Linkage monitoring system and device for cutting surface roughness and tool wear

Abstract

The invention relates to the technical field of numerical control machine tool machining quality monitoring, and discloses a cutting machining surface roughness and tool wear linkage monitoring system and device, and the system comprises a multi-source heterogeneous data collection module which is used for collecting static data and dynamic data in the cutting machining process of a numerical control machine tool; and the data preprocessing module is used for preprocessing the collected multi-source heterogeneous data, including data cleaning, feature extraction and data compression. According to the cutting surface roughness and tool wear linkage monitoring system, a high-precision virtual measurement model is established through a multi-source heterogeneous data fusion and fuzzy width learning system, the accuracy of surface roughness and tool wear prediction is improved, the dependence on traditional detection equipment is reduced, and the detection cost is reduced; the problem that the number of samples is limited in the actual machining process is solved through the virtual sample generation technology, the generalization ability and prediction performance of the model are improved, and the machining process monitoring requirements under different working conditions are met.

Description

Technical Field

[0001] The present invention relates to the technical field of numerical control machine tool processing quality monitoring, and particularly to a linkage monitoring system and device for cutting surface roughness and tool wear. Background Art

[0002] In the field of modern machining, numerical control machine tool cutting is a key link to achieve high-precision and high-efficiency production. However, during the cutting process, the tool wear state directly affects the machining quality, production efficiency, and machining cost. Traditional monitoring methods mainly rely on manual inspection or the use of contact measurement equipment, and these methods have the following limitations: Low detection efficiency: Manual inspection is not only time-consuming and laborious but also difficult to achieve real-time monitoring. Although contact measurement equipment has high precision, it requires interrupting the machining process, resulting in a decrease in production efficiency.

[0003] High detection cost: High-precision detection equipment is expensive and has high maintenance costs, increasing the production costs of enterprises.

[0004] Strong data dependence: Traditional methods have a strong dependence on sample data, and the actual machining process has a limited number of samples, making it difficult to meet the requirements of model training, resulting in insufficient generalization ability of the model and inability to adapt to the machining process monitoring under different working conditions.

[0005] Lack of linkage monitoring: Surface roughness and tool wear state are usually monitored separately, lacking an effective linkage mechanism, making it difficult to achieve comprehensive monitoring and closed-loop control of the machining process.

[0006] With the rapid development of industrial automation and intelligence, higher requirements are put forward for real-time, efficient, and low-cost monitoring of the machining process. In recent years, multi-source heterogeneous data fusion technology and artificial intelligence have been widely applied in the field of industrial monitoring; these technologies can effectively integrate static and dynamic data from different sensors and achieve real-time monitoring and prediction of the machining process by constructing a high-precision virtual measurement model. However, how to improve the generalization ability and prediction performance of the model under the condition of limited sample quantity, and how to achieve the linkage monitoring of surface roughness and tool wear are still technical problems that need to be solved urgently. Summary of the Invention

[0007] (I) Technical Problems to be Solved Aiming at the deficiencies of the existing technology, the present invention provides a linkage monitoring system and device for the surface roughness and tool wear in cutting processing. Through multi-source heterogeneous data fusion and a fuzzy width learning system, a high-precision virtual measurement model is established to improve the accuracy of surface roughness and tool wear prediction, reduce the dependence on traditional detection equipment, and lower the detection cost. By adopting virtual sample generation technology, the problem of limited sample quantity in the actual processing process is solved, the generalization ability and prediction performance of the model are improved, and the monitoring requirements of the processing process under different working conditions are met. It can monitor the surface roughness of parts and the tool wear state in real time during the cutting processing of CNC machine tools, realize the synchronous monitoring of the two, timely detect abnormal situations in the processing process, and solve the problem of the lack of linkage monitoring of surface roughness and tool wear state.

[0008] (II) Technical Solution To achieve the above object, the present invention provides the following technical solution: A linkage monitoring system for the surface roughness and tool wear in cutting processing, including: A multi-source heterogeneous data acquisition module, which is used to acquire static data and dynamic data during the cutting processing of a CNC machine tool; A data preprocessing module, which is used to preprocess the acquired multi-source heterogeneous data, including data cleaning, feature extraction, and data compression; A virtual measurement modeling module, which constructs a surface roughness virtual measurement model and a tool wear state monitoring model based on a fuzzy width learning system; A small sample enhancement module, which is used to solve the problem of limited sample quantity by adopting virtual sample generation technology; A linkage monitoring module, which is used to integrate the surface roughness virtual measurement model and the tool wear state monitoring model to realize the multi-task synchronous monitoring of surface roughness and tool wear; The process of building the linkage monitoring model for surface roughness and tool wear is as follows: Let the training sample data set be , where , respectively represent the input data and output labels, represents the number of samples, is the number of attributes or features included in each sample, is the number of tasks; at the same time, let there be first-order TSK fuzzy subsystems in the linkage monitoring model, and the th TSK fuzzy subsystem contains fuzzy rules; for the sample input , , in the th fuzzy subsystem, the activation strength of the th fuzzy rule is calculated as follows: (7) Wherein, is the Gaussian membership function corresponding to the fuzzy set , and its calculation is as follows: (8) Wherein, and respectively represent the center and width of the Gaussian membership function; For the th fuzzy subsystem, the K-means clustering method is used to initialize the center of the Gaussian membership function; in order to reduce the model complexity, the variance of the membership function is set to 1; First, the activation intensity of the fuzzy rules is normalized as follows: (9) Before defuzzification in the th fuzzy subsystem, the intermediate output vector of the th training sample is expressed as: (10) Wherein, is the consequent of the rule of the th fuzzy subsystem, expressed as a linear combination of input features, that is (11) Wherein, is a random number obeying the uniform distribution ; The display and feedback module is used to display the monitoring results in real time and feedback the monitoring data to the numerical control machine tool control system to realize the closed-loop control of the machining process.

