Precision deduction method for collaborative machining of multi-tools in machine tools based on virtual-real difference convergence

Abstract

The present application discloses a method for deducing the precision of collaborative machining of multiple tools of a machine tool based on the convergence of virtual and real differences, and relates to the technical field of deducing the precision of machining of machine tools. The method effectively avoids the interference of massive non-machining redundant data through the machine tool machining state identification model, and provides high-quality data support for the tool wear state identification process. At the same time, a hierarchical tool wear state integrated learning model is used to accurately identify the tool wear state, and an accurate and explainable tool wear prediction model is established on this basis, thereby realizing the accurate prediction of tool wear. In addition, by constructing a knowledge graph of collaborative machining of multiple tools, existing tools are used to guide the wear prediction of new tools, and the virtual and real difference convergence mechanism is used to drive the online update of the new tool wear prediction model, which effectively overcomes the problem of needing to carry out tool wear characteristic tests for each new tool and the long test cycle, thereby realizing the accurate, fast and efficient deduction of machine tool machining precision.

Description

Technical Field

[0001] The present application relates to the technical field of machine tool processing accuracy deduction, and in particular to a method for deducing machine tool multi-tool collaborative processing accuracy based on virtual-real difference convergence. Background Art

[0002] With the rapid development of the manufacturing industry, higher requirements are placed on the machining accuracy of CNC machine tools. Among them, as an important component of CNC machine tools, the wear of cutting tools during industrial machining has a great impact on the dimensional accuracy of the target workpiece, and may even cause serious accidents such as machine tool damage. However, since precision parts are usually manufactured by the collaborative processing of multiple types of cutting tools such as milling cutters and grinding wheels, the differences in the structure and size of each cutting tool are destined to have significantly different wear characteristics between different cutting tools. Therefore, how to scientifically and effectively monitor the wear status of machine tool cutting tools and predict the wear amount is of great significance to reducing unnecessary downtime and ensuring machining accuracy.

[0003] At present, in order to avoid the influence of tool wear on machine tool processing accuracy, the direct method and indirect method are mainly used at home and abroad. Among them, the direct method mainly uses tools such as microscopes to measure the tool wear during shutdown. This type of method is usually suitable for experimental environments, but it is difficult to apply in actual production due to its shortcomings such as poor timeliness and complicated operations. Compared with the direct method, the indirect method analyzes the monitoring data that can reflect the tool state and size changes, constructs a mapping relationship between it and the wear amount, and realizes the indirect deduction of tool wear. This type of method has achieved great success, but there are still certain shortcomings in actual engineering applications: (1) The existing methods lack the distinction between processing state and non-processing state data, and a large amount of redundant data in the non-processing state exists, which seriously limits the effect of tool wear prediction methods; (2) The existing methods generally have the problem of insufficient interpretability, and do not predict the wear amount according to the tool wear state, which is difficult to adapt to the tool prediction task with significant changes in wear characteristics; (3) The existing methods do not have the ability to migrate predictions under multi-tool collaborative processing. Usually, a separate wear amount prediction model is established for a certain type of tool, which cannot avoid the problems of long cycle and high cost in tool wear tests. Summary of the invention

[0004] The purpose of this application is to provide a method for deducing the precision of multi-tool collaborative machining of machine tools based on the convergence of virtual and real differences, which can realize accurate, fast and efficient deduction of the machining precision of machine tools under multi-tool collaborative machining.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] The present application provides a method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence, comprising the following steps:

[0007] The multi-source sensor data set is obtained during the machine tool processing process, and the multi-source sensor data under the processing state is screened through the machine tool processing state identification model, and the multi-source sensor characteristics under the processing state are determined.

[0008] A hierarchical tool wear state integrated learning model is constructed to learn the mapping relationship between the multi-source sensor characteristics under any existing tool processing state of the machine tool and the tool wear state of the existing tool; the hierarchical tool wear state integrated learning model can determine the tool wear state of the existing tool based on the multi-source sensor characteristics under the input processing state.

[0009] For any existing tool of the machine tool, a tool wear prediction model under different wear conditions is constructed and trained.

[0010] For newly added tools, the nearest tool to the newly added tool is determined among the existing tools of the machine tool based on the knowledge graph of collaborative machining of multiple tools.

[0011] The tool wear prediction model of the nearest tool under different wear conditions is used as the wear prediction guidance model of the new tool under different wear conditions; the wear prediction guidance model can determine the predicted wear of the new tool according to the multi-source sensor characteristics under the input processing state.

[0012] For any wear prediction guidance model of the newly added tool, the wear prediction guidance model is updated online according to the root mean square error between the actual wear of the newly added tool and the predicted wear. The virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear prediction guidance model, so as to realize the accurate deduction of machine tool processing accuracy under multi-tool collaborative processing.

