A wear pattern prediction method and a wear testing machine thereof

Abstract

The application discloses a wear form prediction method and a wear testing machine thereof. The prediction method comprises the following steps: performing a wear test to obtain three kinds of test data: a friction coefficient, a temperature and a noise; performing label coding on a plurality of wear forms; extracting time domain features of the three kinds of test data respectively, and then converting the time domain features into frequency domain data, and finally extracting frequency domain features of the frequency domain data; forming a data set and generating a plurality of training subsets, then constructing a decision tree and repeatedly obtaining a plurality of decision trees, finally selecting a category of majority prediction as a final result, and obtaining a preliminary prediction function; judging whether the three preliminary prediction functions have a mode, if yes, determining that the mode is the final prediction function, otherwise, establishing a weighted function to determine the final prediction function; and performing wear form prediction according to the final prediction function. The application can perform real-time data analysis while performing the friction and wear test, and predict all wear forms of the material in the whole working condition process.

Description

Technical Field

[0001] This invention relates to a prediction method in the field of mechanical wear detection technology, particularly a method for predicting wear patterns, and also to a wear testing machine. Background Technology

[0002] Currently, to investigate the wear patterns that may occur in specific materials under given working conditions, simulation experiments using a friction and wear testing machine are typically required. However, existing friction and wear testing machines themselves do not have wear pattern prediction capabilities. After the test, subsequent observation equipment such as scanning electron microscopy (SEM) is needed to analyze the wear track morphology in order to determine the specific wear pattern. This method is cumbersome and cannot directly and conveniently identify the wear pattern during or after the test.

[0003] Furthermore, in real-world operating conditions, the wear patterns of materials often evolve dynamically over service time, with different wear patterns occurring at different stages. For example, one type of wear in the initial stage may induce other wear patterns in subsequent stages. Observation after testing with existing friction and wear testing machines can only obtain the final wear morphology at the end of the test, and cannot fully reveal and record the multiple wear patterns that occur sequentially or alternately throughout the entire friction and wear process and their evolution laws. Summary of the Invention

[0004] To address the technical problem that existing friction and wear tests cannot predict wear patterns and require the use of equipment such as scanning electron microscopes to determine wear patterns, this invention provides a method for predicting wear patterns and a wear testing machine thereof.

[0005] This invention is achieved using the following technical solution: a wear pattern prediction method, comprising the following steps:

[0006] Wear tests were conducted to obtain three types of test data: coefficient of friction, temperature, and noise.

[0007] Label and encode various wear patterns;

[0008] First, extract the time-domain features of the three types of experimental data respectively, then convert the time-domain features into frequency-domain data, and finally extract the frequency-domain features of the frequency-domain data.

[0009] First, the time-domain / frequency-domain features of the three experimental data and the wear form labels are encoded to form a dataset and generate multiple training subsets. Then, a decision tree is built for each training subset and the process is repeated to obtain multiple decision trees. Each decision tree is recursively split from the root node until the node meets a preset stopping condition and is marked as a leaf node. Finally, the majority of predicted categories are selected as the final result, and a preliminary prediction function is obtained.

[0010] Determine whether the three preliminary prediction functions have a mode. If they do, the mode of the three preliminary prediction functions is determined as the final prediction function. Otherwise, establish a weighted function to determine the final prediction function.

[0011] Wear pattern prediction is performed based on the final prediction function.

[0012] This invention first obtains three types of experimental data: friction coefficient, temperature, and noise. Then, it extracts features from each set of experimental data using time-domain and frequency-domain feature extraction methods, corresponding to different wear patterns. A dataset is then constructed based on these data features and their corresponding wear patterns. A decision tree is built and trained on the data to obtain three preliminary prediction functions. The existence of a mode for these three preliminary prediction functions yields a final prediction function and a corresponding final prediction model. Finally, the final prediction model is used to predict wear patterns. This invention solves the technical problem that existing friction and wear tests cannot predict wear patterns and require the use of equipment such as scanning electron microscopes for wear pattern determination. It allows for real-time acquisition of experimental data and enables real-time data analysis during friction and wear tests, predicting all wear patterns of the material throughout the entire operating process. This reduces the use of experimental equipment, saves time and costs, and provides a reference for product design, manufacturing, and engineering applications.

