A milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering

Abstract

The present invention discloses a milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering, which belongs to the field of intelligent manufacturing and processing technology. The method comprises the following steps: taking the original time domain signal of the spindle current during the milling process of the milling cutter as the model input, referring to the synchronous observation results and quantitative standards of a super-depth-of-field three-dimensional microscope, extracting the characteristic parameter vectors of the current signals of the milling cutter at different wear levels, considering the fuzziness and uncertainty of experimental data in the identification of critical states of each level of milling cutter wear, introducing a dynamic fuzzy clustering algorithm, obtaining the initial fuzzy clustering division by outputting a threshold λ, further considering the problem that the dynamic fuzzy clustering algorithm is prone to fall into local optimal values, adopting an immune algorithm with global search and parallel capabilities for optimization, obtaining the optimal threshold λ, and finally establishing an immune-optimized dynamic fuzzy clustering model, which provides new ideas for the actual engineering milling cutter wear state level assessment and equipment safety quantitative evaluation.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing and processing technology, and relates to the problems of tool surface wear and workpiece processing quality efficiency during milling. Specifically, it is a milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering. Background Art

[0002] Milling, as a major processing method in modern manufacturing, is widely used in the processing of core components of large-scale equipment such as nuclear power and aerospace. As time goes by, the scratching of the tool by the hard points of the workpiece material and the adhesion and diffusion between the two contact surfaces during the milling process can easily cause the degradation of the milling cutter. If an excessively worn milling cutter is continuously used for production processing, it will not only cause significant changes in milling parameters such as force, temperature, and vibration, but also determine the processing quality and efficiency of the workpiece, and even endanger the personal safety of workers. Therefore, selecting a reasonable and effective method for monitoring the wear status of the milling cutter is of great significance to the development of modern manufacturing.

[0003] The monitoring and identification of milling cutter wear conditions can be divided into two main methods: direct and indirect measurement. Among them, indirect measurement methods are widely used in experiments and engineering due to their real-time detection characteristics. That is, by obtaining indirect measurement methods such as current, force, and acoustic emission signals, the relationship between each milling cutter wear state and the signal is determined, completing the monitoring and identification of the milling cutter wear state. In the research on tool wear conditions, data-driven methods based on machine learning account for a large proportion, mainly including artificial neural networks, support vector machines, and decision trees. A method for classifying end mill wear conditions based on a support vector machine of force and current signals is proposed, and the accuracy of the proposed method in data classification is verified (e.g., in the literature MSchwenzer, K Miura, et al. Machine Learning for Tool Wear Classification in Milling Based on Force and Current Sensors[A]. 2019 4th International Conference on Design Engineering and Science (ICDES2019).). A multi-sensor information fusion system was proposed to overcome the difficulty of rapidity and accuracy in online life prediction of milling tools (e.g. Wu J, Su Y, et al. Multi-sensor Information Fusion for Remaining Useful Life Prediction of Machining Tools by Adaptive Network based Fuzzy Inference System [J]. Applied Soft Computing, 2018, 68: 13-23.). A BP neural network multi-signal fusion model based on cutting force and vibration signals was established to effectively realize the classification of tool wear status (e.g. Chen Gang, Jiao Li, et al. Research on Milling Cutter Wear Status Monitoring Based on Multi-sensor Data Fusion [J]. New Technology and New Process, 2017 (11): 23-28.). The above scholars used shallow machine learning to monitor milling cutter wear status and predict life. Although this method has good recognition and prediction effects, it requires a lot of prior knowledge and repeated testing to ensure the accuracy of the final model, and the generalization ability of the model is dependent on the training samples.In the field of deep learning research, deep learning theories for industrial big data processing and analysis have been proposed, effectively solving the local optimality problem in shallow neural network models (e.g., Hinton GE, Osindero S, et al. AFast Learning Algorithm for Deep Belief Nets[J]. Neural Computation, 2006, 18(7):1527-1554.). A deep learning-based online classification method for milling cutter wear has been proposed, using force signals at different wear stages as input samples, which improves the efficiency of offline monitoring of milling cutter wear (e.g., Terrazas, German, et al. Online Tool Wear Classification during Dry Machining Using Real Time Cutting Force Measurements and a CNN Approach[J]. Journal of Manufacturing and Materials Processing, 2018, 2(4):72.). The identification of high-speed milling cutter wear status is completed based on deep learning methods, which improves the prediction accuracy of shallow models (such as the literature Lin Yang, Gao Siyu, et al. High-speed milling cutter wear status prediction method based on deep learning [J]. Machinery and Electronics, 2017, 35(7): 12-17.). Deep learning is used to perform feature dimensionality reduction, and a milling cutter wear prediction model based on least squares support vector machine and cuckoo optimization algorithm is established (such as the literature Dai Wen, Zhang Chaoyong, et al. Milling cutter wear prediction model based on support vector machine based on deep learning and feature post-processing [J]. Computer Integrated Manufacturing Systems 2020, 26(09): 2331-2343). Although the above-mentioned deep learning methods can obtain more accurate milling cutter wear status monitoring models, such methods require professional experience and knowledge background, and with the improvement of industrial precision requirements, the universality of such methods will be reduced due to their increased complexity. Furthermore, to ensure that the tool condition monitoring model developed in this study has good recognition performance, it is necessary to preprocess the raw data from the sampling process. Furthermore, the model primarily analyzes signals under constant operating conditions. However, in actual engineering, milling cutter processing conditions vary depending on process requirements, and the data preprocessing process can also result in the loss of certain signal features. Therefore, a milling cutter wear condition monitoring method that uses raw signals as input and is applicable to a variety of operating conditions is urgently needed. Summary of the Invention

