Cutter wear monitoring method based on IWOA-GRU

Abstract

The invention discloses a tool wear monitoring method based on IWOA-GRU, belongs to the field of metal cutting monitoring, and aims to solve the problem of low tool wear state monitoring precision caused by incapability of capturing a long-term dependency relationship in the prior art. The method comprises the following steps of 1, collecting tool machining process data to obtain an original data set; 2, processing the original data set, and dividing the original data set into a training set and a test set; 3, inputting the training set into the GRU model for training; 4, optimizing a weight matrix of the GRU model by adopting an IWOA whale population algorithm; and 5, the trained IWOA-GRU model is used for monitoring tool wear in the tool machining process, and the tool wear state is judged according to the tool wear value Yt output by the model.

Description

Technical Field

[0001] The invention relates to a tool wear monitoring method based on IWOA-GRU, and belongs to the field of metal cutting monitoring. Background Art

[0002] With the development of manufacturing technology in the field of modern aerospace, the performance of aircraft materials has gradually improved. Most of them use materials with comprehensive performance of high temperature resistance and corrosion resistance that are difficult to process. The parts have complex structures and high requirements for processing accuracy. In the process of part cutting, due to the interaction between the tool and the workpiece, the cutting edge of the tool will inevitably be worn, thus affecting the processing accuracy, so it is very important to monitor the wear state of the tool. The use of traditional processing methods will cause the tool to wear quickly and severely during the cutting process, thereby seriously reducing the material performance and bearing capacity. Among these difficult-to-process materials, due to workpiece damage and machine tool failure caused by tool failure, most tools are forced to be scrapped before reaching their service life, which results in about 20% to 25% of the tool life not being effectively used, increasing the processing cost. Therefore, it is a big challenge to manufacture and process high-precision aviation parts that meet the requirements of the current scientific and technological level. The tool wear state affects the processing accuracy, efficiency and economic benefits. Therefore, it is of great significance to monitor the tool wear state during the processing of aviation structural parts, which can greatly improve the quality and production efficiency of cutting processing and reduce costs. Generally, tool replacement is determined based on the operator's subjective judgment. If the tool is replaced too early, resources will be wasted and processing costs will increase; if the tool is not replaced in time, the workpiece processing quality will decline and production efficiency will be reduced. Studies have found that if the tool wear can be accurately monitored, the accuracy of machined parts can be improved, 10%-40% of the total processing cost can be saved, and the cutting speed can be increased by 10%-50%. When processing long time series, traditional RNNs are prone to the problem of gradient disappearance, that is, during the back propagation process, the gradient gradually decreases, resulting in the model being unable to effectively learn the long-term dependency between tool wear values ​​and sensor data length. Summary of the invention

[0003] In view of the problem that the prior art cannot capture long-term dependencies and thus leads to low accuracy in tool wear status monitoring, the present invention provides a tool wear monitoring method based on IWOA-GRU.

[0004] The present invention provides a tool wear monitoring method based on IWOA-GRU, the method comprising the following steps:

[0005] Step 1: Collect tool processing data to obtain the original data set, the original data set is Y t =F(X t ), where Y t is the tool wear value, Xt is the sensor data for online monitoring, in is the vibration signal, is the acoustic emission signal, Is the spindle current signal;

[0006] Step 2: Process the original data set in step 1 and divide it into a training set and a test set;

[0007] Step 3: Input the training set into the IWOA-GRU model, wherein the GRU model is used to train the training set, and the IWOA whale population algorithm is used to optimize the weight matrix of the GRU model, wherein the individuals of the initial population of the IWOA whale population algorithm come from the training set; Step 4: The trained IWOA-GRU model is used to monitor the tool wear during tool processing, and the tool wear value Y output by the model is used to monitor the tool wear during tool processing. t Determine the tool wear status.

[0008] Preferably, the process of processing the original data set in step 2 is:

[0009] Step 2: Preprocessing step: the time domain signal x collected for each cutting process i Filtering and smoothing are performed, and null values, abnormal values ​​and repeated values ​​are removed; the time domain signal x i Refers to the sensor data collected during each cutting process, i = 1, 2, ..., N, N is the length of the intercepted sensor data;

[0010] Step 22: extracting three-domain eigenvalues, wherein the three-domain eigenvalues ​​include time domain eigenvalues, frequency domain eigenvalues ​​and time-frequency domain eigenvalues;

[0011] The time domain characteristic values ​​include mean value, standard deviation, root mean square, peak value, pulse factor, kurtosis, shape coefficient and skewness. The time domain characteristic values ​​are obtained by i Processing acquisition;

