method for establishing a drilling tool wear prediction model in the present invention is based on the establishment of a mapping relationship between drilling parameters and drilling tool wear. Deep learning, with parameters while drilling as input, drilling tool health and wear as output, training to obtain a prediction model of drilling tools, can solve the problem of complex working conditions and difficult to measure drilling tool wear in the existing technology, and subsequent applications , the health condition of the drilling tool can be obtained by collecting the parameters while drilling, and there is no need to lay out components such as sensors, so the practicability is good; the drilling tool wear prediction system of the present invention only needs to collect the parameters while drilling and input it into the drilling tool wear prediction model of the wear prediction unit , the wear amount and health of the drilling can be displayed on the display unit, and the monitoring and control of the wear of the drilling tool can be realized.

Description

A method for establishing a drill wear prediction model and a drill wear prediction system

Technical Field

The invention relates to the technical field of rock drilling construction, and in particular to a method for establishing a drill tool wear prediction model and a drill tool wear prediction system.

Background Art

In projects where drilling and blasting is the main excavation method, digital and intelligent rock drilling rigs have replaced the traditional extensive manual operation mode, achieving more accurate, efficient and safer goals. During the rock drilling process, the drill bit and drill rod of the rock drilling rig are the components that are in direct contact with the rock. By setting the thrust pressure, rotation pressure, impact pressure, drilling speed and other drilling parameters in the control system of the rock drilling rig, the hydraulic thrust system and the hydraulic rock drill can produce thrust, rotation, impact and other actions to achieve the rock crushing function. However, due to the complex tunnel construction environment and limited geological survey information, the actual service process of drill bits and drill rods often leads to rock cracks and cavities, uneven rock hardness, poor soil quality, and drill bit sticking, drill wear, drill rod bending, etc. Frequent replacement of drill tools requires interrupting the construction process. If a rod breaks or a serious equipment failure occurs due to drill tool wear, the equipment will be shut down for maintenance, which will seriously affect the construction period. Therefore, it is necessary to monitor the health and wear of the drill tools in real time during the construction process to determine whether the drill tools need to be replaced in advance and to perform effective predictive maintenance.

Depending on the industry, the application scenarios of drill tools vary greatly. For example, in geological exploration, petroleum industry, coal industry, tunnel construction, etc., the drill tools used are different in terms of materials, processes, shapes, sizes and construction environments. Therefore, the complexity of the environment makes it difficult to implement a universal and effective method to identify the health of drill tools. Also, due to the harsh service environment of drill tools and complex movement forms, it is difficult to directly deploy sensors on the drill tools to measure their wear. Therefore, monitoring or predicting the health of drill tools through indirect means is a major problem that needs to be solved.

In summary, there is an urgent need for a method for establishing a drill tool wear prediction model and a drill tool wear prediction system to solve the problem of drill tool wear prediction in the prior art.

Summary of the invention

The purpose of the present invention is to provide a method for establishing a drill wear prediction model and a drill wear prediction system to solve the problem of drill wear prediction in the prior art. The specific technical solution is as follows:

A method for establishing a drill tool wear prediction model comprises the following steps:

Step S1, obtaining all single drilling data of the current shift and photos of drill tool wear after a single drilling work is completed;

Step S2, preprocessing all single drilling data;

Step S3, extracting features of the drilling parameters in the single drilling data to obtain data features of the drilling parameters;

Step S4, calculating the wear amount of the drill tool after a single drilling according to the drill tool wear photo in step S1, and defining the health of the drill tool after each drilling according to the wear amount;

Step S5: Based on the deep learning method, a mapping relationship between the data characteristics of the drilling parameters and the health degree is established to construct a wear prediction model for the drilling tool.

Preferably, in the above technical solution, in step S2, the drilling parameters in the single drilling data include one or more of thrust pressure, rotary pressure, impact pressure, and drilling speed.

Preferably, the above technical solution comprises step S2.1 and step S2.2;

Step S2.1: Perform data cleaning on all single drilling data. The cleaning rule is: determine the data length of the single drilling data. If the data length of the single drilling data is within the range of [300, 500], then keep the single drilling data. Otherwise, remove the single drilling data.