[0004] In order to effectively solve the problems existing in the prior art, the present invention provides a milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering. The original time domain signal of the spindle current during the milling process of the milling cutter is used as the model input, and the synchronous observation results and quantitative standards of the super-depth of field three-dimensional microscope are referred to. The characteristic parameter vectors of the current signals of the milling cutter with different wear levels are extracted. Considering the fuzziness and uncertainty of the experimental data in the identification of the critical states of each level of milling cutter wear, a dynamic fuzzy clustering algorithm is introduced, and the initial fuzzy clustering division is obtained by outputting the threshold λ. Further considering the problem that the dynamic fuzzy clustering algorithm is prone to fall into the local optimal value, the immune algorithm with global search and parallel capabilities is used for optimization to obtain the optimal threshold λ. Finally, an immune-optimized dynamic fuzzy clustering model is established, which provides new ideas for the actual engineering milling cutter wear state level assessment and equipment safety quantitative evaluation.

[0005] The technical solution of the present invention is:

[0006] A milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering includes the following steps:

[0007] Step 1: Development of experimental plan;

[0008] Through simulation experiments, the wear amount corresponding to different tool wear states was determined to guide processing. The full life signal of a new tool from the start of cutting to severe wear was collected. The wear amount of the tool flank was measured during the experiment, and a corresponding relationship between the current monitoring signal and the wear amount of the tool in different wear states was established. The tool used was a carbide flat-end end mill with two flutes, a diameter of 12mm, a helix angle of 30°, and a total length of 72mm. The workpiece material used was 40Cr. During the experiment, the machine was stopped in real time, and the wear amount of the tool flank was measured using an ultra-depth-of-field microscope.

[0009] To explore the ability to monitor and identify different tool wear states during milling under variable working conditions, a crossover experiment method was used to set various milling parameters and repeat the machining process to obtain the spindle current signal. The resulting milling scheme is shown below:

[0010] Table 1 Milling scheme

[0011]

[0012] Step 2: Analysis of experimental data;

[0013] The spindle current signals obtained in step 1 above are sorted and classified, and the three-phase current signals of the spindle motor under various working conditions are integrated and solved by using the current effective value solution formula. u , I v , I w Fusion into three-phase current effective value Irms ;

[0014] According to the flank wear, the tool wear degree is divided into four states: wear state I, wear state II, wear state III and wear state IV;

[0015] The effective value Irms data of the spindle current signal is preprocessed and divided into samples according to the wear state. Then, the signal samples are trained and tested in sequence. Among them, the total length of the current signal contains 36 groups of data. The samples are sampled to obtain signals with different wear states. 1400 samples are obtained, with 350 samples for each wear state. 250 samples are selected for each of the four wear states as training samples, and the rest are used as test samples. The distribution of the samples is shown in Table 2:

[0016] Table 2 Samples of four tool wear states

[0017]

[0018] In view of the fuzziness and uncertainty in the identification of critical wear states of various tool grades under variable working conditions, the dynamic fuzzy clustering algorithm does not require the prior determination of the number of clusters. Fuzzy clustering analysis is used to describe the uncertainty of sample classification. At the same time, the characteristics of each characteristic indicator and the influence of classification decision are considered, so that the final classification result has a better fuzzy similarity relationship.