[0013] Optionally, a multi-source sensor data set is obtained during the machining process of a machine tool, and the multi-source sensor data under the machining state is screened through a machine tool machining state identification model, and multi-source sensor features under the machining state are determined, specifically including the following steps:

[0014] A multi-source sensor data set is obtained during a machine tool processing process; the multi-source sensor data set includes a plurality of multi-source sensor data of unit time length.

[0015] Based on the synchronous compression wavelet transform algorithm, the multi-source sensor data of each section in the machine tool processing process are converted into two-dimensional image data.

[0016] The two-dimensional image data corresponding to each segment of multi-source sensor data is input into the machine tool processing state identification model to determine the machine tool working state corresponding to each segment of multi-source sensor data.

[0017] According to the machine tool working state corresponding to each segment of multi-source sensor data, the multi-source sensor data under the processing state is screened from the multi-source sensor data set in the machine tool processing process, and the multi-source sensor features under the processing state are determined; the multi-source sensor feature refers to a feature whose Pearson coefficient between the tool wear state and the tool wear state is greater than a preset threshold, which is determined from several features represented in the time domain, frequency domain and time-frequency domain of the multi-source sensor data.

[0018] Optionally, based on the synchronous compression wavelet transform algorithm, each section of multi-source sensor data in the machine tool processing process is converted into two-dimensional image data respectively, which specifically includes the following steps:

[0019] For any section of multi-source sensor data in the machine tool processing process, the multi-source sensor data is decomposed by wavelet transform to obtain components at different frequencies and time scales.

[0020] The sparse representation method is used to compress the components of multi-source sensor data at different frequencies and time scales to obtain the time-frequency feature matrix.

[0021] The time-frequency feature matrix is ​​converted into a two-dimensional image form to obtain the two-dimensional image data corresponding to the multi-source sensor data.

[0022] Optionally, the machine tool processing state identification model includes a convolution module, a bidirectional gated loop module and a strong classification module; before inputting the two-dimensional image data corresponding to each segment of multi-source sensor data into the machine tool processing state identification model and determining the machine tool working state corresponding to each segment of multi-source sensor data, the machine tool multi-tool collaborative processing accuracy deduction method based on virtual-real difference convergence also includes the following steps:

[0023] A processing state identification data set is constructed; the processing state identification data set includes a number of processing state identification samples; the processing state identification samples include two-dimensional image data corresponding to multi-source sensor data under the processing state and the corresponding machine tool working state.

[0024] Initialize the weight of each processing state identification sample in the processing state identification data set to obtain a processing state identification data set after weight initialization; the weights of each processing state identification sample in the processing state identification data set after weight initialization are the same.

[0025] Based on the processing state identification sample data set after weight initialization, a weak classifier is trained and used as the current weak classifier.

[0026] According to the classification results of each processing state identification sample by the current weak classifier, the weight of each processing state identification sample is adjusted to obtain a processing state identification data set after the weight is adjusted; when adjusting the weight of each processing state identification sample, the weight of the processing state identification sample that is incorrectly classified by the weak classifier is increased; and the weight of the processing state identification sample that is correctly classified by the weak classifier is reduced.

[0027] Based on the weight-adjusted processing state identification sample data set, a weak classifier is trained and obtained as the current weak classifier.

[0028] Determine whether the iteration termination condition is met. If so, determine the classification error rate of each weak classifier during the iterative training process based on the processing state identification sample data set after weight adjustment; otherwise jump to the step "adjust the weight of each processing state identification sample according to the classification result of each processing state identification sample by the current weak classifier to obtain the processing state identification data set after weight adjustment".

[0029] According to the classification error rate of each weak classifier, several weak classifiers are combined to obtain a machine tool processing state identification model with strong classification ability.

[0030] Optionally, a hierarchical tool wear state integrated learning model is constructed to learn the mapping relationship between the multi-source sensor features under any existing tool processing state of the machine tool and the tool wear state of the existing tool, specifically including the following steps:

[0031] A wear state identification data set is constructed; the wear state identification data set includes a number of wear state identification samples; the wear state identification samples include multi-source sensor features and corresponding tool wear states under any existing tool processing state of the machine tool.

[0032] According to the wear state identification data set, a number of wear state identification data subsets are obtained by partitioning with replacement, and a low-level base learner is trained respectively to obtain a number of low-level base learners.

[0033] For any wear state identification sample, the multi-source sensor states under the processing state in the wear state identification sample are respectively input into several low-level base learners to obtain several low-level prediction results of the wear state identification sample.