[0013] As a further improvement to the above scheme, the time-domain characteristics include peak value f. X,1 Mean f X,2 Root mean square f X,3 Standard deviation f X,4 , kurtosis f X,5 skewness f X,6 Margin index f X,7 Waveform index f X,8 、, impulse index f X,9 The frequency domain features include the frequency mean F. S,1 Frequency centroid F S,2 Root mean square frequency F S,3 Frequency variance F S,4 Where X∈{μ,Te,No}, μ is the friction coefficient, Te is the temperature, and No is the noise.

[0014] Furthermore, the formula for calculating the time-domain features is as follows:

[0015] f X，1 =max(|X(t)|)

[0016]

[0017]

[0018] Where X(t) represents the time-domain data of the experiment, t represents time, T represents the end time point, and n represents the number of valid samples.

[0019] Furthermore, the frequency domain data is S(k), and the calculation formula is:

[0020]

[0021] The formula for calculating the frequency domain features is as follows:

[0022]

[0023]

[0024] Among them, f k Let be the frequency value of the k-th spectral line.

[0025] Furthermore, the wear mode is one of adhesive wear, abrasive wear, fatigue wear, and oxidative wear, and the four wear modes are labeled and coded according to 1, 2, 3, and 4;

[0026] H training subsets are generated from the dataset using Bootstrap sampling. Each training subset accounts for 2 / 3 of the original data, and the remaining 1 / 3 of the original data is used as out-of-bag data. The method for constructing the decision tree includes the following steps: first, randomly select m features when splitting at each node of the tree; then, select the best feature for node splitting based on the Gini index as the splitting criterion; finally, perform recursive splitting, repeating feature selection and the selection of the best splitting point until the number of node samples is less than a set threshold or the maximum depth is reached.

[0027] Furthermore, the datasets for the three types of experimental data are as follows:

[0028]

[0029] Among them, D μ D Te D No These represent the datasets for the friction coefficient, the temperature, and the noise, respectively. The 13-dimensional feature vector composed of the time-domain / frequency-domain features is represented; Yi∈{1,2,3,4} represents the category label of the four wear forms.

[0030] Furthermore, the method for determining the final prediction function through the weighting function includes the following steps:

[0031] make Calculate the deviation: in, These represent three preliminary prediction functions;

[0032] Calculate the weighting factors:

[0033] Let the weighting function be The final prediction function is then... The round(·) function indicates rounding to the nearest integer.

[0034] Furthermore, the preliminary prediction function for the friction coefficient is:

[0035]

[0036] Among them, T b Let b represent the b-th decision tree, B represent the total number of decision trees, and mode(·) represent the mode.

[0037] Furthermore, the correction model makes time-segmented predictions: a time period ΔT is set, and the prediction results of the correction model are compared with those of the previous time period. If the prediction results of the two models are different in the time period hΔT to (h+1)ΔT, the correction model updates the prediction data of the final prediction model, and the time hΔT when the prediction results of the correction model begin to change is taken as the starting time of the updated final prediction model. The final prediction model is then recalculated. During output, the prediction results of the final prediction model before time hΔT are output as one type of wear pattern, and the prediction results after time hΔT are output as another type of wear pattern, thus predicting multiple wear patterns in different time periods throughout the entire working process.

[0038] The present invention also provides a wear testing machine, which is applied in any of the above-mentioned wear mode prediction methods, the wear testing machine comprising:

[0039] frame;

[0040] A power mechanism, which is mounted on the frame;

[0041] Tooling, which is mounted on the frame;

[0042] The grinding mechanism is fixed on the frame; the power mechanism drives the tooling to reciprocate on the frame, and the tooling drives the test sample block to contact the grinding mechanism, causing the test sample block to wear.

[0043] A friction sensor, mounted on the frame, is used to detect the friction force generated by the wear of the test specimen; the wear testing machine calculates the coefficient of friction based on the friction force.

[0044] A temperature sensor, mounted on the frame, is used to detect the temperature generated by the wear of the test specimen.

[0045] A noise sensor, mounted on the frame, is used to detect noise generated by wear of the test specimen.