Step S2.2: Determine whether to exclude the single drilling data according to the drilling speed in the single drilling data. The rule is:

If the drilling speed is greater than N m/min in a single drilling data, and the subsequent drilling speed remains greater than N m/min within T seconds, the single drilling data will be discarded.

In the above technical solution, step S2 further includes step S2.3, performing drilling state differentiation processing on the retained single drilling data, as follows:

Step 1: According to the drilling speed in the single drilling data, the single drilling data is divided into multiple drilling states along the time domain;

Step 2: The data in the same drilling state in the single drilling data are concentrated as samples of the same type, and the single drilling data containing multiple samples of the same type are used as processing samples in the next step.

Preferably, in the above technical solution, in step S3, the data characteristics of the while-drilling parameters include statistical characteristics of the while-drilling parameters and time domain/frequency domain characteristics of the while-drilling parameters;

Extracting statistical characteristics of the drilling parameters: extracting the maximum value, minimum value, average value, variance, origin moment and central moment of the drilling parameters to obtain the statistical characteristics of the drilling parameters;

Extracting the time domain/frequency domain features of the while-drilling parameters: The time domain/frequency domain features of the while-drilling parameters are obtained by extracting the modulus maximum feature of the multi-scale space based on wavelet transform.

The above technical solution is preferably, the step S4 comprises:

Step S4.1: noise reduction and removal of redundant components of the drill wear photo after the current single drilling work is completed;

Step S4.2: Based on the feature-based image alignment method, all pixels of the drill tool wear photo to be aligned are mapped onto the standard drill tool photo to align the two photos;

Step S4.3: Taking the standard drilling tool photo as a reference, respectively measure the radial dimension _Di of the drilling tool wear photo and the axial dimension _Hi of the drill tooth to the measurement reference; calculate the actual axial dimension _hi of the drilling tool after the current single drilling work is completed through the radial dimension _Di and the axial dimension _Hi ;

Step S4.4: Calculate the wear amount after the current single drilling work is completed according to the actual axial size h _i ;

Step S4.5: Define the health of the drill after the current single drilling work is completed according to the wear amount.

The above technical solution is preferred, in step S4.3, the actual axial dimension h _i is as shown in formula 1):

Where d represents the radial dimension of the standard drilling tool.

The above technical solution is preferred, in the step S4.4, the wear amount Wear _i is as shown in formula 2);

Wherein, _hi represents the actual axial size of the drill tool after the current single drilling work is completed; hi _-1 represents the actual axial size of the drill tool after the previous single drilling work is completed; h represents the actual axial size of a brand new drill tool; S=1 represents that the drill tool after the current drilling work is a newly replaced drill tool; S=0 represents that the drill tool after the current drilling work is an old drill tool that has not been replaced;

In step S4.5, the rules for defining health are as follows:

When 0≤Wear _i ≤0.1mm, the health of the drill tool is wear-free;

When 0.1 mm＜Wear _i ≤1 mm, the health of the drill tool is slightly worn;

When 1mm＜Wear _i ≤2mm, the health of the drill tool is moderately worn;

When 2mm＜Wear _i ≤4mm, the health of the drill tool is severely worn;

When Wear _i ＞4mm, the health of the drill bit is damaged.

The above technical solution is preferred, in step S5,

Based on the deep learning method, the data features of the drilling parameters of the single drilling data in step S3 are used as the input parameters of the prediction model, and the health of the drill tool in step S4 is used as the output of the prediction model to construct a deep learning training set, thereby obtaining a drill tool wear prediction model.

A drilling tool wear prediction system includes a data acquisition unit, a wear prediction unit and a display unit;

The data acquisition unit is used to collect drilling parameters during drilling; the wear prediction unit is provided with a prediction model obtained according to the method for establishing a drill wear prediction model; the prediction model is connected to the data acquisition unit; the prediction model is connected to the display unit, and the display unit is used to display the wear amount and health of the drill.