[0009] Preferably, the static data collected by the multi-source heterogeneous data collection module includes cutting parameters, tool diameter, and workpiece material; The dynamic data includes vibration signals, current signals, cutting force signals, acoustic emission signals, and built-in signals of the numerical control system.

[0010] Preferably, the data preprocessing module selects sensitive features through theoretical methods such as information theory and rough sets to reduce the data dimension.

[0011] Preferably, the small sample enhancement module uses interpolation, noise addition, sampling, and generative adversarial network methods to generate virtual samples; In the process of virtual sample generation, first, feature extraction and selection are performed on the process monitoring signals of the original training samples. Then, a virtual measurement model is constructed using the original small samples, and the optimal weight parameters are saved. Finally, a swarm intelligence optimization algorithm is used to generate virtual samples in combination with the trained virtual measurement model. The specific generation process is as follows: Suppose the feature matrix of the original training samples is , where is the number of training samples, is the number of attributes or features of the training samples. Then, the center of the th attribute is calculated as follows: (1) First, the information diffusion function based on triangular membership is used to asymmetrically expand the attribute domain of the selected features in the original training samples. The abscissa represents the observed values of the sample attributes, and the ordinate represents the possibility of the observed values occurring; The magnitudes of the left and right skewnesses can be regarded as a measure of the asymmetry of the sample distribution. The left and right skewnesses of the th attribute are calculated as follows: (2) (3) In the formula, and are the numbers of samples less than and greater than the center in the th attribute respectively. is used to fine-tune the skewness and is set to 1 here. Then, the upper and lower bounds of the extended domain of the th attribute are defined as: (4) (5) In the formula, and represent the minimum and maximum values of the th observed attribute respectively; Then, in order to generate more reasonable virtual samples, the swarm intelligence is used to optimize the minimum relative error between the virtual measurement value and the actual measurement value, and the optimal combination between the input features and the output labels is searched to serve as the virtual samples. Its mathematical description is as follows: (6) Finally, the original training samples and the generated virtual samples are concatenated together and the virtual measurement model is retrained, and the performance of the prediction results is evaluated.

[0012] Preferably, the linkage monitoring module realizes the multi-task synchronous monitoring of surface roughness and tool wear through information sharing between subtasks and the capture of their respective dynamic characteristics, and issues an alarm in a timely manner and feeds it back to the numerical control machine tool control system in case of an anomaly; The intermediate output matrix of the training samples in the feature layer is represented as follows: (13) Next, the intermediate output matrix of the feature layer is mapped to the reservoir corresponding to each subtask through a non-linear transformation to further capture the dynamic characteristics of each subtask. The state update equation of the reservoir is expressed as follows: (14) In the formula, is the state of the reservoir in the th task, represents the connection weight matrix between the intermediate output of the feature layer and the reservoir, represents the connection weight matrix inside the reservoir. The activation function selected here is the tanh function.

[0013] Preferably, the surface roughness detection device includes a detection component for detecting the roughness of the workpiece, a positioning component for positioning and adjusting the detection component, and an adjustment component for adjusting the initial position of the detection component.

[0014] Preferably, the adjustment component includes an adjustment frame, the adjustment frame is connected to the numerical control machine tool through a mounting seat, an adjustment block is slidably connected to the inside of the adjustment frame in a vertical sliding manner, the adjustment block is connected to the positioning component, a lead screw is rotatably connected to the inside of the adjustment frame, and the lead screw is threadedly connected to the adjustment block.

[0015] Preferably, the positioning component includes a U-shaped seat and an electric cylinder for driving the U-shaped seat up and down, and positioning plates are fixedly connected to both sides of the U-shaped seat; The detection component includes a detection box fixed inside the U-shaped seat and a terminal controller fixed on the mounting seat. The detection box is connected to the terminal controller by a data wire, and a probe is installed at one end of the detection box.

[0016] (III)Advantages Compared with the prior art, the present invention provides a linkage monitoring system and device for the surface roughness and tool wear in cutting machining, and has the following advantages: 1. The present invention establishes a high-precision virtual measurement model through multi-source heterogeneous data fusion and a fuzzy width learning system, improves the accuracy of surface roughness and tool wear prediction, reduces the dependence on traditional detection equipment, and lowers the detection cost. By adopting virtual sample generation technology, the problem of limited sample quantity in the actual machining process is solved, the generalization ability and prediction performance of the model are improved, the monitoring requirements of the machining process under different working conditions are met, the surface roughness and tool wear status during the cutting process of CNC machine tools can be monitored in real time, the synchronous monitoring of the two can be realized, abnormal conditions in the machining process can be detected in time, and machining quality problems caused by tool wear can be avoided.