[0046] Compared with existing technologies, the wear pattern prediction method and wear testing machine of the present invention have the following advantages:

[0047] 1. This wear pattern prediction method first obtains three types of experimental data: friction coefficient, temperature, and noise. Then, it extracts features from each set of experimental data using time-domain and frequency-domain feature extraction methods, corresponding to different wear patterns. A dataset is then constructed based on these features and their corresponding wear patterns. A decision tree is built and trained on the data to obtain three preliminary prediction functions. The final prediction function is obtained by determining if the three preliminary prediction functions have a mode, leading to a final prediction model. Finally, the final prediction model is used to predict the wear pattern. This method solves the technical problem that existing friction and wear tests cannot predict wear patterns and require the use of equipment such as scanning electron microscopes for wear pattern determination. It can acquire experimental data in real time and perform real-time data analysis while conducting friction and wear tests, predicting all wear patterns of the material throughout the entire operating process. This reduces the use of experimental equipment, saves time and costs, and provides a reference for product design, manufacturing, and engineering applications.

[0048] 2. This wear testing machine has a fixed grinding mechanism, and the test sample moves back and forth with the tooling. Temperature and noise sensors are aligned with the contact part between the grinding mechanism and the test sample to collect friction coefficient, temperature and noise data in real time. Finally, the wear mode prediction method is used to predict the wear mode that a certain material will experience under a certain working condition. The wear mode can be predicted while the test is being conducted, which improves the prediction efficiency and timeliness. Attached Figure Description

[0049] Figure 1 This is a flowchart of the wear pattern prediction method of Embodiment 1 of the present invention.

[0050] Figure 2 This is a flowchart of the wear pattern prediction method in Embodiment 2 of the present invention.

[0051] Figure 3 This is a schematic diagram of the wear testing machine according to Embodiment 3 of the present invention.

[0052] Figure 4 for Figure 3 A schematic diagram of the main structure of the wear testing machine.

[0053] Symbol explanation:

[0054] 1. Frame 5. Temperature sensor

[0055] 2. Power mechanism; 6. Noise sensor

[0056] 3. Tooling 7. Test Specimen

[0057] 4. Grinding mechanism Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] Example 1

[0060] Please see Figure 1 This embodiment provides a method for predicting wear patterns. This method mainly achieves the prediction of wear patterns on test samples through the following steps. It should be noted that in practical applications, the test sample can be the finished product to be tested; that is, the method of this embodiment can be applied not only to testing but also directly to product inspection.

[0061] The first step involves conducting wear experiments to obtain three types of test data: friction coefficient, temperature, and noise. In this first step of the wear pattern prediction method, test samples made of different materials are used to collect test data from various materials. This allows for the elimination of the influence of different materials on the accuracy of the prediction results when extracting features to train the model, thus obtaining a wear pattern prediction method applicable to any material. Similarly, multiple sets of working conditions are set to eliminate the influence of working conditions on the accuracy of the prediction results, making the wear pattern prediction method applicable to any working condition. Working condition types include: load, relative speed, and test time.

[0062] The second step is to label and encode various wear patterns. In this embodiment, the test sample can be observed using equipment such as an electron microscope. Based on the sample morphology, the wear pattern is identified as one of adhesive wear, abrasive wear, fatigue wear, and oxidative wear. These four wear patterns are labeled and coded as 1, 2, 3, and 4. Adhesive wear refers to irregular adhesive nodules visible on the sample surface, with localized tearing edges and pits caused by plastic deformation. Abrasive wear refers to parallel furrows and directional scratches visible on the sample surface. Fatigue wear refers to shell-like or fish-scale-like peeling areas on the sample surface with fatigue cracks at the edges. Oxidative wear refers to the formation of a dense oxide layer on the sample surface with localized cracking. The four wear patterns are labeled and coded as follows: Adhesive wear → 1, Abrasive wear → 2, Fatigue wear → 3, Oxidative wear → 4.

[0063] The third step involves first extracting the time-domain features of the three types of experimental data, then converting the time-domain features into frequency-domain data, and finally extracting the frequency-domain features from the frequency-domain data. In this embodiment, the time-domain data X(t) is converted into frequency-domain data S(k) using Fourier transform. t represents time, T represents the end time point, k represents the k-th frequency component, k=0 represents the DC component, k=1 represents the fundamental frequency, and j represents the imaginary unit. The rotation factor is a complex exponential basis function projected from the time-domain signal onto the frequency domain. Time-domain characteristics include the peak value f. X,1 Mean f X,2 Root mean square f X,3 Standard deviation f X,4 , kurtosis f X,5 skewness f X,6 Margin index f X,7 Waveform index f X,8 、, impulse index f X,9 Frequency domain characteristics include the frequency mean F. S,1 Frequency centroid F S,2 Root mean square frequency F S,3 Frequency variance F S,4 Where X∈{μ,Te,No}, μ is the coefficient of friction, Te is the temperature, and No is the noise.