[0012] The frequency domain characteristic values ​​include maximum frequency, minimum frequency, frequency average, center of gravity frequency, frequency deviation and mean square frequency. i Perform Fourier transform to obtain;

[0013] The time-frequency domain eigenvalues ​​are obtained by wavelet packet transform, which is decomposed to generate multi-band data, and then the multi-band data at each moment are spliced ​​into time-frequency domain eigenvalues;

[0014] Step 2 and 3: Use the principal component analysis method to reduce the dimension of the three-domain features, retain the top ten three-domain features with the highest contribution rate, and then filter out the valid data corresponding to the top ten three-domain features with the highest contribution rate in the original data set. The filtered data set is divided into a training set and a test set.

[0015] Preferably, the time domain eigenvalue is obtained according to the following formula:

[0016] average value Standard Deviation RMS

[0017] Peak X pv =max|x i |, Pulse Factor Kurtosis

[0018] Form Factor Skewness

[0019] Preferably, the frequency domain eigenvalue is obtained by: i Perform Fourier transform to obtain the frequency f i and the corresponding amplitude p i , the frequency domain eigenvalue is obtained according to the following formula:

[0020] Maximum frequency f max =max(f1,f2,...,f N ), minimum frequency f min =min(f1,f2,...,f N ),

[0021] Frequency average Center of gravity frequency

[0022] Frequency Deviation Mean square frequency

[0023] Preferably, the process of optimizing the weight matrix of the GRU model using the IWOA whale population algorithm in step 3 is:

[0024] Extract m data from the training set as whale individuals in the population and set the maximum number of iterations;

[0025] The IWOA whale population algorithm finds the optimal whale individual position in each iteration. And update the positions of other individual whales;

[0026] Based on the individual whale positions Get the update gate weight matrix W used to optimize the GRU model zand reset the gate weight matrix W r :

[0027]

[0028] in, is the optimal individual whale position at the tth iteration, W rand is the random position of individual whales, W z (t) and W r (t) is the update gate weight matrix and reset gate weight matrix of the tth iteration, W z (t+1) and W r (t+1) is the update gate weight matrix and reset gate weight matrix of the predicted t+1th iteration, which is used to feed back to the GRU model for updating;

[0029] H is the scaling factor,

[0030] H=2·e -λ·t

[0031] Here, λ is a constant.

[0032] Preferably, the judgment principle of the tool wear state in step 4 is:

[0033] Tool wear value Y t =[0,0.07), indicating that the tool is in the primary wear state;

[0034] Tool wear value Y t =[0.07,0.12), indicating that the tool is in a normal wear state;

[0035] Tool wear value Y t =[0.12,0.3), indicating that the tool is in a state of rapid wear.

[0036] Beneficial effects of the invention: This project intends to study a method for intelligent monitoring of tool wear status, establish an IWOA-GRU deep learning model based on the cutting data set after feature extraction and screening, continuously optimize and train the model, ensure the adaptability and robustness of the model, improve the model training efficiency and monitoring accuracy, so as to achieve the goal of accurately monitoring the tool wear status, thereby extending the tool life, reducing costs and improving production efficiency.

[0037] The parameters of GRU are optimized through the whale population algorithm. Compared with the currently popular prediction models such as SVM, Transformer and BP, IWOA-GRU is more flexible and adaptable in time series prediction. IWOA optimizes the parameters of GRU, which is higher than other models in model training efficiency and prediction accuracy, and its adaptability and robustness are also improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of a tool wear monitoring method based on IWOA-GRU according to the present invention;

[0039] Figure 2 This is the IWOA-GRU model joint training flow chart;

[0040] Figure 3 It is a schematic diagram of the gated recurrent unit GRU;

[0041] Figure 4 This is a schematic diagram of the update gate of GRU;

[0042] Figure 5 This is a schematic diagram of the reset gate of GRU;

[0043] Figure 6 It is a schematic diagram of the implicit output state of GRU;

[0044] Figure 7 It is a schematic diagram of the candidate hidden output states of GRU. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0047] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0048] Specific implementation method 1: The following is combined Figures 1 to 7 The present embodiment describes a tool wear monitoring method based on IWOA-GRU, which comprises the following steps:

[0049] Step 1: Collect tool processing data to obtain the original data set, the original data set is Y t =F(X t ), where Y t is the tool wear value, X t is the sensor data for online monitoring, in is the vibration signal, is the acoustic emission signal, Is the spindle current signal;

[0050] Step 2: Process the original data set in step 1 and divide it into a training set and a test set;

[0051] Step 3: Input the training set into the IWOA-GRU model, wherein the GRU model is used to train the training set, and the IWOA whale population algorithm is used to optimize the weight matrix of the GRU model, wherein the individuals of the initial population of the IWOA whale population algorithm come from the training set;

[0052] Step 4: The trained IWOA-GRU model is used to monitor tool wear during tool processing. The tool wear value Y output by the model is t Determine the tool wear status.