The application of the technical solution of the present invention has the following beneficial effects:

(1) The method for establishing a drill tool wear prediction model in the present invention establishes a mapping relationship between the drilling parameters and the drill tool wear amount. Based on deep learning, the drilling parameters are used as input, and the drill tool health and wear amount are used as output. The prediction model of the drill tool is trained to solve the problem that the drill tool wear condition is difficult to measure under complex working conditions in the prior art. In subsequent applications, the health condition of the drill tool can be obtained by collecting the drilling parameters without the need to deploy sensors and other components, which has good practicality.

(2) In the present invention, the drilling parameters can be identified, classified, and labeled, specifically: the start and end time points of a single drilling are identified, and the corresponding wear amount of the drill tool is identified; various working conditions during the drilling process are classified, and the drilling parameters are classified for better subsequent abnormal data processing and model training robustness; labeling the drilling process and analyzing the data under the same drilling state as similar samples will help improve the accuracy of the analysis and avoid unpredictable errors caused by different drilling methods.

(3) In the present invention, image parameters are used as the basis for judging the amount of drill tool wear. The amount of drill tool wear after each drilling can be obtained in sequence through the algorithm, which corresponds one-to-one with the drilling parameters during drilling, thereby improving the accuracy of model training. The present invention can predict and identify the drill tool loss value and drill tool health through different training models and in various ways such as drilling tool by drilling tool.

(4) The drill tool wear prediction system of the present invention only needs to collect drilling parameters and input them into the drill tool wear prediction model of the wear prediction unit, and then the wear amount and health of the drill tool can be displayed on the display unit, thereby realizing the monitoring and control of the drill tool wear condition.

In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings constituting a part of this application are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

In the attached picture:

FIG1 is a flow chart of a method for establishing a drill wear prediction model in this embodiment;

FIG2 is a schematic diagram of drill bit size measurement in this embodiment;

FIG. 3 is a schematic diagram for distinguishing drilling states in this embodiment.

DETAILED DESCRIPTION

The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

Example:

A method for establishing a drill wear prediction model includes the following steps S1 to S5, as shown in FIG1 , and specifically as follows:

Step S1, collecting the drilling data of the current shift, and dividing the drilling data of the current shift into multiple groups of single drilling data, obtaining all the single drilling data, and taking photos of the drill tool after each single drilling is completed by a camera device, and obtaining photos of the wear of the drill tool after the single drilling work is completed, specifically as follows: collecting the drilling parameter data including the vehicle current, vehicle voltage, propulsion speed, impact pressure, rotation pressure, and propulsion pressure through the rock drilling rig boom data system, and taking photos of the wear of the drill tool side of the rock drilling rig corresponding to the data after each single drilling is completed, that is, taking a photo once after each drilling, one-to-one correspondence, to facilitate the subsequent establishment of a mapping relationship;

The detailed instructions are as follows:

Collect drilling data: The drilling rig boom data system can record the hydraulic data and electrical data of the drilling rig in real time during the construction process, including drilling rate, thrust pressure, impact pressure, rotary pressure, rotary speed, water pressure, water flow, etc. Taking into account the differences in units and orders of magnitude of various parameters, the sample data is normalized and converted into a dimensionless form. The data converted into a dimensionless form has the characteristics of no units and similar orders of magnitude, which is convenient for subsequent mathematical modeling analysis.

Collect photos of drill tool wear: Photos of drill tool wear can be taken at the construction site. First, take a photo of the prototype before the drill bit is installed as a reference for subsequent wear. After each single drilling work is completed, take photos of the drill tool. The drill bit and drill rod can be removed, and side photos can be taken along the radial direction. Preferably, the camera position and drill bit placement position are fixed to reduce the error of subsequent drill tool wear analysis.