[0064] The formula for calculating time-domain features is:

[0065] f X，1 =max(|X(t)|)

[0066]

[0067] Where X(t) represents the time-domain data of the experiment, t represents time, T represents the end time point, and n represents the number of valid samples.

[0068] The frequency domain data is S(k), and the calculation formula is:

[0069]

[0070] The formula for calculating frequency domain characteristics is:

[0071]

[0072] Among them, f k This represents the frequency value of the k-th spectral line. In this embodiment, the third step of the wear pattern prediction method yields 13 features for each of the three data types.

[0073] The fourth step involves building three prediction models for the three types of data. This step first encodes the time-domain / frequency-domain features and wear pattern labels of the three types of experimental data to form a dataset and generates multiple training subsets. Then, a decision tree is constructed for each training subset, and this process is repeated to obtain multiple decision trees. Each decision tree recursively splits from the root node until a node meets a preset stopping condition and is marked as a leaf node. Finally, a voting method is used to select the category with the majority prediction as the final result, and a preliminary prediction function is obtained, resulting in three preliminary prediction models.

[0074] In this embodiment, H training subsets are generated from the dataset using Bootstrap sampling (random sampling with replacement). Each training subset comprises 2 / 3 of the original data, and the remaining 1 / 3 of the original data is used as out-of-bag (OOB) data. A decision tree is constructed for each training subset. The method for constructing a decision tree includes the following steps: First, at each node of the tree, m features are randomly selected (m=3 can be chosen). Then, the optimal feature is selected for node splitting based on the Gini index as the splitting criterion. Where p g It is the proportion of the g-th class samples in the node. Finally, recursive splitting is performed, and feature selection and optimal split point selection are repeated until the number of node samples is less than a set threshold or the maximum depth is reached.

[0075] The datasets for the three types of experimental data are as follows:

[0076]

[0077] Among them, D μ D Te D No Data sets representing friction coefficient, temperature, and noise, respectively; The 13-dimensional feature vector is composed of time-domain / frequency-domain features; Yi∈{1,2,3,4} represents the category label of the four wear forms.

[0078] In this embodiment, the above process is repeated to generate H decision trees. Each decision tree starts from the root node and recursively splits until a stopping condition is met. At internal nodes, samples are split based on the selected optimal feature and threshold. When a node meets the stopping condition, it is marked as a leaf node, and the prediction result is determined based on the majority class.

[0079] The fifth step is to determine whether the three preliminary prediction functions have a mode. If so, the mode of the three preliminary prediction functions is selected as the final prediction function; otherwise, a weighted function is established to determine the final prediction function, thus generating the final prediction model. If the result has a mode, then the final prediction function is... If... The results were all different, so a weighted function was established to determine the final prediction function.

[0080] In this embodiment, the method for determining the final prediction function through a weighted function includes the following steps:

[0081] (1) Let Calculate the deviation: in, Let each represent a preliminary prediction function. The preliminary prediction function for the friction coefficient is:

[0082]

[0083] Among them, T b Let b represent the b-th decision tree, B represent the total number of decision trees, and mode(·) represent the mode.

[0084] (2) Calculate the weighting factors:

[0085] (3) Let the weighting function be The final prediction function is then... In Chinese, round(·) means rounding to the nearest integer.

[0086] The sixth step is to predict the wear pattern based on the final prediction model.

[0087] In summary, the wear pattern prediction method of this embodiment has the following beneficial effects:

[0088] This wear pattern prediction method first obtains three experimental data points: friction coefficient, temperature, and noise. Then, it extracts features from each set of experimental data using time-domain and frequency-domain feature extraction methods, corresponding to different wear patterns. A dataset is then constructed based on these features and their corresponding wear patterns. A decision tree is built and trained on the data to obtain three preliminary prediction functions. The final prediction function is obtained by determining if the three preliminary prediction functions have a mode, leading to a final prediction model. Finally, the final prediction model is used to predict the wear pattern. This method solves the technical problem that existing friction and wear tests cannot predict wear patterns and require equipment such as scanning electron microscopes for wear pattern determination. It can acquire experimental data in real time and perform real-time data analysis while conducting friction and wear tests, predicting all wear patterns of the material throughout the entire operating process. This reduces the use of experimental equipment, saving time and costs, and providing a reference for product design, manufacturing, and engineering applications.