[0034] Several low-level prediction results of the wear state identification sample are combined into a feature vector, and the feature vector is combined with the tool wear state of the wear state identification sample to obtain a high-level feature fusion sample of the wear state identification sample.

[0035] Based on the high-level feature fusion samples of each wear state identification sample, a high-level meta-model is trained to accurately divide the tool wear state of any existing tool of the machine tool; for any high-level feature fusion sample, the feature vector in the high-level feature fusion sample is used as the input data of the high-level meta-model, and the corresponding tool wear state is used as the target output of the high-level meta-model.

[0036] Optionally, constructing and training a tool wear prediction model for existing tools under different wear conditions specifically includes the following steps:

[0037] For any wear state of the existing tool, a wear amount prediction data set under the wear state is constructed; the wear amount prediction data set includes several wear amount prediction samples; the wear amount prediction samples include multi-source sensor features of the existing tool processing state and the actual wear amount of the existing tool under the wear state.

[0038] An initial tool wear prediction model is constructed based on a temporal convolutional network with learnable activation functions.

[0039] The initial tool wear prediction model is trained based on the wear prediction data set to obtain the tool wear prediction model of the existing tool under the wear state.

[0040] Optionally, the temporal convolutional network is based on a recurrent neural network that models time series dependencies, introduces causal convolution to avoid the model's dependence on future time step data, and uses dilated convolution units to expand the model's long-distance time series capture capability; an activation function learnable mechanism is introduced into the tool wear prediction model, and the activation function learnable mechanism transforms the weight of each neuron in the temporal convolutional network into a learnable parameter of a univariate function, thereby enhancing the tool wear prediction model's ability to perceive global and local information.

[0041] Optionally, for the newly added tool, according to the knowledge graph of collaborative processing of multiple tools, the nearest tool to the newly added tool is determined among the existing tools of the machine tool, specifically including the following steps:

[0042] Build a knowledge graph containing the model and size features of multiple tools for collaborative machining, and vectorize new tools and existing tools.

[0043] The matching degree between the newly added tool and each existing tool is calculated based on the cosine similarity algorithm, and the nearest tool to the newly added tool is determined.

[0044] Optionally, based on the root mean square error between the actual wear amount and the predicted wear amount of the newly added tool, the wear amount prediction guidance model is updated online, and the difference convergence mechanism between the actual tool wear process and the virtual wear amount prediction guidance model is used to drive the difference convergence between the actual tool wear process and the virtual wear amount prediction guidance model, so as to realize the precise deduction of the machine tool machining accuracy under multi-tool collaborative machining, which specifically includes the following steps:

[0045] The multi-source sensing features of the newly added tool in each machining process are collected, and the actual wear amount of the newly added tool is collected at intervals of a preset number of machining processes.

[0046] Based on the wear prediction guidance model of the newly added tool, the corresponding predicted wear amount is output according to the multi-source sensing characteristics of the newly added tool during the machining process.

[0047] The root mean square error between the actual wear amount and the predicted wear amount of the newly added tool is used as the additional loss function term of the wear amount prediction guidance model. The structure and parameters of the wear amount prediction guidance model are updated online. The virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear amount prediction guidance model, so as to realize the accurate deduction of machine tool processing accuracy under multi-tool collaborative processing.

[0048] Optionally, a multi-source sensor data set is obtained during the machine tool processing process, specifically, acoustic emission sensors, vibration sensors, current sensors and cutting force sensors are arranged near multiple tools in collaborative processing of the machine tool to collect the multi-source sensor data set during the machine tool processing process.

[0049] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0050] The present application provides a method for deducing the precision of collaborative machining of multiple tools of a machine tool based on the convergence of virtual and real differences. First, a multi-source sensor data set in the machining process of the machine tool is obtained, and the data is screened through a machining state identification model of the machine tool to obtain the multi-source sensor data in the machining state, and the multi-source sensor features in the machining state are determined to avoid the interference of the sensor data in the non-machining state; secondly, a hierarchical tool wear state integrated learning model is constructed to learn the mapping relationship between the multi-source sensor features of any existing tool of the machine tool in the machining state and the tool wear state of the existing tool, so as to construct and train a tool wear prediction model for the existing tool in different wear states, so that the tool wear prediction task can adapt to the characteristics of significant changes in wear characteristics; then, according to the knowledge graph of collaborative machining of multiple tools, the nearest tool of the newly added tool is determined among the existing tools of the machine tool, and its tool wear prediction model is used as a guide for the prediction of the wear of the newly added tool; finally, according to the root mean square error between the real wear and the predicted wear, the wear prediction guidance model is updated online to drive the difference between the real tool wear process and the virtual wear prediction guidance model to converge, so as to realize the precise deduction of the machining precision of the machine tool under the collaborative machining of multiple tools.

