Cutter wear state recognition method and device based on improved pelican algorithm

Abstract

The invention relates to a tool wear state recognition method and device based on an improved pelican algorithm. Through combination of the MSCNN and the BiLSTM network, the model can effectively capture the spatio-temporal characteristics of the machine tool cutter wear state data, the attention degree of the model on key information is further enhanced through introduction of an attention mechanism, and the recognition precision and stability are improved. Meanwhile, the IPOA algorithm is introduced to optimize the parameters of the multi-scale convolutional neural network of the model, and the optimization precision and the convergence speed are improved to a certain extent in combination with an adaptive inertia weight factor, Cauchy variation and a sparrow alert mechanism strategy.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for identifying tool wear status based on an improved Pelican algorithm. Background Art

[0002] Tool condition monitoring is of great significance in the manufacturing process and directly affects the quality and production efficiency of product processing. With the development of modern manufacturing technology, tool wear state identification is an important issue. Domestic and foreign scholars have carried out extensive research and proposed a variety of methods, including monitoring methods based on vibration signals, acoustic emission signals, and cutting force signals, as well as machine learning methods based on image processing and signal processing. For example, the tool wear state identification method driven by the fusion of digital twins and support vector machines (SVM); the tool wear state identification model based on the squirrel foraging algorithm (SSA), variational mode decomposition (VMD) and bidirectional long short-term memory network (BiLSTM); the tool wear state identification method based on the convolutional neural network (CNN) of VMD and image coding technology, etc.

[0003] Although these methods have improved the monitoring effect to a certain extent, they still have some shortcomings, such as sensitivity to environmental noise, complex and high-cost equipment, limited feature extraction capabilities, and poor adaptability, which leads to poor accuracy in the existing technology in identifying the wear status of the die cutter. Summary of the invention

[0004] Based on this, it is necessary to provide a tool wear state identification method and device based on the improved Pelican algorithm that can improve the identification accuracy in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for identifying tool wear status based on an improved Pelican algorithm. The method comprises:

[0006] Obtain the wear data of the tool during milling work;

[0007] Extracting temporal features of wear data through multi-scale convolutional neural network;

[0008] The time features are input into the bidirectional long short-term memory network to predict the wear degree of the tool;

[0009] Among them, the multi-scale convolutional neural network adopts the improved Pelican optimization algorithm for parameter optimization. The improved Pelican optimization algorithm introduces adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican optimization algorithm.

[0010] In one embodiment, the wear data includes a vibration signal generated by the tool during milling operation.

[0011] In one embodiment, the improved Pelican optimization algorithm includes:

[0012] In the population initialization stage, an initial spatial position matrix of the pelican population is generated; wherein the spatial position matrix represents the spatial position of each pelican individual in the pelican population;

[0013] In the stage of flying towards the prey, based on the initial spatial position matrix, for each pelican individual in the pelican population, the fitness function value of the pelican individual is compared with the fitness value of the prey, and the spatial position of the pelican individual is updated based on the comparison result; wherein, an adaptive inertia weight factor is introduced in the updating process of the spatial position to adjust the degree of dependence on the spatial position before the update; and the first spatial state matrix of the pelican population is obtained after the spatial positions of all the pelican individuals are updated;

[0014] In the stage of skimming over the water, the pelican population is divided into a first individual set and a second individual set; based on the first spatial state matrix, for the pelican individuals in the first individual set, when the fitness value of the pelican individuals is less than the average fitness value of the pelican population, the Cauchy mutation strategy is used to update the spatial position of the pelican individuals; based on the first spatial state matrix, for the pelican individuals in the second individual set, a sparrow alert mechanism is introduced to update the spatial position of the pelican individuals; the second spatial state matrix of the pelican population is obtained after the spatial positions of the pelican individuals in the first individual set and the second individual set are updated;

[0015] The second spatial state matrix is ​​used as the initial spatial position matrix of the next iteration process, and the steps of the flying toward prey phase and the skimming over the water surface phase are iteratively executed until a preset number of iterations is reached;

[0016] The target spatial position is output according to the fitness value from the second spatial state matrix obtained when the iteration stops.

[0017] In one embodiment, the adaptive inertia weight factor is related to the ratio between the current iteration number and the preset iteration number.

[0018] In one embodiment, during the water skimming stage, when the individual fitness value of the pelican is not less than the average fitness value of the pelican population, the spatial position is updated based on the current spatial position and the current number of iterations.