[0089] Example 2

[0090] Please see Figure 2This embodiment provides a method for predicting wear patterns, which adds some steps to the method described in Embodiment 1. This embodiment also establishes a correction model, which predicts in time periods: a time period ΔT is set, and the prediction result of the correction model is compared with the prediction result of the previous time period. If the prediction results of the two models are different in the time periods hΔT to (h+1)ΔT, the correction model updates the prediction data of the final prediction model, and the time hΔT when the prediction result of the correction model begins to change is taken as the starting time of the updated final prediction model. The final prediction model is then recalculated. During output, the prediction result of the final prediction model before time hΔT is output as one wear pattern result, and the prediction result after time hΔT is output as another wear pattern result, thus predicting multiple wear patterns in time periods throughout the entire operating process.

[0091] Because the predictive method for wear pattern prediction takes into account the possibility that materials may exhibit different wear patterns at different times, a correction model is added. The calculation method of the correction model is the same as that of the prediction model, but the correction model makes predictions in different time periods. In the model prediction stage, under a certain working condition test for a certain material, three types of test data are first collected in real time by sensors. Then, the correction model is used to predict the wear pattern in different time periods. Subsequently, the prediction results of the correction model are checked to see if they change. If they do, the final prediction model updates the data and outputs the results for different time periods; otherwise, the final prediction model continues to calculate and output the results, finally obtaining the wear pattern results for the entire working condition process.

[0092] Example 3

[0093] Please see Figure 3 as well as Figure 4 This embodiment provides a wear testing machine, which is applied to the wear pattern prediction method in Embodiment 1 or 2. The wear testing machine includes a frame 1, a power mechanism 2, a fixture 3, a grinding mechanism 4, a friction force sensor, a temperature sensor 5, and a noise sensor 6. The power mechanism 2 is mounted on the frame 1, the fixture 3 is mounted on the frame 1, and the grinding mechanism 4 is fixed to the frame 1. The power mechanism 2 drives the fixture 3 to reciprocate on the frame 1, causing the fixture 3 to bring the test sample 7 into contact with the grinding mechanism 4, thus causing wear on the test sample 7. The friction force sensor is mounted on the frame 1 and is used to detect the friction force generated by the wear of the test sample 7. The wear testing machine calculates the coefficient of friction based on the friction force. The temperature sensor 5 is mounted on the frame 1 and is used to detect the temperature generated by the wear of the test sample 7. The noise sensor 6 is mounted on the frame 1 and is used to detect the noise generated by the wear of the test sample 7.

[0094] The wear testing machine has a fixed grinding mechanism 4, and the test sample 7 reciprocates with the tooling 3. The temperature sensor 5 and the noise sensor 6 are both aligned with the contact part between the grinding mechanism 4 and the test sample 7 to collect friction coefficient, temperature and noise data in real time. Finally, the wear mode prediction method is used to predict the wear mode that a certain material will undergo under a certain working condition. The wear mode can be predicted while the test is being conducted, which improves the prediction efficiency and timeliness.

[0095] Example 4

[0096] This embodiment provides a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the wear pattern prediction method of Embodiment 1.

[0097] The method in Example 1 can be applied in software form, such as by designing it as a standalone program and installing it on a computer terminal, which can be a computer, smartphone, control system, or other IoT device. Alternatively, the method in Example 1 can be designed as an embedded program and installed on a computer terminal, such as on a microcontroller.

[0098] Example 5

[0099] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it implements the steps of the wear pattern prediction method of Embodiment 1.