[0051] The above solution of this application has the following beneficial effects:

[0052] (1) The multi-source sensor data under the machining state is screened out through the machine tool machining state identification model, and the multi-source sensor characteristics under the machining state are determined, which effectively avoids the interference of massive non-machining redundant data and provides high-quality data support for the tool wear state identification process.

[0053] (2) A hierarchical tool wear state ensemble learning model was used to accurately identify the tool wear state, and based on this, an accurate and explainable tool wear prediction model was established, achieving accurate prediction of tool wear using a progressive strategy.

[0054] (3) In addition, by constructing a knowledge graph for collaborative machining of multiple tools, using tool similarity evaluation to guide the wear prediction of new tools, and using the virtual-real difference convergence mechanism to drive the online update of the new tool wear prediction model, the problem of long test cycles and the need to conduct tool wear characteristic tests for each new tool is effectively overcome, thus achieving accurate, fast, and efficient deduction of machine tool machining accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flowchart of a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences is provided in one embodiment of the present application.

[0057] Figure 2 A flowchart of step A1 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application.

[0058] Figure 3 A flowchart of step A12 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application.

[0059] Figure 4 A flowchart of a machine tool machining state identification model training process in a machine tool multi-tool collaborative machining accuracy deduction method based on virtual-real difference convergence provided in one embodiment of the present application.

[0060] Figure 5A flowchart of step A2 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application.

[0061] Figure 6 A flowchart of step A3 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application.

[0062] Figure 7 A flowchart of step A4 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application.

[0063] Figure 8 A flowchart of step A6 in a method for deducing precision of collaborative machining of multiple tools of a machine tool based on convergence of virtual and real differences provided in one embodiment of the present application. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0065] In an exemplary embodiment, Figure 1 The flowchart shown provides a method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence, including the following steps:

[0066] A1. Obtain a multi-source sensor data set during the machine tool processing process, filter the multi-source sensor data under the processing state through the machine tool processing state identification model, and determine the multi-source sensor features under the processing state. In this embodiment, Figure 2 In the flowchart shown, step A1 specifically includes the following steps:

[0067] A11. Acquire a multi-source sensor data set in a machine tool processing process; the multi-source sensor data set includes a plurality of multi-source sensor data of unit time length. Specifically in this embodiment, an acoustic emission sensor, a vibration sensor, a current sensor, and a cutting force sensor are arranged near multiple cutting tools in collaborative machining of a machine tool to collect a multi-source sensor data set in a machine tool processing process.

[0068] A12. Based on the synchronous compression wavelet transform algorithm, each segment of multi-source sensor data in the machine tool processing process is converted into two-dimensional image data.

[0069] The basic principle of the synchronous compression wavelet transform algorithm adopted in this step is to decompose the signal data into components of different frequencies and time scales through wavelet transform, and then compress these components using sparse representation methods. In this way, on the basis of removing redundant information, the time and frequency characteristics of the sensor data can be extracted with high quality, so as to better understand and analyze the signal. Figure 3 As shown in the flowchart, in this embodiment, step A12 specifically includes the following steps:

[0070] A121. For any section of multi-source sensor data in the machine tool processing process, the multi-source sensor data is decomposed by wavelet transform to obtain components at different frequencies and time scales.

[0071] A122. Use sparse representation methods to compress the components of multi-source sensor data at different frequencies and time scales to obtain a time-frequency feature matrix.

[0072] A123. Convert the time-frequency feature matrix into a two-dimensional image form to obtain two-dimensional image data corresponding to the multi-source sensor data.

[0073] A13. Input the two-dimensional image data corresponding to each segment of multi-source sensor data into the machine tool processing state identification model to determine the machine tool working state corresponding to each segment of multi-source sensor data.

[0074] A14. According to the machine tool working state corresponding to each segment of multi-source sensor data, the multi-source sensor data set in the machine tool processing process is screened to obtain the multi-source sensor data under the processing state, and the multi-source sensor features under the processing state are determined. The multi-source sensor feature refers to a feature whose Pearson coefficient between the tool wear state and the tool wear state is greater than a preset threshold value among several features represented in the time domain, frequency domain and time-frequency domain of the multi-source sensor data.