[0017] 2. Through the setting of the adjustment component, the present invention is used to adjust the height of the detection component, so as to meet the detection work of workpieces with different sizes and thicknesses. Through the setting of the positioning component, the detection end of the detection component is positioned and calibrated, ensuring that the detection end of the detection component contacts the surface to be measured with reasonable force to form a roughness detection work, solving the problem that during the real-time cutting process of the workpiece, the thickness or diameter of the workpiece gradually decreases, resulting in the detection end of the detection component being unable to accurately move to the surface to be measured after cutting, thus affecting the subsequent real-time monitoring of roughness.

[0018] 3. The telescopic rod of the present invention is installed inside the sleeve in an inserted manner, which not only facilitates the telescopic adjustment of the telescopic rod to form the longitudinal adjustment of the position of the detection component and realize the roughness detection work at different positions, but also facilitates the angle adjustment of the telescopic rod to form the angle adjustment of the detection end of the detection component and realize the roughness detection work on different inclined surfaces, further improving the functionality and practicality of the roughness detection device. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic diagram of the principle of the present invention; Figure 2 is a schematic diagram of the multi-source heterogeneous data synchronous acquisition technical route of the present invention; Figure 3 is a schematic diagram of the virtual measurement modeling process based on multi-source heterogeneous data of the present invention; Figure 4 is a schematic diagram of the virtual sample generation technical route of the present invention; Figure 5 is a schematic diagram of the multi-task linkage monitoring technical route of surface roughness and tool wear of the present invention; Figure 6 is a schematic diagram of the structure of the surface roughness detection device of the present invention; Figure 7 is a side view of the structure of the surface roughness detection device of the present invention; Figure 8 is a schematic diagram of the structure of the adjustment component of the present invention; Figure 9 It is a structural schematic diagram of the positioning component of the present invention; Figure 10 It is a transmission schematic diagram of two sliding blocks of the present invention; Figure 11 It is a schematic diagram of the monitoring results of the linkage between the surface roughness and the tool wear of the present invention; Figure 12 For the present invention Figure 4 Schematic diagram of the expansion of the attribute domain in it.

[0020] In the figure: 1. Detection component; 11. Detection box; 12. Terminal controller; 13. Probe; 2. Positioning component; 21. U-shaped seat; 22. Electric cylinder; 23. Positioning plate; 24. Guide frame; 25. Sliding block; 26. Positioning shaft; 27. Buckle part; 28. Pressure sensor; 29. Hinge seat; 210. Transmission frame; 211. Axis positioning light source; 212. Sleeve; 213. Telescopic rod; 214. Locking bolt; 3. Adjustment component; 31. Adjustment frame; 32. Mounting seat; 33. Adjustment block; 34. Lead screw. Specific embodiments

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Embodiment 1: Referring to the attached Figures 1 to 5 , Figure 11 and Figure 12 , a linkage monitoring system for the surface roughness and tool wear in cutting machining includes: A multi-source heterogeneous data acquisition module for acquiring static data and dynamic data during the cutting machining process of a numerically controlled machine tool; A data preprocessing module for preprocessing the acquired multi-source heterogeneous data, including data cleaning, feature extraction, and data compression, and selecting sensitive features through theoretical methods such as information theory and rough sets to provide high-quality data input for subsequent modeling; A virtual measurement modeling module that constructs a surface roughness virtual measurement model and a tool wear state monitoring model based on a fuzzy width learning system; since the two-task linkage monitoring model is data-driven, its input is static process parameters and dynamic monitoring signal features, and the output is the surface roughness measurement value and the tool wear state. The modeling process is as Figure 5 shown; Figure 5 It is a schematic diagram of multi-task synchronous monitoring based on a fuzzy width learning system. First, multi-source heterogeneous data is mapped to each fuzzy subsystem in the feature layer. Then, the outputs of the feature layer are merged and mapped to the enhancement layer of each sub-task. Finally, the outputs of the feature layer and the enhancement layer are merged and connected to the output layer, and the connection weights of the output layer are solved by pseudo-inverse; The small sample enhancement module uses virtual sample generation technology to solve the problem of limited sample quantity, improve the generalization ability and prediction performance of the model. Aiming at the problem of limited sample quantity in the actual machining process, virtual sample generation technology such as interpolation, noise addition, sampling and generative adversarial network method is used to generate virtual samples with a certain scale and reasonable structure, enrich sample diversity, and improve the prediction performance and generalization ability of the model; The linkage monitoring module is used to integrate the surface roughness virtual measurement model and the tool wear state monitoring model to realize multi-task synchronous monitoring of surface roughness and tool wear; through information sharing between sub-tasks and capture of their respective dynamic characteristics, multi-task synchronous monitoring of surface roughness and tool wear is realized. When abnormal conditions of surface roughness or tool wear state are detected, an alarm is sent in time and feedback is sent to the numerical control machine tool control system so that corresponding adjustment measures can be taken. Here, it should be noted that "abnormal" here means that the surface roughness value exceeds the specified requirements of the part processing drawing for its surface roughness, and the tool wear amount exceeds the national standard or special requirements for precision machining; The display and feedback module is used to display the monitoring results in real time and feedback the monitoring data to the numerical control machine tool control system to realize closed-loop control of the machining process; the display and feedback module provides a user interface to facilitate the operator to view the monitoring data and adjust the monitoring parameters; Real-time performance and synchronization: It can monitor the surface roughness and tool wear state in the cutting process of the numerical control machine tool in real time, realize synchronous monitoring of the two, and timely detect abnormal conditions in the machining process to avoid machining quality problems caused by tool wear.