[0053] The original data set in step 1 is constructed by a large amount of data acquisition during the tool processing process, in which the data of each sensor is collected online, while the tool wear value is collected offline.

[0054] The data in the original data set cannot be used directly for model training. The original data set needs to be processed to extract valid data for model training, so that the model training can be accurate and the output results can be accurate.

[0055] The process of processing the original data set in step 2 is:

[0056] Step 2: Preprocessing step: the time domain signal x collected for each cutting process i Filtering and smoothing are performed, and null values, abnormal values ​​and repeated values ​​are removed; the time domain signal x i Refers to the sensor data collected during each cutting process, i = 1, 2, ..., N, N is the length of the intercepted sensor data;

[0057] Step 22: extracting three-domain eigenvalues, wherein the three-domain eigenvalues ​​include time domain eigenvalues, frequency domain eigenvalues ​​and time-frequency domain eigenvalues;

[0058] Time domain analysis reflects the changes in the waveform of the cutting process parameter signal on the time axis. The characteristics retained by the time domain signal are relatively comprehensive. The time domain characteristic values ​​include mean value, standard deviation, root mean square, peak value, pulse factor, kurtosis, shape coefficient and skewness. The time domain characteristic values ​​are obtained by i Processing obtained.

[0059] The time domain eigenvalue is obtained according to the following formula:

[0060] average value Standard Deviation RMS

[0061] Peak X pv =max|x i |, Pulse Factor Kurtosis

[0062] Form Factor Skewness

[0063] Time domain analysis is effective in stable periodic signals and can more intuitively reflect the changing trend of tool wear. However, most of the signals collected in actual processing are non-stationary signals. Therefore, analyzing the time domain characteristics cannot fully represent the characteristics of the signal. As the tool wear increases, the frequency of the signal often changes. The time domain signal is transformed through Fast Fourier Transform (FFT) to obtain the relationship between frequency and amplitude, and then analyze the frequency domain characteristics of the tool as it changes with wear. The frequency domain characteristic values ​​include maximum frequency, minimum frequency, frequency average, center of gravity frequency, frequency deviation and mean square frequency. By analyzing the time domain signal x i Perform Fourier transform to obtain;

[0064] Obtaining frequency domain eigenvalues: For the time domain signal x i Perform Fourier transform to obtain the frequency f i and the corresponding amplitude p i , the frequency domain eigenvalue is obtained according to the following formula:

[0065] Maximum frequency f max =max(f1,f2,...,f N ), minimum frequency f min =min(f1,f2,...,f N ),

[0066] Frequency average Center of gravity frequency

[0067] Frequency Deviation Mean square frequency

[0068] The common features of the time domain or frequency domain describe the changes in the tool wear state during the entire cutting process, but the frequency information of the non-steady-state signal changing with time cannot be observed, nor can the changes in the cutting process parameter signal be analyzed locally, that is, the resolution of the time frequency is not high. With the help of time-frequency domain analysis, this feature extraction method can determine the state information of tool wear. In order to obtain more comprehensive information related to tool wear, it is necessary to analyze the signal locally and as a whole at the same time. This embodiment uses wavelet packet transform to perform time-frequency domain analysis. The wavelet packet transform decomposes to generate multi-band data, and then splices the multi-band data at each moment into time-frequency domain feature values;

[0069] The time domain, frequency domain and time-frequency domain features of the cutting process parameter signal are extracted. The extracted features vary with the different tool wear states, but not all of these features are highly correlated with the tool wear state. Using all of them as input to the model will result in information redundancy, which will reduce the model operation time and even overfitting, resulting in performance degradation. Therefore, in order to effectively monitor the tool state, it is necessary to extract effective information related to tool wear. This implementation method uses principal component analysis for screening.

[0070] Step 2 and 3: Use the principal component analysis method to reduce the dimension of the three-domain features, retain the top ten three-domain features with the highest contribution rate, and then filter out the valid data corresponding to the top ten three-domain features with the highest contribution rate in the original data set. The filtered data set is divided into a training set and a test set.