[0075] In this embodiment, after screening the multi-source sensor data under the processing state, it is also necessary to determine the sensor features that are strongly correlated with the tool wear condition. Specifically, the sensor features refer to the time domain (maximum value, mean, standard deviation, root mean square value), frequency domain (frequency mean, frequency center of gravity, root mean square frequency, standard deviation frequency), and time-frequency domain (sub-band energy percentage after wavelet analysis) of the sensor data. The Pearson coefficient between the above-mentioned sensor features of each sensor and the tool wear state can be determined respectively, and the sensor features that are greater than a preset threshold are selected. For example, when the preset threshold is 0.9, the sensor features with a Pearson coefficient greater than 0.9 between the sensor features and the tool wear state are selected as the sensor features for subsequent wear state identification and wear amount prediction.

[0076] Specifically in this embodiment, the machine tool processing state identification model includes a convolution module, a bidirectional gated loop module and a strong classification module; before executing step A13, the machine tool multi-tool collaborative processing accuracy deduction method based on virtual-real difference convergence proposed in this embodiment also has a training process of the machine tool processing state identification model, such as Figure 4 The flowchart shown includes the following steps:

[0077] B1. Construct a processing state identification data set; the processing state identification data set includes a number of processing state identification samples; the processing state identification samples include two-dimensional image data corresponding to multi-source sensor data under the processing state and the corresponding machine tool working state.

[0078] B2. Initialize the weight of each processing state identification sample in the processing state identification data set to obtain a processing state identification data set after weight initialization; the weights of each processing state identification sample in the processing state identification data set after weight initialization are the same.

[0079] B3. Based on the weight-initialized processing state identification sample data set, a weak classifier is trained and obtained as the current weak classifier.

[0080] B4. According to the classification results of the current weak classifier for each processing state identification sample, the weight of each processing state identification sample is adjusted to obtain a processing state identification data set after the weight is adjusted; when adjusting the weight of each processing state identification sample, the weight of the processing state identification sample that is incorrectly classified by the weak classifier is increased; and the weight of the processing state identification sample that is correctly classified by the weak classifier is reduced.

[0081] B5. Based on the weight-adjusted processing state identification sample data set, a weak classifier is trained and obtained as the current weak classifier.

[0082] B6. Determine whether the iteration termination condition is met. If so, execute step B7; otherwise, jump to step B4 "adjust the weight of each processing state identification sample according to the classification result of each processing state identification sample by the current weak classifier to obtain a processing state identification data set after weight adjustment".

[0083] B7. Determine the classification error rate of each weak classifier in the iterative training process according to the processing state identification sample data set after weight adjustment.

[0084] B8. According to the classification error rate of each weak classifier, several weak classifiers are combined to obtain a machine tool processing state identification model with strong classification ability.

[0085] Specifically, after combining several weak classifiers to obtain a strong classifier, for the two-dimensional image data input into the machine tool processing state recognition model, the convolution module is first used to extract the features of the two-dimensional image data, and the bidirectional gated loop module is used to further extract the timing features. Finally, the strong classifier is used to accurately distinguish the processing state or non-processing state represented by the sensor data corresponding to the two-dimensional image data. Before proceeding to the next step, the sensor data of the non-processing state is first eliminated to avoid interference.

[0086] A2. Construct a hierarchical tool wear state integrated learning model to learn the mapping relationship between the multi-source sensor features of any existing tool in the machining state of the machine tool and the tool wear state of the existing tool; the hierarchical tool wear state integrated learning model can determine the tool wear state of the existing tool according to the multi-source sensor features under the input machining state. In this embodiment, Figure 5 As shown in the flowchart, A2 specifically includes the following steps:

[0087] A21. Construct a wear state identification data set; the wear state identification data set includes a number of wear state identification samples; the wear state identification samples include multi-source sensor features under the processing state of any existing tool of the machine tool and the corresponding tool wear state of any existing tool of the machine tool.

[0088] A22. According to the wear state identification data set, several wear state identification data subsets are obtained by partitioning with replacement, and a low-level base learner is trained for each of them to obtain several low-level base learners. The collected machine tool processing data is divided into multiple subsets with replacement, and low-level base learners such as decision trees, support vector machines, neural networks, etc. are used for training, which can all be used to generate classification results for input data.

[0089] A23. For any wear state identification sample, the multi-source sensor features under the processing state in the wear state identification sample are respectively input into a number of low-level base learners to obtain a number of low-level prediction results of the wear state identification sample.

[0090] A24. Combining several low-level prediction results of the wear state identification sample into a feature vector, combining the feature vector with the tool wear state of the wear state identification sample, and obtaining a high-level feature fusion sample of the wear state identification sample. Specifically, the prediction results output by several low-level base learners according to the original input data are stacked in columns, fused to form a feature vector as new input data, and then the new input data is used as the input of the high-level meta-model. The label of the new input data is still the label of the original input data, and is used as the target output of the high-level meta-model.