[0019] In one embodiment, the method further comprises:

[0020] Construct a model that combines multi-scale convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms.

[0021] In a second aspect, the present application also provides a tool wear state identification method device based on the improved Pelican algorithm. The device comprises:

[0022] A data acquisition module is used to obtain the wear data of the tool during milling operation;

[0023] A feature extraction module, used to extract the temporal features of wear data through a multi-scale convolutional neural network;

[0024] A prediction output module is used to input the time features into a bidirectional long short-term memory network to predict the wear degree of the tool;

[0025] Among them, the multi-scale convolutional neural network adopts the improved Pelican optimization algorithm for parameter optimization. The improved Pelican optimization algorithm introduces adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican optimization algorithm.

[0026] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the tool wear state identification method based on the improved Pelican algorithm when executing the computer program.

[0027] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the tool wear state identification method based on the improved Pelican algorithm when executed by a processor.

[0028] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the steps in the tool wear state identification method based on the improved Pelican algorithm when executed by a processor.

[0029] The above-mentioned tool wear state recognition method and device based on the improved Pelican algorithm can effectively capture the spatiotemporal characteristics of machine tool tool wear state data by combining a multi-scale convolutional neural network and a bidirectional long short-term memory network. At the same time, the improved Pelican optimization algorithm is introduced to optimize the parameters of the multi-scale convolutional neural network of the model, combined with the adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism strategy, which has a certain improvement in optimization accuracy and convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flowchart of an improved Pelican optimization algorithm in one embodiment;

[0031] Figure 2 A schematic diagram of the structure of an IPOA-MSCNN-BiLSTM-Attention model in one embodiment;

[0032] Figure 3 is a test curve of the F6 benchmark function of CEC2005 in one embodiment;

[0033] Figure 4 is a box plot corresponding to the CEC2005 test set in one embodiment;

[0034] Figure 5 A schematic diagram of the structure of a device platform for collecting wear data in one embodiment;

[0035] Figure 6 is a MSCNN-BiLSTM classification confusion matrix in one embodiment;

[0036] Figure 7 is a MSCNN-BiLSTM-Attention classification confusion matrix in one embodiment;

[0037] Figure 8 is a POA-MSCNN-BiLSTM classification confusion matrix in one embodiment;

[0038] Fig. 9 is a POA-MSCNN-BiLSTM-Attention classification confusion matrix in one embodiment;

[0039] Fig.10 is an IPOA-MSCNN-BiLSTM classification confusion matrix in one embodiment;

[0040] Fig.11 is the IPOA-MSCNN-BiLSTM-Attention classification confusion matrix in one embodiment;

[0041] Fig.12 is the accuracy of POA-MSCNN-BiLSTM-Attention in one embodiment;

[0042] Fig.13 It is the accuracy rate of IPOA-MSCNN-BiLSTM-Attention in one embodiment. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] The embodiment of the present application provides a method for identifying tool wear status based on an improved Pelican algorithm, comprising the following steps:

[0045] Step 102, acquiring wear data of the tool during milling operation.

[0046] Tool wear is a common but very important problem that directly affects machining quality and production efficiency. Different degrees of wear correspond to different working conditions. For example, at the same rotation speed, feed speed, and the same material workpiece at the same cutting depth, the milling capabilities of tools with different degrees of wear are different. Wear data can be obtained by monitoring the working conditions of the tool during milling.

[0047] In one embodiment, the wear data includes a vibration signal generated by the tool during milling operation.

[0048] Step 104: extracting the time characteristics of the wear data through a multi-scale convolutional neural network.

[0049] Multi-Scale Convolutional Neural Network (MSCNN) is a deep learning model that captures multi-level information in an image by extracting features at different scales. The main idea of ​​a multi-scale convolutional neural network is to capture features of different sizes and shapes by processing image data at multiple scales simultaneously. The convolution operation at each scale can extract different levels of information, thereby improving the robustness and accuracy of the model.

[0050] Step 105, input the time feature into the bidirectional long short-term memory network to predict the wear degree of the tool.

[0051] Bidirectional Long Short-Term Memory (BiLSTM) is a special recurrent neural network (RNN) structure that is very powerful in processing sequence data. Traditional LSTM models can only make predictions based on contextual information before that moment, while BiLSTM can understand and process sequence data more comprehensively by considering both forward and backward contextual information.