[0023] The present invention adopts a method combining theoretical research and experimental verification, develops a multi-source heterogeneous data acquisition system, explores the correlation between multi-granularity perception information in the machining process and the surface roughness of parts, proposes a virtual measurement method for cutting surface roughness based on multi-source heterogeneous data fusion, uses virtual sample generation technology to realize small sample enhancement of surface roughness, and deeply conducts multi-task synchronous monitoring of surface roughness and tool wear, specifically as follows: Multi-granularity synchronous acquisition and preprocessing technology of multi-source heterogeneous data The data of numerically controlled machine tools can be divided into program information such as spindle speed, feed rate, cutting parameters, position, etc., internal information such as motor current, spindle torque, servo error, etc., and external sensor information such as spindle vibration, current, machine tool temperature, etc. The sources, formats, and sampling frequencies of these information are different. The present invention takes G-code instructions as the basis for data sampling, establishes data sampling rules, conducts multi-granularity synchronous sampling of multi-source heterogeneous data, and then realizes the unification of the data time scale, provides a benchmark for subsequent analysis, and reduces the storage and transmission pressure of data through data compression technology. Its technical route is as Figure 2 shown; Figure 2 Taking G-code instructions as the basis for data sampling, multi-granularity synchronous acquisition of multi-source heterogeneous data of built-in sensors, additional sensors, and process parameters in the numerical control system during the machining process, and reducing the storage and transmission pressure of data through technologies such as feature extraction and selection.

[0024] It should be noted here that when conducting data acquisition, based on the G-code instructions of the numerical control system, a trigger sampling rule for data can be established to achieve synchronous sampling and time matching of multi-source heterogeneous data during the numerical control machining process; Virtual measurement modeling of surface roughness based on multi-source heterogeneous data fusion Regarding the problems such as the complex forming mechanism of the surface roughness of cutting surfaces, diverse influencing factors resulting in low model prediction accuracy and poor generalization ability, relevant additional sensors are deployed at safe and sensitive parts of the numerically controlled machine tool for process monitoring, and sensitive features are extracted from the monitoring signals in the time domain, frequency domain, and spatial domain respectively. Then, the static cutting parameters and dynamic monitoring information are fused, and theoretical methods such as stepwise selection and rough set are used for feature selection. Furthermore, a virtual measurement model of surface roughness based on multi-source heterogeneous data fusion is established. Its modeling process is as Figure 3 shown; Figure 3 It is a schematic diagram of the virtual measurement modeling process of surface roughness based on multi-source heterogeneous data fusion. First, the static process parameters and dynamic monitoring signal features during the machining process of the numerically controlled machine tool are fused, and then a width learning system based on feature selection is constructed for the prediction of surface roughness.

[0025] Surface roughness small sample enhancement technology based on virtual sample generation Regarding the small sample problems such as the small number of surface roughness samples, uneven distribution, and incomplete information existing in the actual cutting machining of numerically controlled machine tools, the present invention regards the generation of virtual samples as a constrained optimization problem, designs surface roughness virtual sample generation schemes starting from cutting parameters and process monitoring signals respectively, and enriches sample diversity by filling the information gaps between original small samples. The technical route of surface roughness virtual sample generation is as Figure 4 shown; Figure 4 Generate a technical roadmap for virtual samples. First, construct a virtual measurement model for the original small samples, extract and select features from the monitoring signals during the processing, and expand the attribute domain of the selected features; then use the swarm intelligence optimization algorithm combined with the small sample model to generate effective virtual samples; finally, merge the original small samples with the generated virtual samples and reconstruct the virtual measurement model; Specifically as follows: During the generation process of virtual samples, it can be regarded as an optimization problem with constraints. First, extract and select features from the process monitoring signals of the original training samples, then construct a virtual measurement model using the original small samples and save the optimal weight parameters, and finally use the swarm intelligence optimization algorithm combined with the trained virtual measurement model to generate virtual samples. The specific generation process is as follows: Assume that the feature matrix of the original training samples is , where is the number of training samples, is the number of attributes or features of the training samples, then the center of the th attribute can be calculated as follows: (1) First, use the information diffusion function based on triangular membership to asymmetrically expand the attribute domain of the selected features in the original training samples, Figure 12 As shown, the abscissa represents the observed values of the sample attributes, and the ordinate represents the possibility of the observed values occurring; From Figure 12 it can be intuitively seen that the skewness of the triangle in the TMIE function is closely related to the number of samples on both sides of the center of each attribute. This shows that the magnitudes of the left and right skewnesses can be regarded as a measure of the asymmetry of the sample distribution, where the left and right skewnesses of the th attribute are calculated as follows: (2) (3) In the formula, and are respectively the number of samples less than and greater than the center in the th attribute, is used to fine-tune the skewness, and is set to 1 here. Then, the upper and lower bounds of the extended domain of the th attribute can be defined as: (4) (5) In the formula, and respectively represent the The minimum and maximum values of an observed attribute.