[0091] A25. Based on the high-level feature fusion samples of each wear state identification sample, a high-level meta-model is trained to accurately classify the tool wear state of any existing tool of the machine tool; for any high-level feature fusion sample, the feature vector in the high-level feature fusion sample is used as the input data of the high-level meta-model, and the corresponding tool wear state is used as the target output of the high-level meta-model. In this embodiment, the logistic regression method is used as the high-level meta-model, and the classification results of each low-level base learner are integrated to achieve accurate prediction of the tool wear state.

[0092] A3. For any existing tool of a machine tool, a tool wear prediction model under different wear conditions is constructed and trained. Figure 6 In the flowchart shown, step A3 specifically includes the following steps:

[0093] A31. For any wear state of the existing tool, a wear amount prediction data set under the wear state is constructed; the wear amount prediction data set includes a number of wear amount prediction samples; the wear amount prediction samples include multi-source sensor features of the existing tool processing state and the actual wear amount of the existing tool under the wear state.

[0094] A32. Based on the temporal convolutional network with learnable activation functions, an initial tool wear prediction model is constructed.

[0095] A33. The initial tool wear prediction model is trained based on the wear prediction data set to obtain a tool wear prediction model for the existing tool under the wear state.

[0096] The temporal convolutional network used in step A3 is based on the recurrent neural network that models time series dependencies. It introduces causal convolution to avoid the model's dependence on future time step data, and uses dilated convolution units to expand the model's long-distance time series capture capability.

[0097] In an exemplary embodiment, an activation function learnable mechanism is introduced into the tool wear prediction model. The activation function learnable mechanism transforms the weight of each neuron in the temporal convolutional network into a learnable parameter of a univariate function, thereby enhancing the tool wear prediction model's ability to perceive global and local information.

[0098] A4. For the newly added tool, the nearest tool to the newly added tool is determined from the existing tools of the machine tool according to the knowledge graph of collaborative processing of multiple tools. Figure 7 In the flowchart shown, step A4 specifically includes the following steps:

[0099] A41. Construct a knowledge graph containing the model and size features of collaborative processing multiple tools, and vectorize the newly added tools and each existing tool. Specifically, the constructed knowledge graph contains data such as the model and size of collaborative processing multiple tools (such as milling cutters, grinding wheels, etc.).

[0100] A42. Calculate the matching degree between the new tool and each existing tool based on the cosine similarity algorithm, and determine the nearest neighboring tool of the new tool. Determine the matching degree between the new tool and each tool based on the knowledge graph, so as to use the wear prediction model of the nearest tool to guide the wear prediction of the new tool processing process and reduce the implementation of tool wear characteristic tests.

[0101] A5. The tool wear prediction model of the nearest tool under different wear conditions is used as the wear prediction guidance model of the newly added tool under different wear conditions; the wear prediction guidance model can determine the predicted wear of the newly added tool according to the multi-source sensor characteristics under the input processing state.

[0102] A6. For any wear prediction guidance model of the newly added tool, the wear prediction guidance model is updated online according to the root mean square error between the actual wear of the newly added tool and the predicted wear, and the virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear prediction guidance model, so as to achieve accurate deduction of machine tool processing accuracy under multi-tool collaborative processing. Figure 8 As shown in the flowchart, in this embodiment, step A6 specifically includes the following steps:

[0103] A61. Collect the multi-source sensor characteristics of the newly added tool during each processing process, and collect the actual wear amount of the newly added tool at intervals of a preset number of processing processes.

[0104] A62. Based on the wear prediction guidance model of the newly added tool, the corresponding predicted wear amount is output according to the multi-source sensing characteristics of the newly added tool during the processing.

[0105] A63. The root mean square error between the actual wear amount and the predicted wear amount of the newly added tool is used as the additional loss function term of the wear amount prediction guidance model, and the structure and parameters of the wear amount prediction guidance model are updated online. The virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear amount prediction guidance model, so as to realize the precise deduction of machine tool processing accuracy under multi-tool collaborative processing.

[0106] Compared with the prior art, the solution provided by the above-mentioned embodiments of the present application, by collecting multi-source sensor data near the machine tool and converting it into images, extracting the sensor features under the processing state, effectively avoiding the interference of massive non-processing redundant data, and providing high-quality data support for the tool wear identification process; by establishing a hierarchical tool wear state classification model, a precise and explainable tool wear prediction model is established by adopting a temporal convolutional network with an activation function learning mechanism, and accurate prediction of tool wear is achieved with a progressive strategy; based on the virtual-real difference convergence strategy, the structure and parameters of the model are updated online, which effectively overcomes the problem of long tool wear characteristic test cycle and realizes the online deduction of machine tool processing accuracy under multi-tool collaboration.