[0052] Among them, the multi-scale convolutional neural network adopts the improved Pelican Optimization Algorithm (IPOA) for parameter optimization. The improved Pelican Optimization Algorithm (POA) introduces adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican Optimization Algorithm.

[0053] By combining the MSCNN and BiLSTM networks, the model in this embodiment can effectively capture the spatiotemporal characteristics of machine tool tool wear status data, introduce the IPOA algorithm to optimize the parameters of the multi-scale convolutional neural network, and combine the adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism strategy to improve the performance of the multi-scale convolutional neural network in terms of optimization accuracy and convergence speed, thereby improving the overall effect of tool wear identification.

[0054] In one embodiment, the improved Pelican optimization algorithm includes:

[0055] In the population initialization stage, an initial spatial position matrix of the pelican population is generated; wherein the spatial position matrix represents the spatial position of each pelican individual in the pelican population;

[0056] In the stage of flying towards the prey, based on the initial spatial position matrix, for each pelican individual in the pelican population, the fitness function value of the pelican individual is compared with the fitness value of the prey, and the spatial position of the pelican individual is updated based on the comparison result; wherein, an adaptive inertia weight factor is introduced in the updating process of the spatial position to adjust the degree of dependence on the spatial position before the update; and the first spatial state matrix of the pelican population is obtained after the spatial positions of all the pelican individuals are updated;

[0057] In the stage of skimming over the water, the pelican population is divided into a first individual set and a second individual set; based on the first spatial state matrix, for the pelican individuals in the first individual set, when the fitness value of the pelican individuals is less than the average fitness value of the pelican population, the Cauchy mutation strategy is used to update the spatial position of the pelican individuals; based on the first spatial state matrix, for the pelican individuals in the second individual set, a sparrow alert mechanism is introduced to update the spatial position of the pelican individuals; the second spatial state matrix of the pelican population is obtained after the spatial positions of the pelican individuals in the first individual set and the second individual set are updated;

[0058] The second spatial state matrix is ​​used as the initial spatial position matrix of the next iteration process, and the steps of the flying toward prey phase and the skimming over the water surface phase are iteratively executed until a preset number of iterations is reached;

[0059] The target spatial position is output according to the fitness value from the second spatial state matrix obtained when the iteration stops.

[0060] The Pelican Optimization Algorithm imitates the natural predation behavior of pelicans. Based on this algorithm, this embodiment proposes an improved Pelican Optimization Algorithm, which combines the adaptive inertia weight factor, Cauchy mutation and warning mechanism strategy. The execution process is as follows: Figure 1 shown.

[0061] The basic idea of ​​IPOA is to divide the behavior of pelicans into three stages: the first stage is the population initialization stage; the second stage is the flight to prey stage; and the third stage is the skimming over the water surface stage.

[0062] In the population initialization stage, there are n pelicans in an m-dimensional space. Assume that the spatial position of the i-th pelican in this space is X i =[X i1 ,X i2 ,...,X in ], the position of the i-th pelican in the j-th dimension is x ij , where i = 1, 2, ..., n; j = 1, 2, ..., m, n is the number of pelicans, m is the dimension of the problem, α is a random number in the interval [0, 1], l j 、u j They represent the lower and upper boundaries of the problem in the jth dimension respectively. Then the mathematical model of the initial spatial position matrix X of these pelicans in this space can be expressed as:

[0063]

[0064] Pelican position initialization:

[0065] x ij = l j +α·(u j -l j ) (2)

[0066] In the flying-towards-prey phase, the pelican will identify the location of the prey and fly towards it. The random distribution of prey in the search space helps improve the global optimization ability of the algorithm. Assume that the position of the i-th pelican in the j-th dimension after the update in this phase is σ represents a random number in the interval [0,1], P j represents the position of the prey in the jth dimension, I represents a random integer in the interval [1,2], and F P represents the fitness value of the prey, F i represents the fitness value of the i-th pelican individual, ω represents the adaptive inertia weight factor, X best represents the best position vector found in the current iteration. During each iteration, the mathematical model of the new position of the pelican can be expressed as:

[0067]