[0100] When applying the method of Example 1, it can be applied in the form of software, such as by designing it as a program that can run independently on a computer-readable storage medium. The computer-readable storage medium can be a USB flash drive, designed as a USB security token, and the program can be designed to start the entire method through an external trigger.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting wear patterns, characterized in that, It includes the following steps: Wear tests were conducted to obtain three types of test data: coefficient of friction, temperature, and noise. Label and encode various wear patterns; First, extract the time-domain features of the three types of experimental data respectively, then convert the time-domain features into frequency-domain data, and finally extract the frequency-domain features of the frequency-domain data. First, the time-domain / frequency-domain features of the three experimental data and the wear form labels are encoded to form a dataset and generate multiple training subsets. Then, a decision tree is built for each training subset and the process is repeated to obtain multiple decision trees. Each decision tree is recursively split from the root node until the node meets a preset stopping condition and is marked as a leaf node. Finally, the majority of predicted categories are selected as the final result, and a preliminary prediction function is obtained. Determine whether the three preliminary prediction functions have a mode. If they do, the mode of the three preliminary prediction functions is determined as the final prediction function. Otherwise, establish a weighted function to determine the final prediction function and generate the final prediction model. Wear patterns are predicted based on the final prediction model.

2. The wear pattern prediction method as described in claim 1, characterized in that, The time-domain features include peak value f X,1 Mean f X,2 Root mean square f X,3 Standard deviation f X,4 , kurtosis f X,5 skewness f X,6 Margin index f X,7 Waveform index f X,8 Pulse index f X,9 The frequency domain features include the frequency mean F. S,1 Frequency centroid F S,2 Root mean square frequency F S,3 Frequency variance F S,4 Where X∈{μ,Te,No}, μ is the friction coefficient, Te is the temperature, and No is the noise.

3. The wear pattern prediction method as described in claim 2, characterized in that, The formula for calculating the time-domain features is as follows: f X，1 =max(|X(t)|) Where X(t) represents the time-domain data of the experiment, t represents time, T represents the end time point, and n represents the number of valid samples.

4. The wear pattern prediction method as described in claim 3, characterized in that, The frequency domain data is S(k), and the calculation formula is: The formula for calculating the frequency domain features is as follows: Among them, f k Let be the frequency value of the k-th spectral line.

5. The wear pattern prediction method as described in claim 4, characterized in that, The wear mode is one of adhesive wear, abrasive wear, fatigue wear and oxidative wear, and the four wear modes are labeled and coded as 1, 2, 3 and 4; H training subsets are generated from the dataset using Bootstrap sampling. Each training subset accounts for 2 / 3 of the original data, and the remaining 1 / 3 of the original data is used as out-of-bag data. The method for constructing the decision tree includes the following steps: first, randomly select m features when splitting each node of the tree; then, select the best feature for node splitting based on the Gini index as the splitting criterion; finally, perform recursive splitting, repeating feature selection and the selection of the best splitting point until the number of node samples is less than a set threshold or the maximum depth is reached.

6. The wear pattern prediction method as described in claim 5, characterized in that, The datasets for the three types of experimental data are as follows: Among them, D μ D Te D No These represent the datasets for the friction coefficient, the temperature, and the noise, respectively. The 13-dimensional feature vector composed of the time-domain / frequency-domain features is represented; Yi∈{1,2,3,4} represents the category label of the four wear forms.

7. The wear pattern prediction method as described in claim 6, characterized in that, The method for determining the final prediction function using the weighting function includes the following steps: make Calculate the deviation: in, These represent three preliminary prediction functions; Calculate the weighting factors: Let the weighting function be The final prediction function is then... The round(·) function indicates rounding to the nearest integer.

8. The wear pattern prediction method as described in claim 7, characterized in that, The preliminary prediction function for the friction coefficient is: Among them, T b Let b represent the b-th decision tree, B represent the total number of decision trees, and mode(·) represent the mode.

9. The wear pattern prediction method as described in claim 8, characterized in that, A correction model is also established, which makes time-segmented predictions: a time period ΔT is set, and the prediction results of the correction model are compared with the prediction results of the previous time period. If the prediction results of the two models are different in the time period hΔT to (h+1)ΔT, the correction model updates the prediction data of the final prediction model, and the hΔT time when the prediction results of the correction model begin to change is taken as the starting time of the updated final prediction model. The final prediction model is recalculated. During output, the prediction results of the final prediction model before hΔT time are output as one type of wear pattern, and the prediction results after hΔT time are output as another type of wear pattern, thus predicting multiple wear patterns in time periods throughout the entire working process.

Images