[0019] Step 3: Analysis of dynamic fuzzy clustering algorithm;

[0020] Dynamic fuzzy clustering can quickly and preliminarily estimate the sample's category. By iteratively calculating the fuzzy similarity matrix, the category and distribution parameters are continuously updated until convergence. Experiments have revealed that the fuzziness and uncertainty between current signals within each category make it impossible to establish strict classification boundaries, and there is a lack of effective quantitative methods to resolve the fuzzy relationships between signals. To address this issue, a fuzzy similarity matrix and transitive closure are established, and dynamic fuzzy clustering is performed on the initial dataset using different thresholds λ.

[0021] Considering the shortcomings of dynamic fuzzy clustering algorithm that it is easy to fall into local optimal value and the number of classification levels is uncertain, combined with the characteristics of global convergence, diversity and parallelism of immune algorithm, after establishing the initial partition clustering data set through dynamic fuzzy clustering algorithm, the high-dimensional samples are mapped to a two-dimensional plane, so that the Euclidean distance between samples approaches fuzzy similarity. According to the affinity of antibodies, the optimal threshold λ is determined to obtain the global optimal solution and the optimal number of classification levels, and the optimal clustering division is output.

[0022] The details are as follows:

[0023] By establishing the fuzzy similarity matrix and transitive closure, the initial data set is dynamically fuzzy clustered using different thresholds λ; let the domain U={x1,x2,···,x n} is the sample space, the total number of samples is n, and each sample corresponds to m features, that is, x i ={x1,x2,···,x im}, get the original data matrix (x ij ) n×m , where x ij represents the jth characteristic index of the i-th sample, i = 1, 2, ..., n; j = 1, 2, ..., m; the original data is standardized and the translation-range transformation formula (2) is used to compress the data into the interval [0, 1] standardization matrix:

[0024]

[0025] Use the absolute value subtraction method-Euclidean distance to calculate the sample x i with x j The similarity r ij for:

[0026]

[0027] Among them, d(x i ,x j ) represents the sample x i with x j The Euclidean distance, c is the weight parameter, so that 0≤r ij ≤1, c is 0.1;

[0028] In the similarity r ij Based on this, we establish the fuzzy similarity matrix R(x i ,x j ):

[0029]

[0030] Transform R into a fuzzy equivalent matrix t(R), that is, the transitive closure t(R) of R; use the square method of formula (5) to calculate in sequence to find the equivalent matrix t(R) containing the fuzzy similarity matrix R:

[0031] R→R·R→(R 2 ) 2 →…(R 2 ) k →… (5)

[0032] In the established fuzzy equivalence relation, fuzzy clustering is performed on the transitive closure t(R) of R. In order to objectively reflect the clustering state of the sample data set, a threshold λ∈[0,1] is introduced and the λ-section matrix R of R is obtained.λ =(λ rij ) n×n , because different thresholds λ correspond to different classification levels, the purpose of dynamic classification is achieved, as shown in formula (6):

[0033]

[0034] According to formula (6), the clustering data set is initially divided and the classification results of dynamic fuzzy clustering are obtained; however, due to different threshold λ values, the number of classification levels is also different, so it is necessary to further perform optimal clustering analysis to optimize the threshold λ and obtain the best classification results and the number of classification levels;

[0035] Step 4: Model training and validation;

[0036] An immune-optimized dynamic fuzzy clustering algorithm is used to evaluate the affinity of individuals, and individuals for immune operations are selected based on affinity and antibody concentration to increase the diversity of the population, ensure that the result obtained when the algorithm terminates is the global optimal solution, and determine the optimal number of classification levels to improve classification accuracy.