[0026] Then, in order to generate more reasonable virtual samples, the minimum relative error between the virtual measurement value and the actual measurement value is optimized using swarm intelligence, and the optimal combination between the input features and the output labels is searched for to serve as the virtual sample. Its mathematical description is as follows (6) Finally, the original training samples and the generated virtual samples are concatenated together and the virtual measurement model is retrained, and the performance of the prediction results is evaluated.

[0027] Multi-task virtual measurement method for surface roughness and tool wear According to the complex correlation between the surface roughness of the part and tool wear, using multi-source heterogeneous monitoring information in the cutting process, feature extraction is carried out respectively from the time domain, frequency domain, and time-frequency domain, and then kernel principal component analysis is used for dimensionality reduction. Finally, through information sharing between sub-tasks and capturing of their respective dynamic characteristics, a multi-task virtual measurement model for surface roughness and tool wear is established, thereby improving data utilization rate and reducing computational cost. The multi-task virtual measurement technical route for surface roughness and tool wear is as Figure 5 shown.

[0028] The method combining theoretical research and experimental verification is adopted. The specific experimental process and results are as follows: An end milling experiment was carried out on a three-axis machining center. When the milling cutter showed severe wear, a total of 816 tool passes were made, and 63 measurements were taken irregularly during this period. That is, a total of 63 groups of effective sample data were collected in this experiment. In order to verify the effectiveness of the proposed method for the joint monitoring of surface roughness and tool wear, 48 groups of sample data were randomly selected for model training, and the remaining 15 groups of sample data were used for model testing. First, the vibration, current, and cutting force signals during the milling process were segmented, and the signals of the last tool pass before each measurement were intercepted, and the unstable parts when the tool entered and exited were removed. Then, 8 time-domain features and 5 frequency-domain features were extracted from each channel of the intercepted effective signals, that is, a total of 117 signal features were extracted for each sample. Since there are some irrelevant or redundant features in the extracted signal features, kernel principal component analysis was used to perform non-linear dimensionality reduction on them, and the first 12 principal components were selected as the input of the joint monitoring model, and their cumulative contribution rate reached 95.11%. In addition, the milling cutter state was divided into two categories: normal and worn according to the wear condition of the milling cutter.

[0029] According to the above process, the results of the joint monitoring of surface roughness and tool wear status by the proposed model for the test samples are as Figure 11 shown, where Figure 11(a) shows the model prediction results and actual measurement results of surface roughness, Figure 11 (b) shows the actual measurement results and model prediction results of the tool wear state. It can be seen that the prediction results of the proposed linkage monitoring model are consistent with the actual measurement results, with the MAPE for surface roughness prediction being 5.58% and the monitoring accuracy for tool wear state being 100%.

[0030] The process of jointly monitoring surface roughness and tool wear by modeling is as follows: Let the training sample data set be , where and represent the input data and output labels respectively, represents the number of samples, is the number of attributes or features included in each sample, is the number of tasks. At the same time, let there be first-order TSK fuzzy subsystems in the linkage monitoring model, and the rd TSK fuzzy subsystem contains fuzzy rules. For the sample input , , in the th fuzzy subsystem, the activation strength of the rd fuzzy rule is calculated as follows: (7) In the formula, is the Gaussian membership function corresponding to the fuzzy set , and its calculation is as follows: (8) In the formula, and represent the center and width of the Gaussian membership function respectively.

[0031] For the th fuzzy subsystem, the K-means clustering method is used to initialize the center of the Gaussian membership function. To reduce the model complexity, the variance of the membership function is set to 1.

[0032] First, the activation strengths of the fuzzy rules are normalized as follows: (9) Before defuzzification in the th fuzzy subsystem, the intermediate output vector of the th training sample can be expressed as: (10) Wherein, is the consequent of the -th fuzzy subsystem, which can be expressed as a linear combination of input features, i.e., (11) Wherein, is a random number subject to a uniform distribution .

[0033] Furthermore, the intermediate output matrix of all training samples in the -th fuzzy subsystem can be written as: (12) Thus, the intermediate output matrix of all training samples at the feature layer can be expressed as follows: (13) Next, the intermediate output matrix at the feature layer is mapped to the reservoir corresponding to each subtask through a non-linear transformation to further capture the dynamic characteristics of each subtask. The state update equation of the reservoir is expressed as follows: (14) Wherein, is the state of the reservoir in the -th task, represents the connection weight matrix between the intermediate output of the feature layer and the reservoir, represents the connection weight matrix inside the reservoir. The activation function selected here is the tanh function.