Among them, in step S1, the drilling data of the current shift needs to be divided into multiple groups of single drilling data, specifically: in this embodiment, the collected drilling data is an uninterrupted data stream of the drilling rig debugging and drilling process, and in a single drilling process, if the drilling data (drilling speed) is greater than N m/min (N m/min in this embodiment is 5m/min), and there is no new normal drilling data within the next 2 or 3 seconds (here the normal drilling data is, for example, a drilling speed of 2-4m/min), it is considered that the single drilling has been completed. In this way, the data stream (i.e., drilling data) is divided into multiple groups of single drilling data. Of course, in addition to this, other existing means can also be used to divide the overall drilling data into single drilling data.

Step S2: pre-process all single drilling data collected in the current shift, that is, clean the data (eliminate abnormal values, invalid values, redundant values, etc.), as follows:

Step S2.1: Data cleaning is performed on all separated single drilling data. The cleaning rule is: determine the data length of the single drilling data. If the data length of the single drilling data is within the range of [300,500], the single drilling data is retained, otherwise it is discarded. The principle explanation of this step S2.1 is: this step S2.1 is based on the particularity of the drilling trolley data, and improves the traditional outlier analysis method. Since the traditional method only eliminates the data that is too high or too low by setting a threshold, it does not fully consider the data anomalies caused by over-excavation, under-excavation, trial drilling, etc. that may occur during the actual drilling process. Therefore, this embodiment adopts a global abnormal data cleaning method to determine the length of the single group drilling data (including the length of the electrical and hydraulic data characteristics of the trolley when the whole vehicle is running). If there is an obvious reduction in the amount of data, the data is discarded to ensure the validity of the training data. In this embodiment, the reason for the reduction in the amount of data can be determined in combination with the on-site record situation, and the database notes are "under-excavation and supplementary drilling", "trial drilling", "drill bit breakage", "drill rod breakage", "other anomalies", etc.

Step S2.2: judging whether to remove the single drilling data by the drilling speed in the single drilling data, the judgment rule is: if the drilling speed in the single drilling data is greater than N m/min (that is, N m/min here is also preferably 5 m/min), and the subsequent drilling speed remains greater than 5 m/min within T seconds (T seconds here is preferably 2 or 3 seconds), then remove the single drilling data;

It should be noted that: if the drilling speed in a single drilling data is greater than 5m/min, but the subsequent drilling speed returns to the normal drilling speed after T seconds (the normal drilling speed here is, for example, 2.5m/min-3.5m/min), then the single drilling data is retained, that is, it is considered that a hole in the rock is encountered at this time, and the data should be retained;

Step S2.3: Perform drilling status differentiation processing on the retained single drilling data, as follows:

Step 1: Divide the single drilling data into multiple drilling states along the time domain according to the drilling speed in the single drilling data. That is to say, for each group of single drilling data, it is necessary to divide it into drilling states;

Step 2: The data in the same drilling state in the single drilling data are concentrated as samples of the same type, and the single drilling data containing multiple samples of the same type are used as processing samples for the next step (i.e., step S3). The explanation here is that a group of single drilling data can be divided into multiple drilling states, and the multiple drilling states divided from a group of single drilling data are classified, that is, the samples of the group of single drilling data after classification can be obtained;

A further explanation of this step S2.3 is:

The single drilling data is divided into multiple drilling states along the time domain by the drilling speed, and the data is decomposed into common drilling states of on-site construction such as high impact, low impact, anti-stuck drill, and drill withdrawal along the time domain direction. The data under the same drilling state in the single drilling data are concentrated as similar samples for analysis, which will help improve the accuracy of the analysis and avoid unpredictable errors due to different drilling methods. Among them, different drilling states are mainly reflected by the drilling speed. As shown in Figure 3, according to the system setting of the rock drilling rig, the average feed speed in the high impact state is usually 4m/min, and the average feed speed in the low impact state is usually 2.5m/min. The rapid drop in the drilling process is the anti-stuck drill action. The specific means of dividing the drilling state according to the drilling speed here is well known to those skilled in the art, and will not be repeated in this embodiment.