[0037] To accurately monitor tool wear during milling, we extracted both dimensioned and dimensionless characteristic parameters from the current time-domain signal. Using the multidimensional characteristic parameters of the current signal, we investigated their correlation with the wear state. The current signals of the tool under different wear states during milling were collated, and several groups of characteristic parameters were randomly selected from the total sample to serve as training samples. The wear on the milling cutter flank was compared under various operating conditions, and the sample parameters were categorized.

[0038] To further validate the model's ability to identify critical wear states for milling cutters, several additional sets of milling data were used as test samples to verify the model. By comparing the actual and predicted levels of the sample data, the dynamic immune fuzzy clustering model's effectiveness in categorizing milling cutter wear states was verified, providing a theoretical basis for accurately distinguishing damage states in practical engineering applications.

[0039] Furthermore, the process of the immune optimized dynamic fuzzy clustering algorithm is as follows:

[0040] (1) Assign sample coordinate values ​​according to the characteristic parameters of the current signal, execute the dynamic fuzzy clustering algorithm, establish the fuzzy similarity matrix through equations (2) to (4), calculate the λ-intercept matrix of the transitive closure matrix t(R) according to equations (5) to (6), obtain the threshold λ, and output the initial cluster division;

[0041] (2) Use binary coding to obtain the initial population;

[0042] (3) Calculation of antibody affinity:

[0043] (4) Determine whether the termination condition is met. If so, use antibodies with high affinity to classify, determine the optimal threshold, and output the optimal clustering. Otherwise, continue the optimization calculation.

[0044] (5) Calculate the antibody concentration and excitation degree, where the antibody concentration is The excitation degree is sim(x i ,x j )=a·aff(x i ,x j )-(1-a)·den(x i ,x j ), a is the calculation parameter;

[0045] (6) Select antibodies with high affinity and low antibody concentration for immunization operation;

[0046] (7) The population is refreshed, and the newly generated antibodies replace the antibodies with lower excitation in the population, forming a new generation of antibodies, and then go to step (3).

[0047] Beneficial effects of the present invention: The method of the present invention uses the original time domain signal of the spindle current under different working conditions as the model input, and completes the identification and monitoring of different wear status levels of the milling cutter by training and testing the multi-dimensional feature parameter vector of the sample data. Experimental results show that the monitoring model can effectively capture the characteristics of the original data, thereby correctly classifying the tool wear conditions. Overall, the proportion of cases that can be accurately classified in the test sample data is very high, and under normal circumstances, the immune dynamic fuzzy clustering model predicts the damage level with an accuracy rate of 95%, which provides a new idea for the actual engineering milling cutter wear level assessment and equipment safety quantitative evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the dynamic immune fuzzy clustering algorithm provided by the present invention;

[0049] Figure 2 It is a structural diagram of the experimental system of the present invention;

[0050] Figure 3 is a characteristic distribution diagram of training samples of the present invention;

[0051] Figure 4 It is a two-dimensional schematic diagram of the wear state classification of the present invention. DETAILED DESCRIPTION

[0052] The specific implementation of the present invention is described in detail below in conjunction with the technical solutions and drawings.

[0053] In this embodiment, a milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering is proposed, wherein the specific algorithm operation process is as follows: Figure 1 As shown, it mainly includes the following steps:

[0054] Step 1: Development of experimental plan;

[0055] A 2-flute carbide flat-end milling cutter with a diameter of 12 mm and a helix angle of 30° was selected as the experimental tool. 40Cr was selected as the workpiece material. The signal acquisition equipment included a closed-loop Hall effect current sensor, a 12-channel data acquisition module, and DASP software. The sampling frequency was set to 4096 Hz. The experimental environment was a vertical CNC machining center. The specific operation process is as follows: Figure 2 The milling parameters were set using a cross-experimental method, and the machining process was repeated cyclically to obtain the spindle current signal. The milling parameters are shown in Table 1.

[0056] Table 1 Milling scheme

[0057]

[0058] To accelerate tool wear, no cutting fluid was used during the entire milling process. That is, the tool was worn to different states by dry milling. The wear of the tool flank was then measured and marked for each working condition to define the different wear states of the milling cutter. The specific classification is shown in Table 2.