[0071] Step 3: Input the training set into the GRU model for training;

[0072] The Gated Recurrent Unit (GRU) is a deep learning model specifically designed for processing time series data. Figure 3 As shown. GRU contains 1 memory unit 2 gates: Update gate z t and reset gate r t , x t represents the input data at the current time t, h t-1 Represents the implicit state output at the historical time t-1. σ and tanh represent the sigmoid function and the hyperbolic tangent function respectively.

[0073] See also Figure 4 , update gate z t : Determine how much historical information of the last training dataset and the current training dataset information to use to update the current implicit state. The update gate at time t is:

[0074] z t =σ(W z ·[h t-1 ,x t ])

[0075] Among them, z t is the gated update signal, whose size determines the memory degree of the candidate hidden state, h t-1 is the historical implicit state, x t represents the new input data at time t, W z is the weight matrix.

[0076] See also Figure 5 , reset gate r t: It mainly determines how much of the trained vibration, acoustic emission, current signal and data set information is retained, and how much of the past vibration, acoustic emission, current signal and data set information needs to be forgotten. The reset gate at the tth moment is:

[0077] r t =σ(W r ·[h t-1 ,x t ])

[0078] Among them, r t is the reset signal. The larger the reset signal, the more historical information needs to be remembered. r is the weight matrix.

[0079] See also Figure 6 , implicit output state h t : Under the action of the update gate and the reset gate, the implicit output state h at the current moment t Can be updated to:

[0080]

[0081] Among them, h t Represents the final implicit output state of the data set at the current moment, including the two processes of memory and forgetting, z t The larger the element value in z is, the t and h t-1 The larger the Hadamard product, the higher the vibration, acoustic emission, current signal and data set at the previous moment. t-1 The higher the degree of information retention in the t The smaller the element value in is, the more the new implicit state comes from

[0082] See also Figure 7 , candidate hidden states

[0083]

[0084] The input new vibration, acoustic emission, current signal and data and the historical implicit state vibration, acoustic emission, current signal and its data set feature information, this operation and the reset signal r obtained by the reset gate t Related, the candidate hidden state is obtained through the tanh function.

[0085] GRU is good at processing sequence data, but it is easy to fall into local optimal solutions and has low convergence accuracy. This paper uses the optimization algorithm of IWOA to enhance the parameter tuning ability of GRU. The improved whale algorithm optimizes the gated recurrent unit time prediction. The model has good generalization ability. The core idea is to optimize the parameters of GRU through the whale algorithm.

[0086] The process of using the IWOA whale population algorithm to optimize the weight matrix of the GRU model in step 3 is:

[0087] Extract m data from the training set as whale individuals in the population and set the maximum number of iterations;

[0088] The IWOA whale population algorithm finds the optimal whale individual position in each iteration. And update the positions of other individual whales;

[0089] Based on the individual whale positions Get the update gate weight matrix W used to optimize the GRU model z and reset the gate weight matrix W r :

[0090]

[0091] in, is the optimal individual whale position at the tth iteration, W rand is the random position of individual whales, W z (t) and W r (t) is the update gate weight matrix and reset gate weight matrix of the tth iteration, and the initial value W at t = 0 z (0) and W r (0) is a random value, W z (t+1) and W r (t+1) is the update gate weight matrix and reset gate weight matrix of the predicted t+1th iteration, which is used to feed back to the GRU model for updating;

[0092] H is the scaling factor, which is used to control the search step size;

[0093] H=2·e -λ·t

[0094] Here, λ is a constant.

[0095] The maximum number of iterations is set in the IWOA optimization algorithm. Each iteration updates the optimal position parameters of the population individuals. The optimal solution is the individual position closest to the prey position among the individuals in the population composed of the data set, that is, the most suitable vibration, acoustic emission and current signals and their data sets. Then the positions of other individuals are changed. Then IWOA improves the traditional WOA model through adaptive weight strategy and random difference mutation strategy. The model is prone to fall into local optimality and premature convergence during the later local development. The individual position is updated again to achieve better optimization effect. Then it is mapped back to the GRU model, and the weight matrix and other parameters of the GRU model are optimized to replace the original parameters. For the specific process of IWOA, see Figure 2 shown.

[0096] The judgment principle of tool wear status in step 4 is:

[0097] Tool wear value Y t =[0,0.07), indicating that the tool is in the primary wear state;

[0098] Tool wear value Y t =[0.07,0.12), indicating that the tool is in a normal wear state;

[0099] Tool wear value Y t =[0.12,0.3), indicating that the tool is in a state of rapid wear.