[0034] For the -th training sample , the defuzzified output of the -th fuzzy subsystem is calculated as follows: (15) Wherein, is used to adjust the consequent of the -th fuzzy rule, which can be expressed as: (16) At this time, (17).

[0035] For all training samples, the defuzzified output of the -th fuzzy subsystem can be expressed as: (18) Therefore, the defuzzified outputs of all fuzzy subsystems can be integrated as follows: (19) where (20).

[0036] Finally, the defuzzified outputs of all fuzzy subsystems are concatenated with the outputs of the corresponding reservoir for each subtask and mapped to the final output layer. At this time, the final output of FESBTTLS can be expressed as: (21) where represents the connection weight matrix from the reservoir to the output layer in the -th task.

[0037] Denote and Then can be regarded as the output of the hidden layer in the -th task. Then can be abbreviated as: (22) Given the actual training sample labels , the output weight matrix can be solved by the pseudoinverse as follows: (23) (24) where and represent the regularization parameter and the identity matrix, respectively.

[0038] During the training process of the multi-task linkage monitoring model, its loss function is expressed as the weighted sum of the loss functions of multiple subtasks. Generally, the loss function of each subtask is made as small as possible, that is: (25) where is the weighting coefficient, is the loss function of the -th subtask.

[0039] Referring to Appendix Figures 6 to 10 , the surface roughness detection device includes a detection component 1 for detecting the roughness of the workpiece, a positioning component 2 for positioning and adjusting the detection component 1, and an adjustment component 3 for adjusting the initial position of the detection component 1; By adjusting the settings of the adjusting component 3, the height of the detection component 1 can be adjusted to meet the detection requirements for workpieces of different sizes and thicknesses. Through the setting of the positioning component 2, the detection end of the detection component 1 can be positioned and calibrated to ensure that the detection end of the detection component 1 contacts the surface to be measured with a reasonable force, forming a roughness detection operation. This solves the problem that during real-time cutting of a workpiece, the thickness or diameter of the workpiece gradually decreases, resulting in the detection end of the detection component 1 being unable to accurately move to the surface to be measured after cutting, thus affecting subsequent real-time roughness monitoring.

[0040] Embodiment 2: Different from Embodiment 1; Refer to the attached Figures 6 to 10 The adjusting component 3 includes an adjusting frame 31. The adjusting frame 31 is connected to the numerical control machine tool through a mounting seat 32. An adjusting block 33 is slidably connected inside the adjusting frame 31 in a vertical sliding manner. The adjusting block 33 is connected to the positioning component 2. A lead screw 34 is rotatably connected inside the adjusting frame 31, and the lead screw 34 is threadedly connected to the adjusting block 33; A driving handle is installed at the top of the lead screw 34, which facilitates the rotation drive of the lead screw 34 by the operator, enabling the adjusting block 33 to move up and down to adjust the height of the detection end of the detection component 1.

[0041] The positioning component 2 includes a U-shaped seat 21 and an electric cylinder 22 for driving the U-shaped seat 21 up and down. Positioning plates 23 are fixedly connected to both sides of the U-shaped seat 21; The electric cylinder 22 is connected to an external power supply and a controller, and is set using the connection method and coding method of the existing technology. It is used to drive the U-shaped seat 21 up and down, and then drive the detection end of the detection component 1 up and down to contact the workpiece for subsequent roughness detection. Through the setting of the two positioning plates 23, they are used to press against the surface to be measured, which not only prevents the problem of damage to the detection end caused by the continuous descent of the detection end of the detection component 1, but also enables the detection end of the detection component 1 to accurately contact the surface to be measured, improving the accuracy of subsequent detection; It should be noted here that the bottom of the positioning plate 23 is flush with the bottom of the detection end of the detection component 1; The detection component 1 includes a detection box 11 fixed inside the U-shaped seat 21 and a terminal controller 12 fixed on the mounting seat 32. The detection box 11 is connected to the terminal controller 12 using a data wire, and a probe 13 is installed at one end of the detection box 11; The detection component 1 uses a contact-type roughness detection instrument in the existing technology, which is composed of a detection box 11, a terminal controller 12, and a probe 13. It is controlled by the terminal controller 12 and detected by the probe 13.