Step S3: For all the single drilling data classified in step S2.3, feature extraction is performed on the drilling parameters in these single drilling data to obtain data features of the drilling parameters. The data features of the drilling parameters here include statistical features of the drilling parameters and time domain/frequency domain features of the drilling parameters, as follows:

Extract statistical features of drilling parameters: including the maximum, minimum, average, variance, origin moment and central moment of drilling parameters in a single drilling data. For example, for the drilling parameter of propulsion pressure, extract the maximum, minimum, average and variance of propulsion pressure respectively;

Extracting time domain/frequency domain features of while drilling parameters: A modulus maximum feature extraction method of a multi-scale space based on wavelet transform is used to extract the time domain/frequency domain features of the while drilling parameters. The extraction of time domain/frequency domain features here can refer to the prior art. This embodiment provides a specific step of extracting time domain/frequency domain features, which is as follows:

Step 1: Wavelet basis function is scaled and translated:

Where Ψ(·) is the wavelet basis function, t is the horizontal coordinate of the original data, a is the scaling factor, and b is the translation factor.

Step 2: The continuous wavelet transform of any original data f(t) is:

Where _Wf (a,b) is the data function after wavelet transformation, which contains two independent variables: scaling factor a and translation factor b. It can be seen that the continuous wavelet transform is a mapping from f(t)→ _Wf (a,b), and its inverse transform is:

为满足逆变换的约束条件，

为ψ(t)的傅里叶变换；In the formula,

To satisfy the constraints of the inverse transformation,

is the Fourier transform of ψ(t);

Step 3: Based on the Mallat algorithm, the single drilling data is subjected to wavelet transformation and then decomposed into high-frequency and low-frequency components to obtain time domain/frequency domain features (i.e., the single drilling data can be decomposed into frequency domain components after wavelet transformation through the above steps);

Step S4: Calculate the wear amount of the drill tool after a single drilling according to the drill tool wear photo in step S1, and define the health of the drill tool after each drilling according to the wear amount. The general idea of this step S4 is: perform image processing on the drill tool wear photo taken, remove redundant components and noise points of the image, calibrate the drill tool (drill rod) reference calculation part through the alignment algorithm, and measure the radial wear amount of the drill tool (drill bit); at the same time, combined with the actual situation of the drill bit, establish the health of the drill tool loss after a single drilling, that is, define the health index and record the loss situation, as follows:

Step S4.1: After a single drilling is completed, the photograph of the drill wear is subjected to noise reduction and removal of redundant components. The noise reduction and removal of redundant components can be performed with reference to the prior art;

Step S4.2: Based on the feature-based image alignment method, all pixels of the drill wear photo to be aligned are mapped to the standard drill photo to align the two photos, as follows:

First, in step S4.2, the general idea of the feature-based image alignment method is:

Obtain a new drill bit picture (i.e., a photo of a standard drill tool), and use Photoshop tools to convert the picture into a standard side view, which will serve as the reference picture for aligning all the drill tool wear photos;

Use ORB to extract feature points. ORB is a feature point detector, which consists of two parts: one is the locator, which finds points on the image that are rotationally invariant, scale-invariant, and affine invariant; the other is the descriptor, which obtains the appearance encoding of the feature points to distinguish each other. In this way, the feature points can be represented by the descriptor. Ideally, the same physical point corresponding to different photos should have the same descriptor.

The specific steps are as follows:

1): Read the reference image (i.e., the photo of the standard drilling tool) and the image to be aligned (i.e., the photo of the wear of the drilling tool) into the memory respectively;

2) Detect ORB feature points for the two images, use the parameter MAX_FEATURES to control the number of feature points detected, and use the detectAndCompute function to detect feature points and calculate descriptors;

3) Find the matching feature points in the two images, arrange them according to the matching degree, retain the most matching minority, and use the Hamming distance to measure the similarity of the two feature point descriptors.

4) The matching feature points generated in the previous step may have certain errors. A random sample consensus algorithm is used to calculate the homography matrix in the case of certain matching errors.