[0068] An adaptive inertia weight factor ω is introduced at this stage. Balancing the local optimization and global search capabilities of the metaheuristic algorithm is a key factor in improving the optimization accuracy and speed of the algorithm. Since the position update of the individual pelican depends on its current position, it is necessary to adjust this dependence through the adaptive inertia weight factor ω. In the initial iteration stage of the algorithm, the ω value is small, which means that the individual position update is less affected by the current pelican position, which helps to expand the search range and enhance the global optimization ability; as the iteration proceeds, the ω value gradually increases, and the individual pelican position update depends more on the current position, which narrows the search range and helps to accurately find the optimal solution. This not only improves the local optimization ability of the algorithm, but also speeds up the convergence speed. t represents the current number of iterations, T represents the preset number of iterations, and the adaptive inertia weight factor ω can be expressed as:

[0069]

[0070] Assume that the new position of the i-th pelican is It means that after updating the position, the fitness value of the i-th pelican is If the fitness of the individual pelican is improved in the new position, the mathematical model of its position update can be expressed as:

[0071]

[0072] x i =x i ,Others(7)

[0073] After the positions of all pelicans in the pelican individual are updated during the stage of flying toward the prey, a first spatial state matrix is ​​constructed based on the updated spatial positions of the pelicans.

[0074] When the pelicans reach the water surface, they enter the skimming stage. They spread their wings and collect prey into their throat pouches. This process enhances the algorithm's local optimization ability. Assume that the position of the i-th pelican in the j-dimension after the update in the flying to prey stage is β represents a random number in the interval [0,1], R represents 0.2, t represents the current iteration number, and T represents the preset iteration number. For the first half of the pelican individuals in the population, i ranges from 1 to Loop, n represents the number of pelicans, X best represents the best position vector found in the current iteration. The mathematical model of the new position of the pelican can be expressed as:

[0075]

[0076] This embodiment introduces the Cauchy mutation strategy in the water-skimming stage of the pelican optimization algorithm. Specifically, in each iteration, the fitness value of the current pelican is compared with the average fitness value of the population. When the fitness value of the current pelican is less than the average fitness value of the population, it means that the pelican is in a gathering state. In order to increase diversity, the Cauchy mutation strategy is adopted; when the fitness value is not less than the average fitness value of the population, the original pelican position update method is continued. F avg Represents the average value of fitness, given the random variable rr i Generated by the Cauchy distribution, with mean 0, scale parameter σ=0.5, and dimension [N,1], the mathematical model of the Cauchy mutation can be expressed as:

[0077] rr=cauchyrnd(0,σ,[N,1]) (10)

[0078]

[0079] Assume that the new position of the i-th pelican is It means that after updating the position, the fitness value of the i-th pelican is If the fitness value of the individual pelican is improved in the new position, the mathematical model of its position update can be expressed as:

[0080]

[0081] x i =x i ,Others(13)

[0082] For the second half of the population of pelicans, that is, the second set of individuals, i is The sparrow alert mechanism is added to the loop from 1 to n. By introducing the sparrow alert mechanism, the convergence speed of the pelican algorithm can be accelerated. When the pelicans sense danger, the pelicans at the edge of the group will quickly move to a safe area to find a better position, while the pelicans in the center of the group will move randomly and get closer to other pelicans. The implementation logic of the sparrow alert mechanism is as follows: First, find the individual with the highest current fitness value. Then, the method of position update is determined by comparing the fitness value of each individual with the global optimal fitness value. If the fitness value of the individual is higher than the global optimal fitness value, the position of the individual is updated to X. best Add some random perturbations, that is, generate new positions by using normal distribution random numbers; otherwise, use the update strategy based on distance and fitness difference to update the position. The updated position will be limited to the upper and lower bounds of the variable. Then calculate the fitness value of the new position, and decide whether to accept the new position based on the comparison of the fitness values. Through this warning mechanism, the algorithm can enhance the global search capability while maintaining diversity.

[0083] After the positions of all pelicans in the pelican group are updated during the stage of skimming over the water surface, a second spatial state matrix is ​​constructed based on the updated spatial positions of the pelicans.

[0084] The second spatial state matrix is ​​used as the initial state matrix to iterate the steps of the flying to prey stage and the skimming water surface stage. When the maximum number of iterations is reached, the target spatial position is output according to the fitness value from the second spatial state matrix obtained when the iteration stops.

[0085] This embodiment uses the IPOA algorithm to optimize the parameters of MSCNN, thereby improving the performance of MSCNN.