[0100] Decide whether to replace the tool based on the tool wear status.

[0101] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that the features described in conjunction with a single embodiment may be used in other described embodiments.

Claims

1. A tool wear monitoring method based on IWOA-GRU, characterized in that: The method comprises the following steps: Step 1: Collect tool processing data to obtain the original data set, the original data set is Y t =F(X t ), where Y t is the tool wear value, X t is the sensor data for online monitoring, in is the vibration signal, is the acoustic emission signal, Is the spindle current signal; Step 2: Process the original data set in step 1 and divide it into a training set and a test set; Step 3: Input the training set into the IWOA-GRU model, wherein the GRU model is used to train the training set, and the IWOA whale population algorithm is used to optimize the weight matrix of the GRU model, wherein the individuals of the initial population of the IWOA whale population algorithm come from the training set; Step 4: The trained IWOA-GRU model is used to monitor tool wear during tool processing. The tool wear value Y output by the model is t Determine the tool wear status.

2. The tool wear monitoring method based on IWOA-GRU according to claim 1, characterized in that: The process of processing the original data set in step 2 is: Step 2: Preprocessing step: the time domain signal x collected for each cutting process i Filtering and smoothing are performed, and null values, abnormal values ​​and repeated values ​​are removed; the time domain signal x i Refers to the sensor data collected during each cutting process, i = 1, 2, ..., N, N is the length of the intercepted sensor data; Step 22: extracting three-domain eigenvalues, wherein the three-domain eigenvalues ​​include time domain eigenvalues, frequency domain eigenvalues ​​and time-frequency domain eigenvalues; The time domain characteristic values ​​include mean value, standard deviation, root mean square, peak value, pulse factor, kurtosis, shape coefficient and skewness. The time domain characteristic values ​​are obtained by i Processing acquisition; The frequency domain characteristic values ​​include maximum frequency, minimum frequency, frequency average, center of gravity frequency, frequency deviation and mean square frequency. i Perform Fourier transform to obtain; The time-frequency domain eigenvalues ​​are obtained by wavelet packet transform, which is decomposed to generate multi-band data, and then the multi-band data at each moment are spliced ​​into time-frequency domain eigenvalues; Step 2 and 3: Use the principal component analysis method to reduce the dimension of the three-domain features, retain the top ten three-domain features with the highest contribution rate, and then filter out the valid data corresponding to the top ten three-domain features with the highest contribution rate in the original data set. The filtered data set is divided into a training set and a test set.

3. The tool wear monitoring method based on IWOA-GRU according to claim 2, characterized in that: The time domain eigenvalue is obtained according to the following formula: average value Standard Deviation RMS Peak X pv =max|x i |, Pulse Factor Kurtosis Form Factor Skewness 4. The tool wear monitoring method based on IWOA-GRU according to claim 2, characterized in that: Obtaining frequency domain eigenvalues: For the time domain signal x i Perform Fourier transform to obtain the frequency f i and the corresponding amplitude p i , the frequency domain eigenvalue is obtained according to the following formula: Maximum frequency f max =max(f1,f2,...,f N ), minimum frequency f min =min(f1,f2,...,f N ), Frequency average Center of gravity frequency Frequency Deviation Mean square frequency 5. The tool wear monitoring method based on IWOA-GRU according to claim 2, characterized in that: The process of using the IWOA whale population algorithm to optimize the weight matrix of the GRU model in step 3 is: Extract m data from the training set as whale individuals in the population and set the maximum number of iterations; The IWOA whale population algorithm finds the optimal whale individual position in each iteration. And update the positions of other individual whales; Based on the individual whale positions Get the update gate weight matrix W used to optimize the GRU model z and reset the gate weight matrix W r : in, is the optimal individual whale position at the tth iteration, W rand is the random position of individual whales, W z (t) and W r (t) is the update gate weight matrix and reset gate weight matrix of the tth iteration, W z (t+1) and W r (t+1) is the update gate weight matrix and reset gate weight matrix of the predicted t+1th iteration, which is used to feed back to the GRU model for updating; H is the scaling factor, H=2·e -λ·t Here, λ is a constant.

6. The tool wear monitoring method based on IWOA-GRU according to claim 1, characterized in that: The judgment principle of tool wear status in step 4 is: Tool wear value Y t =[0,0.07), indicating that the tool is in the primary wear state; Tool wear value Y t =[0.07,0.12), indicating that the tool is in a normal wear state; Tool wear value Y t =[0.12,0.3), indicating that the tool is in a state of rapid wear.

Images