5) Use the warpPerspective function to map all pixels of the image to be aligned to the reference image to align the image, so that the reference image and the image to be aligned can be aligned;

Step S4.3: respectively measure the radial dimension _Di of the drill tool wear photo and the axial dimension _Hi from the drill tooth to the measurement reference; calculate the actual axial dimension _hi of the drill tool after the current single drilling work is completed through the radial dimension _Di and the axial dimension _Hi , specifically: use the Photoshop tool to open the aligned image _Pi , and respectively measure its radial dimension _Di and the axial dimension _Hi from the drill tooth to the measurement reference, as shown in Figure 2;

Among them, it is known that the radial dimension d of the standard drill bit (i.e., a brand new drill bit) is 45 mm, and the actual axial dimension h _i of the image P _i is as shown in formula 1):

Step S4.4: Calculate the wear amount after the current single drilling work is completed according to the actual axial size h _i. In this embodiment, the drill wear amount is defined as the difference in the axial size of the drill bit between two consecutive drillings. The wear amount Wear _i is shown in Formula 2):

Wherein, _hi represents the actual axial size of the drill after the current single drilling work is completed; hi _-1 represents the actual axial size of the drill after the last single drilling work is completed; h represents the actual axial size of the brand new drill (that is, the standard size);

S=1 means that the drilling tool after the single drilling is completed is a newly replaced drilling tool; S=0 means that the drilling tool after the single drilling is completed is an old drilling tool that has not been replaced;

Formula 2) is further explained as follows:

If the drill tool is a new drill tool (i.e., S=1), the difference between the actual axial dimension h _i of the drill tool after the current single drilling work is completed and the actual axial dimension h _i of the standard drill tool is the wear amount after the current single drilling work is completed;

If the drill tool is an old drill tool (i.e., S=0), the difference between the actual axial dimension h _i of the drill tool after the current single drilling work is completed and the actual axial dimension h _i-1 of the drill tool after the previous single drilling work is completed is the wear amount after the current single drilling work is completed;

Among them, the standard for distinguishing new drilling tools from old drilling tools is: if the drilling times of the drilling tools are less than or equal to one, it is a new drilling tool, otherwise it is an old drilling tool.

Step S4.5: Define the health of the drill after the current single drilling work is completed according to the wear amount, and record the wear amount as follows:

When 0≤Wear _i ≤0.1mm, the health of the drill tool is wear-free;

When 0.1 mm＜Wear _i ≤1 mm, the health of the drill tool is slightly worn;

When 1mm＜Wear _i ≤2mm, the health of the drill tool is moderately worn;

When 2mm＜Wear _i ≤4mm, the health of the drill tool is severely worn;

When Wear _i ＞4mm, the health of the drill tool is damaged;

Preferably, when defining the health of the drilling tool, the actual use time of the drilling tool may also be combined for evaluation.

Step S5: Based on the deep learning method, a mapping relationship between the data characteristics of the drilling parameters and the health degree is established, and a wear prediction model of the drilling tool is obtained by training, as follows:

In step S3, the data feature matrix (i.e., time domain/frequency domain features) and statistical features of the current drilling state of the drilling tool are obtained according to the drilling parameters. The data feature matrix and statistical features are used as input parameters of the mathematical model (i.e., the convolutional neural network model). At the same time, the wear amount after each drilling is solved based on the image alignment algorithm described in step S4, and the health degree is defined as the output of the mathematical model. A training set for deep learning mathematical model training is constructed. The mathematical model is trained and the parameters are adjusted and optimized based on the training set to obtain a health degree prediction model for drilling tool wear of a rock drilling rig.

The present embodiment also discloses a drill tool wear prediction system, which includes a data acquisition unit, a wear prediction unit and a display unit; the data acquisition unit is used to collect drilling parameters as input during drilling work; the wear prediction unit is provided with a drill tool wear prediction system, which is established according to the above-mentioned method for establishing a drill tool wear prediction model; the drill tool wear prediction model is connected to the data acquisition unit, and the drilling parameters are input into the wear prediction model, and the wear prediction model can output the wear amount and health of the drill tool, wherein the prediction model is connected to the display unit, and the display unit is used to display the wear amount and health of the drill tool.