[0086] In one embodiment, the method further includes: constructing a model combining a multi-scale convolutional neural network, a bidirectional long short-term memory network and an attention mechanism. This embodiment introduces an attention mechanism to enhance the model's attention to key information and improve recognition accuracy.

[0087] In one embodiment, an improved Pelican optimization algorithm-multi-scale convolutional neural network-bidirectional long short-term memory network-attention mechanism (IPOA-MSCNN-BiLSTM-Attention) model is proposed. The structural diagram of the model is shown in Figure 2 As shown in the figure. First, the model uses MSCNN to extract the temporal features of tool wear data, and effectively captures the local spatiotemporal patterns in the data through multi-scale convolution operations. Subsequently, BiLSTM further learns the global temporal features of the data, and captures the front-to-backward dependencies in the data by forward and backward processing sequences. Attention is introduced to dynamically adjust the weights of input features by calculating weights at different time steps, so that the model pays more attention to key wear information. In order to optimize the model parameters, the IPOA algorithm is proposed, which combines the adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism strategy to optimize the parameters of MSCNN, strike a balance between global search and local optimization, and ensure that the model can maintain efficient and stable performance under different conditions.

[0088] In order to verify the application effect of the IPOA-MSCNN-BiLSTM-Attention model in the actual tool wear state identification, the model was deployed on an industrial control computer. In terms of data processing, the model preprocessed, extracted features and analyzed the state classification of tool wear data. The LCD display is used to display real-time monitoring results and wear state information on site, so that maintenance personnel can intuitively understand the equipment status. The industrial control computer transmits the processed data remotely to China Mobile's Internet of Things Open Platform (OneNET) through a 4G communication module, and realizes cloud storage, management and analysis of data on the cloud platform. The cloud platform not only provides powerful computing and storage capabilities, but also supports remote monitoring and early warning functions. Through mobile applications or web interfaces, operation and maintenance personnel can view the equipment operation status and wear identification results anytime and anywhere, thereby improving the monitoring protection level and management efficiency of manufacturing equipment.

[0089] The present invention aims to test six algorithms, Improved Pelican (IPOA), Pelican (POA), Particle Swarm (PSO), Whale (WOA), Sand Cat Swarm (SCSO), Subtraction Optimizer (SABO) on the benchmark functions provided by the five classic test sets of CEC2005, CEC2017, CEC2019, CEC2021, and CEC2022. These benchmark functions include unimodal, multimodal, mixed and composite functions, covering different types of optimization problems. The local optimization performance of the unimodal function evaluation algorithm, the global optimization performance of the multimodal function evaluation algorithm, the global and local optimization performance of the hybrid function evaluation algorithm, and the comprehensive performance of the composite and multimodal function evaluation algorithm are evaluated. For each algorithm, the population size is set to 30, the number of iterations is set to 500, and a comprehensive evaluation is performed by executing each test function multiple times to examine its performance in terms of optimization accuracy and convergence speed. In addition to statistics on the optimal value (min), standard deviation (std), average (avg), median (median), and worst value (worse) of each algorithm, the results of the rank sum test are also specially counted. In the five classic test sets of CEC, the Pelican Optimization Algorithm (IPOA) that combines the Cauchy mutation and sparrow alert mechanism strategy has excellent performance in all aspects and achieved good comprehensive results. The box plot of the CEC2005 test is shown in the figure below. Figure 4 As shown in the figure, the theoretical minimum value (fmin) of the functions F6, F12, and F13 should be 0, which can test the quality of an algorithm. From the figure, it can be seen that the IPOA algorithm is the best in the optimal value test of these three functions. Specifically, the F6 function test curve is as follows Figure 3 The test results of F6, F12 and F13 functions are shown in Table 1.

[0090] Table 1 Test results of the F6, F12, and F13 benchmark functions of the six algorithms in CEC2005

[0091]

[0092]