[0059] Table 2 Samples of four tool wear states

[0060]

[0061]

[0062] Step 2: Acquisition and analysis of experimental data;

[0063] The above experimental scheme is used to obtain the three-phase current signal of the spindle motor under multiple working conditions. At the same time, according to formula (1), the three-phase current signal I u , I v , I w Fusion into three-phase current effective value I rms .

[0064]

[0065] The current data obtained from the experiments was organized and classified, and training and test samples were created for the corresponding characteristic parameters. To address the fuzziness and uncertainty in identifying critical wear states at various levels of tool wear under variable operating conditions, a dynamic fuzzy clustering algorithm, which does not require a predetermined number of clusters, was used to describe the uncertainty of sample assignment to different categories. This approach also considered the characteristics of each characteristic indicator and the impact of the classification decision, resulting in a final classification result with a better fuzzy similarity relationship.

[0066] Step 3: Analysis of dynamic fuzzy clustering algorithm;

[0067] By building a fuzzy similarity matrix and transitive closure, dynamic fuzzy clustering is performed on the initial data set using different thresholds λ. Let the domain U = {x1, x2, ···, x n} is the sample space, the total number of samples is n, and each sample corresponds to m features, that is, x i ={x1,x2,···,x im}, we can get the original data matrix (x ij ) n×m , where x ij (i = 1, 2, ..., n; j = 1, 2, ..., m) represents the jth characteristic index of the i-th sample. Different data correspond to different dimensions. To make it possible to compare different dimensions, it is necessary to perform appropriate transformations on the data, that is, to standardize the original data. The translation-range transformation formula (2) is used to compress the data into a standardized matrix in the interval [0, 1].

[0068]

[0069] Use the absolute value subtraction method-Euclidean distance to calculate the sample x i with x j The similarity r ij for:

[0070]

[0071] Where: d(x i ,x j ) represents the sample x i with x j The Euclidean distance, c is the weight parameter, so that 0≤r ij ≤1, in this paper c is taken as 0.1.

[0072] In the similarity r ij Based on this, we establish the fuzzy similarity matrix R(x i ,x j ):

[0073]

[0074] Since R generally only satisfies reflexivity and symmetry, in order to perform fuzzy clustering analysis, R must be transformed into a fuzzy equivalent matrix t(R), that is, the transitive closure t(R) of R. Using the square method of formula (5), we can calculate the equivalent matrix t(R) containing the fuzzy similarity matrix R.

[0075] R→R·R→(R 2 ) 2 →…(R2 ) k →… (5)

[0076] In the established fuzzy equivalence relation, fuzzy clustering is performed on the transitive closure t(R) of R. In order to objectively reflect the clustering state of the sample data set, a threshold λ∈[0,1] is introduced and the λ-section matrix R of R is obtained. λ =(λ rij ) n×n , because different thresholds λ correspond to different classification levels, the purpose of dynamic classification can be achieved, as shown in formula (6).

[0077]

[0078] According to formula (6), the initial clustering data set is divided and the classification results showing dynamic fuzzy clustering can be obtained. However, since the number of classification levels varies with the value of the threshold λ, further optimal clustering analysis is required to optimize the threshold λ and obtain the optimal classification results and the number of classification levels.

[0079] according to Figure 1 As shown, the dynamic fuzzy clustering algorithm process of immune optimization is completed, and its specific implementation process is as follows:

[0080] (1) Assign sample coordinate values ​​according to the characteristic parameters of the current signal, execute the dynamic fuzzy clustering algorithm, establish the fuzzy similarity matrix through formulas (2-4), calculate the λ-intercept matrix of the transitive closure matrix t(R) according to formulas (5-6), obtain the threshold λ, and output the initial cluster division;

[0081] (2) Use binary coding to obtain the initial population;

[0082] (3) Calculation of antibody affinity:

[0083] (4) Determine whether the termination condition is met. If so, use the antibody classification with high affinity to determine the optimal threshold λ and output the optimal clustering division. Otherwise, continue the optimization calculation.