[0042] Embodiment 3: Different from Embodiment 2; Refer to the attached Figure 9 and Figure 10 , the positioning component 2 includes a guiding frame 24. Two symmetric sliding blocks 25 are slidably connected inside the guiding frame 24. The U-shaped seat 21 is fixedly connected to the upper sliding block 25. A positioning shaft 26 is rotatably connected to the lower sliding block 25 through a rotating shaft, and a buckle member 27 for locking the rotating shaft is installed on the sliding block 25; Through the setting of the guiding frame 24, it is used to improve the smoothness and guiding property of the symmetric sliding of the two sliding blocks 25. The detection end of the detection component 1 is installed on the upper sliding block 25, and a positioning shaft 26 is installed on the other sliding block 25, which is convenient for when the cylindrical workpiece is being machined by cutting, the diameter of the workpiece may gradually decrease, resulting in the problem that the detection end of the subsequent detection component 1 cannot be accurately positioned to contact the outer surface of the workpiece; Through the symmetric setting of the detection end of the detection component 1 and the positioning shaft 26, it is convenient that when the detection end and the positioning shaft 26 move relative to each other, when the positioning shaft 26 contacts the outer surface of the cylindrical workpiece, the detection end of the detection component 1 just reaches the detection position, forming an automatic calibration and positioning operation, solving the problem that when the cylindrical workpiece is cut and its diameter gradually decreases, the detection end of the detection component 1 cannot be accurately positioned to contact the outer surface of the workpiece, and then the probe 13 is likely to have too large or too small contact force with the cylindrical workpiece, thus affecting the accuracy of its roughness detection and easily causing damage to the probe 13; By installing the positioning shaft 26 in a rotating manner, it is convenient that when the workpiece is a non-cylindrical workpiece, the positioning shaft 26 can be hidden to prevent the problem that the existence of the positioning shaft 26 and its synchronous movement with the detection end cause interference to the subsequent movement route; The electric cylinder 22 is fixed to one side of the guiding frame 24 through a bracket. The telescopic end of the electric cylinder 22 is fixedly connected to a hinge seat 29 through a pressure sensor 28, and two inclined transmission frames 210 are hinged to the hinge seat 29. The two transmission frames 210 are respectively hinged to the two sliding blocks 25, and an axial center positioning light source 211 is installed on one side of the hinge seat 29; The axial center positioning light source 211 includes, but is not limited to, an infrared generator for positioning the axial center of the cylindrical workpiece, which is convenient for positioning the initial position of the positioning component 2 and improving the positioning and calibration effect of the subsequent detection component 1; With the setting of the electric cylinder 22, it is used to drive the articulated seat 29 to perform telescopic movement, and then can drive the two inclined transmission frames 210 to perform sector movement, and then can drive the two sliding blocks 25 to move relatively or away from each other. Through the relative movement of the two sliding blocks 25, the probe 13 and the positioning shaft 26 in the detection component 1 can be driven to move relatively or away from each other. By arranging the probe 13 and the positioning shaft 26 symmetrically, when the positioning shaft 26 contacts the outer surface of the cylindrical workpiece, the probe 13 just contacts the outer surface of the cylindrical workpiece, forming a positioning and calibration operation, effectively preventing the problem that the contact force between the probe 13 and the cylindrical workpiece is too large or too small, which affects the accuracy of roughness detection and is likely to cause damage to the probe 13; With the setting of the pressure sensor 28, it is used to detect the pressure during positioning. When the positioning shaft 26 contacts the outer surface of the cylindrical workpiece, as the electric cylinder 22 continues to contract, the pressure detected by the pressure sensor 28 will gradually increase. When the set threshold is reached, it means that the detection end of the detection component 1 is positioned.

[0043] Embodiment 4: What is different from Embodiment 3 is that; Refer to the attached Figure 8 and Figure 9 One side of the adjusting block 33 is fixedly connected with a sleeve 212, and a telescopic rod 213 for inserting into the inside of the sleeve 212 is fixedly connected to the guiding frame 24. A locking bolt 214 for locking the telescopic rod 213 for telescopic adjustment or rotational adjustment is arranged on the sleeve 212; By installing the telescopic rod 213 in the sleeve 212 in an inserted manner, it is not only convenient to perform telescopic adjustment on the telescopic rod 213, thereby forming longitudinal adjustment of the position of the detection component 1 to realize roughness detection work at different positions, but also convenient to perform angle adjustment on the telescopic rod 213 to form adjustment of the detection end angle of the detection component 1 to realize roughness detection work on different inclined surfaces, further improving the functionality and practicality of the roughness detection device.

[0044] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The cutting surface roughness and tool wear linkage monitoring system is characterized by: include: Multi-source heterogeneous data acquisition module, used to collect static and dynamic data during the CNC machine tool cutting process; The data preprocessing module is used to preprocess the collected multi-source heterogeneous data, including data cleaning, feature extraction and data compression; Virtual measurement modeling module, based on fuzzy width learning system, builds surface roughness virtual measurement model and tool wear status monitoring model; Small sample enhancement module, used to solve the problem of limited sample number by using virtual sample generation technology; Linkage monitoring module, which is used to integrate the surface roughness virtual measurement model with the tool wear status monitoring model to achieve multi-task synchronous monitoring of surface roughness and tool wear; The modeling process of surface roughness and tool wear linkage monitoring is as follows: Assume that the training sample data set is ,in , Represent the input data and output label respectively, represents the number of samples, is the number of attributes or features contained in each sample, is the number of tasks; at the same time, suppose the linkage monitoring model contains A first-order TSK fuzzy subsystem, The TSK fuzzy subsystem contains Fuzzy rules; for sample input , , in In the fuzzy subsystem The activation strength of a fuzzy rule is calculated as follows: （7） In the formula, Corresponding to the fuzzy set The Gaussian membership function is calculated as follows: （8） In the formula, and Respectively represent the center and width of the Gaussian membership function; For Fuzzy subsystems, using K-means clustering method to initialize the center of Gaussian membership function ; In order to reduce the complexity of the model, the variance of the membership function is Set to 1; First, the activation strength of the fuzzy rules is normalized as follows: （9） In the Before defuzzifying the fuzzy subsystem, The intermediate output vector of training samples is expressed as: （10） In the formula, For the The rule consequent of the fuzzy subsystem is expressed as a linear combination of the input features, that is, （11） In the formula, To obey uniform distribution A random number; The display and feedback module is used to display the monitoring results in real time and feed back the monitoring data to the CNC machine tool control system to achieve closed-loop control of the machining process.