[0093] The present invention uses experiments to verify the advantages of the IPOA-MSCNN-BiLSTM-Attention model. The experiment uses data from the tool milling competition (PHM2010). The equipment is a high-speed CNC milling machine, and the cutting material is stainless steel with a hardness of HRC52. The milling experiment parameters include spindle speed, feed speed, Y-axis cutting depth, Z-axis cutting depth and cooling conditions. The specific parameters are as follows: spindle speed 10400r / min, feed speed 1555mm / min, Y-axis cutting depth 0.125mm, Z-axis cutting depth 0.2mm, and cooling condition is dry cutting. Three milling cutters (marked as C1, C2 and C3 respectively) were used in the experiment for processing under the same working conditions, and each milling cutter passed 315 times. Each time the cubic workpiece is milled, a length of 108mm will be milled, and the time consumed by the pass is the same. Whenever the milling reaches 10 -3 After the tool is cut to mm, a microscope is used to measure the tool flank wear. The data acquisition card collects cutting force and vibration signals in the x, y, and z directions at a frequency of 50KHz. Since the vibration signal is non-stationary and varies greatly, the industrial control computer mainly uses the vibration signal and the recorded tool wear as the data set for the MSCNN-BiLSTM-Attention model training. The device platform architecture is as follows Figure 5 shown.

[0094] Since the signal data of some time periods are incomplete, it is necessary to remove the sections with abnormal data volume. In addition, the amount of original data is very large. If it is directly input into the neural network training, it may exceed the performance tolerance of the industrial control computer. Therefore, some data of the vibration sensor are taken for experiments. The data preprocessing is to normalize the original data and use a sample length of 10,000 to cut the samples without overlapping. In order to conduct experiments under balanced data sets, 1,000 samples are taken for each state. The ratio of training set to test set is 4:1, the batch size is 64, the optimizer is Adam, the learning rate is 0.001, and the number of training rounds is set to 100. This data set is used as the input of the MSCNN-BiLSTM-Attention model to test the effects of each model. According to the wear of the flank surface, the tool can be divided into three stages: initial wear, normal wear and rapid wear. They correspond to labels 0, label 1 and label 2 respectively. In order to further process and analyze the data, unique hot encoding is used to convert and represent these label data. The specific division is as follows: when the label is 0, it means that the tool is in the initial wear state, and the corresponding milling times are 1 to 54 times; when the label is 1, it means that the tool is in the normal wear state, and the corresponding milling times are 55 to 205 times; when the label is 2, it means that the tool is in the rapid wear state, and the corresponding milling times are 206 to 315 times. This division method helps to judge the degree of tool wear according to the number of milling times and perform corresponding processing. In the MSCNN-BiLSTM-Attention model, the multi-scale convolutional neural network (MSCNN) uses 8 parameters, namely filters1 to filters4 and kernel_size1 to kernel_size4. Initially, these parameters are assigned an initial value, and then optimized by the POA and IPOA algorithms to obtain better parameter settings, as shown in Table 2.

[0095] Table 2 Network structure parameters of IPOA-MSCNN-BiLSTM-Attention model

[0096]

[0097]

[0098] The IPOA algorithm combines adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism. The nonlinear increasing adaptive inertia weight factor ω is dynamically adjusted through formula (5), which enhances global exploration in the early stage and improves local development in the later stage, thus balancing the algorithm performance. In contrast, the linear decreasing weight factor has a faster global to local transition, but may lead to falling into the local optimum. The experiment was repeated 10 times under the same conditions. The results showed that the average accuracy (Aacc) of the IPOA-MSCNN-BiLSTM-Attention model with nonlinear increasing weight factor ω was 97.1%, the linear decreasing factor model Aacc was 96.1%, and the model without weight factor Aacc was 91.7%. The best accuracy (Bacc) of the three models can reach 98.2%. The model with nonlinear increasing weight factor performed best. The Cauchy mutation strategy uses the Cauchy distribution with a large tail to effectively improve the search diversity and avoid falling into the local optimum, especially when dealing with high-dimensional and multi-peak problems. By adjusting the scale parameter σ of the Cauchy distribution formula (10), the influence of the variation amplitude on the model effect is explored. Under the condition of nonlinear increasing inertia weight, the cases of σ being 0, 0.5, 1 and 1.5 were tested respectively, and 10 repeated tests were performed. The results show that when σ=0, the model Aacc is 95.2%; when σ=0.5, the model Aacc is 97.1%; when σ=1, the model Aacc is 96.2%; the three models Bacc can reach 98.2%. When σ=1.5, the model Aacc is 93.1% and the model Bacc can reach 97.5%. The results show that increasing the σ parameter to 0.5 can improve the model performance, but further increasing the parameter will lead to performance degradation, so the IPOA algorithm σ parameter is taken as 0.5.