[0084] (5) Calculate the antibody concentration and excitation degree, where the antibody concentration is: The degree of motivation is: sim(x i ,x j )=a·aff(x i ,x j )-(1-a)·den(x i ,x j ), a is the calculation parameter;

[0085] (6) Select antibodies with high affinity and low antibody concentration for immunization operations;

[0086] (7) The population is refreshed, and the newly generated antibodies replace the antibodies with lower excitation in the population, forming a new generation of antibodies, and then go to step (3).

[0087] Step 4: Model training and validation;

[0088] In order to accurately monitor the wear status of the tool during the milling process, the dimensional and dimensionless characteristic parameters of the current time domain signal are extracted respectively. By using the multidimensional characteristic parameters of the current signal, the correlation between the parameter and the wear status is explored.

[0089] The current signals of the tool under different wear states in the milling process are sorted out, and the characteristic parameters of the current signals of each state in the total sample are randomly extracted as training samples. Taking the data in the training sample as an example, 16 groups of data are selected for training, and the size of the milling cutter flank wear under various working conditions is compared. The categories of the sample parameters are divided into the following specific categories: Figure 3 As shown in the figure, Ⅰ to Ⅳ represent the tool wear status respectively. According to the above current signal characteristic parameters, the coordinate values ​​of the training samples are assigned, and the corresponding fuzzy similarity matrix R is established through equations (2) to (4).

[0090]

[0091] The immune algorithm is used to optimize each sample data set, and the immune parameters are set as 200 initial population, 100 maximum iterations and 10 clones. The above test samples show that λ = 0.9441 is the best threshold. At this time, the sample X is divided into four categories: {x1, x2, x3, x4}, {x5, x6, x7, x8}, {x9, x 10 ,x 11 ,x 12}, {x 13 ,x 14 ,x 15 ,x 16} respectively correspond to the four wear states of the milling cutter, and the training data results are in good agreement with the actual results.

[0092] In order to further verify the recognition effect of the model on different critical wear states of milling cutters, another 20 sets of milling data of the cutters were taken as test samples, as shown in Table 2.

[0093] Table 3 Test samples

[0094]

[0095]

[0096] The results show that when the number of iterations reaches 22, the affinity reaches the maximum value, that is, the optimal threshold λ = 0.9479 is obtained, and the optimal clustering result is output. The status monitoring of the sample is as follows: Figure 4 As shown, they are: {x1,x2,x3,x4,x5}, {x6,x7,x8,x9,x 10}, {x 11 ,x 12 ,x 13 ,x 14 ,x 15 ,x 16}, {x 17 ,x 18 ,x 19 ,x 20 Comparing the actual grade of the sample data with the predicted grade, the classification accuracy of the test samples reached 95%, verifying the effectiveness of the dynamic immune fuzzy clustering model in classifying the wear status of milling cutters and providing a theoretical basis for accurately distinguishing damage states in actual engineering.