2. The cutting surface roughness and tool wear linkage monitoring system according to claim 1 is characterized in that: The static data collected by the multi-source heterogeneous data collection module includes cutting parameters, tool diameter, and workpiece material; Dynamic data include vibration signals, current signals, cutting force signals, acoustic emission signals and CNC system built-in signals.

3. The cutting surface roughness and tool wear linkage monitoring system according to claim 1 is characterized in that: The data preprocessing module selects sensitive features through theoretical methods such as information theory and rough sets to reduce data dimensions.

4. The cutting surface roughness and tool wear linkage monitoring system according to claim 1 is characterized in that: The small sample enhancement module generates virtual samples by using interpolation, noise addition, sampling and generative adversarial network methods; In the process of generating virtual samples, the features of the process monitoring signals of the original training samples are first extracted and selected, and then the virtual measurement model is constructed using the original small samples, and the optimal weight parameters are saved. Finally, the swarm intelligence optimization algorithm is used to combine the trained virtual measurement model to generate virtual samples. The specific generation process is as follows: Assume that the feature matrix of the original training sample is ,in is the number of training samples, is the number of attributes or features of the training sample, then The centroid of an attribute is calculated as follows: （1） Firstly, the attribute domain of the selected features in the original training samples is asymmetrically expanded using the information diffusion function based on triangular membership. The horizontal axis represents the observed value of the sample attribute, and the vertical axis represents the possibility of the observed value. The size of the left and right skewness can be regarded as a measure of the asymmetry of the sample distribution. The left and right skewness of each attribute is calculated as follows: （2） （3） In the formula, and Respectively The number of samples that are smaller and larger than the center in the attribute, Used to fine-tune the skewness, here it is set to 1; thus, The upper and lower bounds of the attribute extension domain are defined as: （4） （5） In the formula, and Respectively represent The minimum and maximum values ​​of the observed attributes; Then, in order to generate more reasonable virtual samples, swarm intelligence is used to optimize the minimum relative error between the virtual measurement value and the actual measurement value, so as to search for the optimal combination of input features and output labels to serve as virtual samples. The mathematical description is as follows: （6） Finally, the original training samples and the generated virtual samples are spliced ​​together to retrain the virtual measurement model, and the prediction results are evaluated for performance.

5. The cutting surface roughness and tool wear linkage monitoring system according to claim 1 is characterized in that: The linkage monitoring module realizes multi-task synchronous monitoring of surface roughness and tool wear by sharing information between subtasks and capturing their respective dynamic characteristics, and issues an alarm in time under abnormal circumstances and feeds back to the CNC machine tool control system; The intermediate output matrix of the training sample at the feature layer is expressed as follows: （13） Next, the intermediate output matrix of the feature layer After nonlinear transformation, it is mapped to the reserve pool corresponding to each subtask to capture the dynamic characteristics of each subtask. The state update equation of the reserve pool is expressed as follows: （14） In the formula, For the The state of the reserve pool in each task, Represents the connection weight matrix between the intermediate output of the feature layer and the reserve pool, Represents the connection weight matrix inside the reservoir; the activation function selected here is the tanh function.

6. The surface roughness detection device used in the cutting surface roughness and tool wear linkage monitoring system according to any one of claims 1 to 5, characterized in that: The surface roughness detection device comprises a detection component (1) for detecting the roughness of a workpiece, a positioning component (2) for adjusting the positioning of the detection component (1), and an adjustment component (3) for adjusting the initial position of the detection component (1).

7. The surface roughness detection device according to claim 6, characterized in that: The adjustment assembly (3) comprises an adjustment frame (31), the adjustment frame (31) being connected to the numerical control machine tool via a mounting seat (32), an adjustment block (33) being slidably connected to the interior of the adjustment frame (31) in an up-and-down sliding manner, the adjustment block (33) being connected to the positioning assembly (2), a lead screw (34) being rotatably connected to the interior of the adjustment frame (31), and the lead screw (34) being threadedly connected to the adjustment block (33).

8. The surface roughness detection device according to claim 7, characterized in that: The positioning assembly (2) comprises a U-shaped seat (21) and an electric cylinder (22) for driving the U-shaped seat (21) up and down, and positioning plates (23) are fixedly connected to both sides of the U-shaped seat (21); The detection assembly (1) comprises a detection box (11) fixed inside a U-shaped seat (21) and a terminal controller (12) fixed on a mounting seat (32); the detection box (11) is connected to the terminal controller (12) using a data wire, and a probe (13) is installed at one end of the detection box (11).

Images