[0099] In order to identify the tool wear status and verify the results, the parameters of feature extraction and algorithm optimization of each model are kept the same. The accuracy of the POA-MSCNN-BiLSTM-Attention and IPOA-MSCNN-BiLSTM-Attention models and the confusion matrix of each model test are shown in Figure 2. Figures 6 to 13 As shown in the figure, the classification number 0 refers to initial wear, 1 refers to normal wear, and 2 refers to rapid wear. The value of each grid indicates the number of correct recognitions and misrecognitions of each category by the model, with a total of 600 test samples. This comparative analysis helps to comprehensively evaluate the classification performance of different models.

[0100] In order to avoid the contingency of the experimental results, each model was repeated 10 times and the average value was taken for comparison. The Aacc of the MSCNN-BiLSTM model without the Attention mechanism is 91.6%, and the Bacc is 93.3%; the Aacc of the POA-MSCNN-BiLSTM model is 93.4%, and the Bacc is 97.2%; the Aacc of the IPOA-MSCNN-BiLSTM model is 95.2%, and the Bacc is 97.2%. After adding the Attention mechanism, the Aacc of the MSCNN-BiLSTM-Attention model is 92.4%, and the Bacc is 93.2%; the Aacc of the POA-MSCNN-BiLSTM-Attention model is 95.7%, and the Bacc is 98.2%; the Aacc of the IPOA-MSCNN-BiLSTM-Attention model is 97.1%, and the Bacc is 98.2%. The Attention mechanism significantly improves the model performance, especially in the IPOA-MSCNN-BiLSTM model, where Aacc is increased from 95.2% to 97.1%, verifying its key role and versatility in time series data classification. Figure 7 , 9 , 11 classification confusion matrices show that the MSCNN-BiLSTM-Attention model accurately predicts 165, 176, and 215 test samples for classification 0, 1, and 2, respectively; the POA-MSCNN-BiLSTM-Attention model accurately predicts 169, 191, and 215 test samples; and the IPOA-MSCNN-BiLSTM-Attention model accurately predicts 177, 191, and 215 test samples. With the introduction of the optimization algorithm from POA to IPOA, the classification accuracy gradually improves, especially for classification 0, where the IPOA-MSCNN-BiLSTM-Attention model performs best. Fig.12 and 13 It shows that the accuracy of the training set and test set of the IPOA-MSCNN-BiLSTM-Attention model is closer than that of the POA-MSCNN-BiLSTM-Attention model, and it has better stability, showing good generalization ability. The optimization algorithm improves the classification performance in complex feature space.

[0101] The performance evaluation of each model is shown in Table 3. The data in the table show that the indicators of the POA-MSCNN-BiLSTM-Attention model are higher than those of the MSCNN-BiLSTM-Attention model, indicating that the feature extraction after POA optimization is more comprehensive and the classification accuracy is significantly improved. The average values ​​of the evaluation indicators such as accuracy (Acc), precision (Pre), recall (Rec), and F1 value of the IPOA-MSCNN-BiLSTM-Attention model reached 97.1%, which is significantly better than the POA-MSCNN-BiLSTM-Attention and MSCNN-BiLSTM-Attention models, and the best accuracy can reach 98.2%. This shows that the IPOA algorithm has better parameter optimization results for the MSCNN-BiLSTM-Attention model, and the IPOA-MSCNN-BiLSTM-Attention model has stronger feature extraction capabilities and can more accurately correspond vibration signals to actual wear states. Therefore, this model performs better in tool wear state recognition.

[0102] Table 3 Comparative evaluation of model performance

[0103]

[0104]

[0105] By combining MSCNN and BiLSTM networks, the model of the present invention can effectively capture the spatiotemporal characteristics of machine tool tool wear status data. The introduction of the attention mechanism further enhances the model's attention to key information and improves the accuracy and stability of recognition. The IPOA algorithm is introduced to optimize the number of 4 convolution kernels and 4 convolution kernel size parameters of the multi-scale convolutional neural network of the model. Combined with the adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism strategy, it performs better than traditional POA and other algorithms in many function performance tests from CEC2005 to CEC2022, proving its advantages in optimization accuracy and convergence speed. The trained model is deployed on an industrial control computer to verify its effectiveness and reliability in real-time monitoring of machine tool tools and wear status identification, providing strong technical support for improving processing safety and production efficiency.

[0106] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0107] Based on the same inventive concept, the embodiment of the present application also provides a tool wear state identification method device based on the improved Pelican algorithm for implementing the tool wear state identification method based on the improved Pelican algorithm involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more tool wear state identification method device embodiments based on the improved Pelican algorithm provided below can refer to the limitations of the tool wear state identification method based on the improved Pelican algorithm above, and will not be repeated here.