In this embodiment, a specific example is provided for the method of establishing the above-mentioned drill wear prediction model, as follows:

As shown in Table 1, Table 1 shows the drilling parameter data collected in the above step S1;

Table 1 Some drilling parameters of single drilling data

Step S2: pre-process the collected while-drilling parameter data, clean the collected while-drilling parameter data, segment the cleaned data according to the drilling tool construction actions, and obtain the data blocks corresponding to the construction actions (drilling status) such as preparation action, high impact, low impact, anti-stuck drill, and drill withdrawal. As shown in Table 1, rows 1 and 2 are preparation actions, rows 1000-1002 represent the high impact stage, and row 4204 represents the drill withdrawal status.

Step S3: Based on the data obtained by the preprocessing in the previous step, a modulus maximum feature extraction method based on multi-scale space of wavelet transform is selected to construct a characteristic matrix of the working loss of the drilling rig. Table 2 shows part of the data of the characteristic matrix X of the construction parameters of the drilling rig at a certain stage;

Table 2 Partial data of the characteristic matrix X of the construction parameters of the rock drilling rig

Among them, the matrix dimension of the construction parameter feature matrix X is (29, 400, 120), and Table 2 shows part of the data of X(0, 400, 120).

Step S4: Perform image processing on the drill tool damage images corresponding to the 30 sets of single drilling data to remove redundant components and noise points of the images, calibrate the drill rod reference calculation position through the alignment algorithm, and measure the wear amount of the drill bit, as shown in Table 3-1 and Table 3-2:

Table 3-1 Wear of the first 15 groups of drilling tools

Table 3-2 Wear of the last 15 groups of drilling tools

At the same time, combined with the actual use time of the drill bit, a health label for the loss of the drill tool after a single drilling is established, that is, the health index is defined and the loss is recorded. The health definition is shown in Table 4 below:

Table 4 Drilling tool health

Step S5: Use sample data and health to train the convolutional neural network model, and the accuracy of the sample set is 95%; select another set of 30 sets of sample drilling parameter data sets as the test set, and use the trained convolutional neural network model to verify the test set, and the accuracy of the verification result reaches 80%.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. The method for establishing the drilling tool wear prediction model is characterized by comprising the following steps of:

step S1, acquiring all single drilling data of the current shift and drilling tool abrasion photos after the single drilling work is completed;

s2, preprocessing all single drilling data;

s3, extracting characteristics of the while-drilling parameters in the single drilling data to obtain data characteristics of the while-drilling parameters;

s4, calculating the abrasion loss of the drilling tool after single drilling according to the abrasion photo of the drilling tool in the step S1, and defining the health degree of the drilling tool after each drilling according to the abrasion loss;

and S5, establishing a mapping relation between the data characteristics of the while-drilling parameters and the health degree based on a deep learning method, and constructing a wear prediction model of the drilling tool.

2. The method according to claim 1, wherein in the step S2, the while-drilling parameters in the single-pass drilling data include one or more of a push pressure, a swing pressure, a percussion pressure, and a drilling rate.

3. The method of creating a predicted model of drill wear according to claim 2, wherein step S2 includes steps S2.1 and S2.2;

step S2.1: data cleaning is carried out on all single drilling data, and the cleaning rules are as follows: judging the data length of the single-time drilling data, if the data length of the single-time drilling data is within the range of [300,500], reserving the single-time drilling data, otherwise, eliminating the single-time drilling data;

step S2.2: judging whether to reject the single drilling data according to the drilling speed in the single drilling data, wherein the rule is as follows:

and if the drilling speed is larger than N m/min in the single drilling data and the subsequent drilling speeds are all kept larger than N m/min within T seconds, eliminating the single drilling data.

4. The method for building a predicted model of drill wear according to claim 3, wherein step S2 further comprises step S2.3 of performing a drilling state discrimination process on the retained single-pass drilling data, specifically as follows:

the first step: dividing the single drilling data into a plurality of drilling states along a time domain according to the drilling speed in the single drilling data;

and a second step of: and taking the data set in the same drilling state in the single drilling data as a similar sample, and taking the single drilling data containing a plurality of similar samples as a processing sample of the next step.