[0107] 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.

[0108] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence, characterized in that: include: Acquire a multi-source sensor data set during the machine tool processing process, filter the multi-source sensor data under the processing state through the machine tool processing state identification model, and determine the multi-source sensor features under the processing state; Constructing a hierarchical tool wear state integrated learning model to learn the mapping relationship between the multi-source sensor characteristics of any existing tool of the machine tool under the processing state and the tool wear state of the existing tool; the hierarchical tool wear state integrated learning model can determine the tool wear state of the existing tool according to the multi-source sensor characteristics under the input processing state; For any existing tool of a machine tool, a tool wear prediction model under different wear conditions is constructed and trained; For the newly added tool, the nearest tool to the newly added tool is determined among the existing tools of the machine tool according to the knowledge graph of collaborative machining of multiple tools; The tool wear prediction model of the nearest tool under different wear states is used as the wear prediction guidance model of the newly added tool under different wear states; the wear prediction guidance model can determine the predicted wear amount of the newly added tool according to the multi-source sensor characteristics under the input processing state; For any wear prediction guidance model of the newly added tool, the wear prediction guidance model is updated online according to the root mean square error between the real wear and the predicted wear of the newly added tool, and the virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear prediction guidance model, so as to realize the accurate deduction of machine tool processing accuracy under multi-tool collaborative processing; Obtain a multi-source sensor data set during the machine tool processing process, filter the multi-source sensor data under the processing state through the machine tool processing state identification model, and determine the multi-source sensor features under the processing state, including: Acquire a multi-source sensor data set in a machine tool processing process; the multi-source sensor data set includes a plurality of multi-source sensor data of unit time length; Based on the synchronous compression wavelet transform algorithm, the multi-source sensor data of each section in the machine tool processing process are converted into two-dimensional image data respectively; Inputting the two-dimensional image data corresponding to each segment of multi-source sensor data into the machine tool processing state recognition model to determine the machine tool working state corresponding to each segment of multi-source sensor data; According to the working state of the machine tool corresponding to each section of multi-source sensor data, the multi-source sensor data under the processing state is screened from the multi-source sensor data set in the processing process of the machine tool, and the multi-source sensor feature under the processing state is determined; the multi-source sensor feature refers to the feature whose Pearson coefficient between the tool wear state and the tool wear state is greater than a preset threshold value, determined in the time domain, frequency domain and time-frequency domain representation of the multi-source sensor data; The machine tool processing state identification model includes a convolution module, a bidirectional gated loop module and a strong classification module; before inputting the two-dimensional image data corresponding to each segment of multi-source sensor data into the machine tool processing state identification model and determining the machine tool working state corresponding to each segment of multi-source sensor data, the machine tool multi-tool collaborative processing accuracy deduction method based on virtual-real difference convergence also includes: Constructing a processing state identification data set; the processing state identification data set includes a plurality of processing state identification samples; the processing state identification samples include two-dimensional image data corresponding to multi-source sensor data under the processing state and the corresponding machine tool working state; Initializing the weight of each processing state identification sample in the processing state identification data set to obtain a processing state identification data set after weight initialization; the weights of each processing state identification sample in the processing state identification data set after weight initialization are the same; Based on the processing state identification sample data set after weight initialization, a weak classifier is trained and obtained as the current weak classifier; According to the classification results of the current weak classifier on each processing state identification sample, the weight of each processing state identification sample is adjusted to obtain a processing state identification data set after the weight is adjusted; when adjusting the weight of each processing state identification sample, for the processing state identification sample that is incorrectly classified by the weak classifier, the weight of the processing state identification sample is increased; for the processing state identification sample that is correctly classified by the weak classifier, the weight of the processing state identification sample is reduced; Based on the weight-adjusted processing state identification sample data set, a weak classifier is trained and obtained as the current weak classifier; Determine whether the iteration termination condition is met. If so, determine the classification error rate of each weak classifier in the iterative training process according to the weight-adjusted processing state identification sample data set; otherwise, jump to the step of "adjusting the weight of each processing state identification sample according to the classification result of each processing state identification sample by the current weak classifier to obtain the weight-adjusted processing state identification data set"; According to the classification error rate of each weak classifier, a plurality of the weak classifiers are combined to obtain a machine tool processing state identification model with strong classification ability; A hierarchical tool wear state integrated learning model is constructed to learn the mapping relationship between the multi-source sensor features of any existing tool processing state of the machine tool and the tool wear state of the existing tool, specifically including: Constructing a wear state identification data set; the wear state identification data set includes a plurality of wear state identification samples; the wear state identification samples include multi-source sensing features and corresponding tool wear states under any existing tool processing state of the machine tool; According to the wear state identification data set, a plurality of wear state identification data subsets are obtained by partitioning with replacement, and a low-level base learner is trained respectively to obtain a plurality of low-level base learners; For any wear state identification sample, the multi-source sensor features under the processing state in the wear state identification sample are respectively input into a plurality of low-level base learners to obtain a plurality of low-level prediction results of the wear state identification sample; Combining a plurality of the low-level prediction results of the wear state identification sample into a feature vector, and combining the feature vector with the tool wear state of the wear state identification sample to obtain a high-level feature fusion sample of the wear state identification sample; Based on the high-level feature fusion samples of each wear state identification sample, a high-level meta-model is trained to accurately divide the tool wear state of any existing tool of the machine tool; for any high-level feature fusion sample, the feature vector in the high-level feature fusion sample is used as the input data of the high-level meta-model, and the corresponding tool wear state is used as the target output of the high-level meta-model.

2. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 1 is characterized in that: Based on the synchronous compression wavelet transform algorithm, the multi-source sensor data of each section in the machine tool processing process is converted into two-dimensional image data, including: For any section of multi-source sensor data in the machine tool processing process, the multi-source sensor data is decomposed by wavelet transform to obtain components at different frequencies and time scales; Using a sparse representation method to compress the components of the multi-source sensor data at different frequencies and time scales to obtain a time-frequency feature matrix; The time-frequency feature matrix is ​​converted into a two-dimensional image form to obtain two-dimensional image data corresponding to the multi-source sensor data.

3. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 1 is characterized in that: Constructing and training a tool wear prediction model for the existing tool under different wear conditions, specifically including: For any wear state of an existing tool, a wear amount prediction data set under the wear state is constructed; the wear amount prediction data set includes a plurality of wear amount prediction samples; the wear amount prediction samples include multi-source sensing features under the existing tool processing state and the actual wear amount of the existing tool under the wear state; Based on the temporal convolutional network with learnable activation functions, an initial tool wear prediction model is constructed; The initial tool wear prediction model is trained based on the wear prediction data set to obtain a tool wear prediction model for the existing tool under the wear state.

4. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 3 is characterized in that: The temporal convolutional network is based on a recurrent neural network that models time series dependencies, introduces causal convolution to avoid the model's dependence on future time step data, and uses dilated convolution units to expand the model's long-distance time series capture capability; an activation function learnable mechanism is introduced into the tool wear prediction model, and the activation function learnable mechanism transforms the weight of each neuron in the temporal convolutional network into a learnable parameter of a univariate function, thereby enhancing the tool wear prediction model's perception of global and local information.

5. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 1 is characterized in that: For the newly added tool, according to the knowledge graph of collaborative processing of multiple tools, the nearest tool to the newly added tool is determined among the existing tools of the machine tool, specifically including: Build a knowledge graph containing the model and size features of multiple tools for collaborative machining, and vectorize the newly added tools and existing tools; The matching degree between the newly added tool and each existing tool is calculated based on the cosine similarity algorithm, and the nearest tool to the newly added tool is determined.

6. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 1 is characterized in that: According to the root mean square error between the real wear amount and the predicted wear amount of the newly added tool, the wear amount prediction guidance model is updated online, and the difference convergence mechanism between the real tool wear process and the virtual wear amount prediction guidance model is used to drive the difference convergence between the real tool wear process and the virtual wear amount prediction guidance model, so as to realize the accurate deduction of the machine tool processing accuracy under multi-tool collaborative processing, specifically including: Collecting multi-source sensing features of the newly added tool during each machining process, and collecting the actual wear amount of the newly added tool at intervals of a preset number of machining processes; Based on the wear prediction guidance model of the newly added tool, outputting the corresponding predicted wear amount according to the multi-source sensing characteristics of the newly added tool during the machining process; The root mean square error between the actual wear amount of the newly added tool and the predicted wear amount is used as an additional loss function term of the wear amount prediction and guidance model. The structure and parameters of the wear amount prediction and guidance model are updated online. The virtual-real difference convergence mechanism is used to drive the difference convergence between the real tool wear process and the virtual wear amount prediction and guidance model, so as to realize the accurate deduction of machine tool processing accuracy under multi-tool collaborative processing.

7. The method for deducing the precision of multi-tool collaborative machining of machine tools based on virtual-real difference convergence according to claim 1 is characterized in that: A multi-source sensor data set is obtained during the machine tool processing process. Specifically, acoustic emission sensors, vibration sensors, current sensors and cutting force sensors are arranged near multiple tools in collaborative processing of the machine tool to collect the multi-source sensor data set during the machine tool processing process.

Images