[0108] In one embodiment, a tool wear state identification method and device based on an improved Pelican algorithm is provided, comprising:

[0109] A data acquisition module is used to obtain the wear data of the tool during milling operation;

[0110] A feature extraction module, used to extract the temporal features of wear data through a multi-scale convolutional neural network;

[0111] A prediction output module is used to input the time features into a bidirectional long short-term memory network to predict the wear degree of the tool;

[0112] Among them, the multi-scale convolutional neural network adopts the improved Pelican optimization algorithm for parameter optimization. The improved Pelican optimization algorithm introduces adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican optimization algorithm.

[0113] Each module in the tool wear state identification method based on the improved Pelican algorithm can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0114] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in all the above method embodiments when executing the computer program.

[0115] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in all the above method embodiments are implemented.

[0116] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in all the above method embodiments when executed by a processor.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0118] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processors involved in each embodiment provided in this application may be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, data processing logic devices based on quantum computing, etc., but are not limited thereto.

[0119] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A tool wear state identification method based on the improved Pelican algorithm, characterized in that: The method comprises: Obtain the wear data of the tool during milling work; Extracting the temporal features of the wear data through a multi-scale convolutional neural network; Inputting the time feature into a bidirectional long short-term memory network to predict the wear degree of the tool; Among them, the multi-scale convolutional neural network adopts an improved Pelican optimization algorithm for parameter optimization, and the improved Pelican optimization algorithm introduces an adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican optimization algorithm.

2. The method according to claim 1, characterized in that: The wear data includes a vibration signal generated by the tool during milling operation.

3. The method according to claim 1, characterized in that The improved Pelican optimization algorithm includes: Generate an initial spatial position matrix of the pelican population in the population initialization stage; wherein the spatial position matrix represents the spatial position of each pelican individual in the pelican population; In the stage of flying towards prey, based on the initial spatial position matrix, for each pelican individual in the pelican population, the fitness function value of the pelican individual is compared with the fitness value of the prey, and the spatial position of the pelican individual is updated based on the comparison result; wherein, an adaptive inertia weight factor is introduced in the updating process of the spatial position to adjust the degree of dependence on the spatial position before the update; and a first spatial state matrix of the pelican population is obtained after the spatial positions of all the pelican individuals are updated; In the stage of skimming over the water, the pelican population is divided into a first individual set and a second individual set; based on the first spatial state matrix, for the pelican individuals in the first individual set, when the fitness value of the pelican individuals is less than the average fitness value of the pelican population, the spatial position of the pelican individuals is updated by adopting the Cauchy mutation strategy; based on the first spatial state matrix, for the pelican individuals in the second individual set, a sparrow alert mechanism is introduced to update the spatial position of the pelican individuals; and a second spatial state matrix of the pelican population is obtained after the spatial positions of the pelican individuals in the first individual set and the second individual set are updated; Using the second spatial state matrix as the initial spatial position matrix of the next iteration process, iteratively executing the steps of the flying towards prey stage and the skimming over the water surface stage until a preset number of iterations is reached; The target spatial position is output according to the fitness value from the second spatial state matrix obtained when the iteration stops.

4. The method according to claim 3, characterized in that The adaptive inertia weight factor is related to the ratio between the current iteration number and the preset iteration number.

5. The method according to claim 3, characterized in that: In the water-skimming stage, when the individual fitness value of the pelican is not less than the average fitness value of the pelican population, the spatial position is updated based on the current spatial position and the current number of iterations.

6. The method according to claim 1, characterized in that The method further comprises: Construct a model that combines multi-scale convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms.

7. A tool wear state identification method and device based on the improved Pelican algorithm, characterized in that: The device comprises: A data acquisition module is used to obtain the wear data of the tool during milling operation; A feature extraction module, used for extracting the time features of the wear data through a multi-scale convolutional neural network; A prediction output module, used for inputting the time feature into a bidirectional long short-term memory network to predict the wear degree of the tool; Among them, the multi-scale convolutional neural network adopts an improved Pelican optimization algorithm for parameter optimization, and the improved Pelican optimization algorithm introduces an adaptive inertia weight factor, Cauchy mutation and sparrow warning mechanism into the Pelican optimization algorithm.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images