5. The method according to claim 3 or 4, wherein in step S3, the data features of the parameter while drilling include statistical features of the parameter while drilling and time/frequency domain features of the parameter while drilling;

extracting statistical characteristics of parameters while drilling: extracting the maximum value, the minimum value, the average value, the variance, the origin moment and the center moment of the parameter while drilling to obtain the statistical characteristics of the parameter while drilling;

extracting time domain/frequency domain characteristics of the parameter while drilling: and extracting by adopting a wavelet transform-based mode maximum feature extraction method in a multi-scale space to obtain the time domain/frequency domain features of the parameter while drilling.

6. The method for building a predicted model of drill wear according to claim 1, wherein the step S4 comprises:

step S4.1: noise reduction and redundant component removal are carried out on the drilling tool abrasion photo after the current single drilling work is completed;

step S4.2: mapping all pixels of the drilling tool abrasion photo to be aligned onto a standard drilling tool photo based on a characteristic image alignment method so as to align the two photos;

step S4.3: measuring radial dimension D of abrasion photo of drilling tool based on standard drilling tool photo _i And axial dimension H of the drilling tooth to the measurement reference _i The method comprises the steps of carrying out a first treatment on the surface of the By radial dimension D _i And an axial dimension H _i Calculating the actual axial dimension h of the drilling tool after the current single drilling work is completed _i ；

Step S4.4: according to the actual axial dimension h _i Calculating the abrasion loss after the current single drilling work is completed;

step S4.5: and defining the health degree of the drilling tool after the current single drilling work is completed according to the abrasion loss.

7. The method according to claim 6, wherein in the step S4.3, the actual axial dimension h _i As shown in formula 1):

where d represents the radial dimension of the standard drill.

8. The method for building a predicted model of drill Wear according to claim 7, wherein in the step S4.4, the Wear amount spar _i As shown in formula 2);

wherein h is _i Representing the actual axial dimension of the drilling tool after the current single drilling operation is completed; h is a _i-1 Representing the actual axial dimension of the drilling tool after the last single drilling operation is completed; h represents the actual axial dimension of the brand new drilling tool; s=1, the drilling tool after the current drilling is completed is a new replaced drilling tool; s=0 represents the current drilling after completion of the drillingThe drilling tool is an old drilling tool which is not replaced;

in step S4.5, the rules defining the health degree are as follows:

when 0 is equal to or less than Wear _i When the diameter is less than or equal to 0.1mm, the health degree of the drilling tool is abrasion-free;

when 0.1mm < Wear _i When the thickness is less than or equal to 1mm, the health degree of the drilling tool is slightly worn;

when 1mm < Wear _i When the length is less than or equal to 2mm, the health degree of the drilling tool is moderate abrasion;

when 2mm < Wear _i When the thickness is less than or equal to 4mm, the health degree of the drilling tool is severely worn;

when the weather is _i At > 4mm, the health of the drill is compromised.

9. The method for creating a model for predicting wear of a drilling tool according to any one of claims 6 to 8, wherein in the step S5,

based on a deep learning method, taking the data characteristics of the while-drilling parameters of the single drilling data in the step S3 as input parameters of a prediction model, taking the health degree of the drilling tool in the step S4 as output of the prediction model, and constructing a training set of deep learning so as to obtain a drilling tool abrasion prediction model.

10. The drilling tool wear prediction system is characterized by comprising a data acquisition unit, a wear prediction unit and a display unit;

the data acquisition unit is used for acquiring parameters while drilling in drilling work; the wear prediction unit is provided with a prediction model obtained by the method for establishing a drilling tool wear prediction model according to claim 9; the prediction model is connected with the data acquisition unit; the prediction model is connected with a display unit, and the display unit is used for displaying the abrasion loss and the health degree of the drilling tool.

Images