Claims

1. A milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering, characterized in that: The following steps are involved: Step 1: Development of experimental plan; Through simulation experiments, the wear amount corresponding to different tool wear states was determined to guide processing. The full life signal of a new tool from the start of cutting to severe wear was collected. The wear amount of the tool flank was measured during the experiment, and a corresponding relationship between the current monitoring signal and the wear amount of the tool in different wear states was established. The tool used was a carbide flat-end end mill with two flutes, a diameter of 12mm, a helix angle of 30°, and a total length of 72mm. The workpiece material used was 40Cr. During the experiment, the machine was stopped in real time, and the wear amount of the tool flank was measured using an ultra-depth-of-field microscope. In order to explore the monitoring and identification capabilities of different tool wear states during milling under variable working conditions, a crossover experiment method was used to set various milling parameters and repeat the machining process to obtain the spindle current signal. Step 2: Analysis of experimental data; The spindle current signals obtained in step 1 above are sorted and classified, and the three-phase current signals of the spindle motor under various working conditions are integrated and solved by using the current effective value solution formula. u , I v , I w Fusion into three-phase current effective value I rms ; According to the flank wear, the tool wear degree is divided into four states: wear state I, wear state II, wear state III and wear state IV; The effective value of the spindle current signal I rms The data is preprocessed and divided into samples according to the wear state, and then the training and test signal samples are sequentially used. Among them, the total length of the current signal contains 36 groups of data, which are sampled to obtain signals with different wear states. 1400 samples are obtained, with 350 samples for each wear state. 250 samples are selected for each of the four wear states as training samples, and the rest are used as test samples. The distribution of the samples is shown in Table 2: Table 2 Samples of four tool wear states In view of the fuzziness and uncertainty in the identification of critical wear states of various tool grades under variable working conditions, the dynamic fuzzy clustering algorithm does not require the prior determination of the number of clusters. Fuzzy clustering analysis is used to describe the uncertainty of sample classification. At the same time, the characteristics of each characteristic indicator and the influence of classification decision are considered, so that the final classification result has a better fuzzy similarity relationship. Step 3: Analysis of dynamic fuzzy clustering algorithm; By establishing the fuzzy similarity matrix and transitive closure, the initial data set is dynamically fuzzy clustered using different thresholds λ; let the domain U={x1,x2,···,x n } is the sample space, the total number of samples is n, and each sample corresponds to m features, that is, x i ={x1,x2,···,x im }, get the original data matrix (x ij ) n×m , where x ij represents the jth characteristic index of the i-th sample, i = 1, 2, ..., n; j = 1, 2, ..., m; the original data is standardized and the translation-range transformation formula (2) is used to compress the data into the interval [0, 1] standardization matrix: Use the absolute value subtraction method-Euclidean distance to calculate the sample x i with x j The similarity r ij for: Among them, d(x i ,x j ) represents the sample x i with x j The Euclidean distance, c is the weight parameter, so that 0≤r ij ≤1, c is 0.1; In the similarity r ij Based on this, we establish the fuzzy similarity matrix R(x i ,x j ): Transform R into a fuzzy equivalent matrix t(R), that is, the transitive closure t(R) of R; use the square method of formula (5) to calculate in sequence to find the equivalent matrix t(R) containing the fuzzy similarity matrix R: R→R·R→(R 2 ) 2 →…(R 2 ) k →… (5) In the established fuzzy equivalence relation, fuzzy clustering is performed on the transitive closure t(R) of R. In order to objectively reflect the clustering state of the sample data set, a threshold λ∈[0,1] is introduced and the λ-section matrix R of R is obtained. λ =(λ rij ) n×n , because different thresholds λ correspond to different classification levels, the purpose of dynamic classification is achieved, as shown in formula (6): According to formula (6), the clustering data set is initially divided and the classification results of dynamic fuzzy clustering are obtained; however, due to different threshold λ values, the number of classification levels is also different, so it is necessary to further perform optimal clustering analysis to optimize the threshold λ and obtain the best classification results and the number of classification levels; Step 4: Model training and validation; An immune-optimized dynamic fuzzy clustering algorithm is used to evaluate the affinity of individuals, and individuals for immune operations are selected based on affinity and antibody concentration to increase the diversity of the population, ensure that the result obtained when the algorithm terminates is the global optimal solution, and determine the optimal number of classification levels to improve classification accuracy.

2. The milling cutter wear monitoring method based on order spectrum and dynamic immune fuzzy clustering according to claim 1 is characterized in that: The process of the immune optimized dynamic fuzzy clustering algorithm is as follows: (1) Assign sample coordinate values ​​according to the characteristic parameters of the current signal, execute the dynamic fuzzy clustering algorithm, establish the fuzzy similarity matrix through equations (2) to (4), calculate the λ-intercept matrix of the transitive closure matrix t(R) according to equations (5) to (6), obtain the threshold λ, and output the initial cluster division; (2) Use binary coding to obtain the initial population; (3) Calculation of antibody affinity: (4) Determine whether the termination condition is met. If so, use antibodies with high affinity to classify, determine the optimal threshold, and output the optimal clustering. Otherwise, continue the optimization calculation. (5) Calculate the antibody concentration and excitation degree, where the antibody concentration is The excitation degree is sim(x i ,x j )=a·aff(x i ,x j )-(1-a)·den(x i ,x j ), a is the calculation parameter; (6) Select antibodies with high affinity and low antibody concentration for immunization operation; (7) The population is refreshed, and the newly generated antibodies replace the antibodies with lower excitation in the population, forming a new generation of antibodies, and then go to step (3).

Images