Numerical control lathe tool wear monitoring method and system

Abstract

The invention discloses a numerical control lathe tool wear monitoring method and system. The method comprises the steps that numerical control lathe tool operation process data are monitored in real time; a corresponding preset operation parameter threshold value is matched according to the operation process data, and when it is monitored that the preset operation parameter threshold value is exceeded, the operation process data of the numerical control lathe tool is transmitted to an edge computing node; matching a corresponding preset operation parameter combination and finishing operation parameter screening according to the operation process data by utilizing the edge computing node; based on the screened operation parameters, key features are selected and extracted by using a feature selection algorithm, and an optimal feature set is acquired and uploaded to a cloud platform; and matching a corresponding tool wear prediction model according to the operation process data at the cloud platform, inputting the optimal feature set into the matched tool wear prediction model, obtaining a wear value, a wear degree and a life prediction result, and performing early warning decision and early warning prompt. According to the invention, efficient, accurate and real-time tool wear monitoring under complex working conditions is realized.

Description

Technical Field

[0001] The present application relates to the technical field of tool wear monitoring for CNC lathes, and in particular to a tool wear monitoring method and system for CNC lathes. Background Art

[0002] As an important machine tool equipment, CNC lathe plays a vital role in modern manufacturing. With the continuous improvement of industrial automation level, the application scope of CNC lathe is becoming more and more extensive, which not only improves production efficiency, but also significantly improves product quality and stability. However, the wear problem of CNC lathe tools has always been one of the important factors restricting its performance. Tool wear will lead to reduced processing accuracy, poor surface quality, and may even cause safety accidents. Therefore, it is particularly important to monitor the tool wear status in a timely and accurate manner.

[0003] In the existing technology, in order to deal with the problem of tool wear in CNC lathes, the following methods are usually used: the first is the regular inspection method, that is, the wear condition is judged by manual visual inspection or regular detection of tool size changes using measuring tools; the second is the vibration / temperature / emission monitoring method, which uses sensors to monitor the vibration signal of the tool during work / the temperature rise of the tool during the cutting process / capture the tiny sound wave signal generated by the tool during the cutting process, and analyze the change rules of these signals to evaluate the degree of tool wear; the third is to use intelligent data processing methods and make full use of big data and machine learning technology to predict and warn of tool loss.

[0004] Although the above methods can effectively monitor the wear status of the tool to a certain extent, there are still some shortcomings. For example, the regular inspection method relies on the experience and technical level of the operator, is easily affected by human factors, and cannot achieve real-time monitoring; although the single vibration / temperature / emission monitoring method can achieve online monitoring, the signal interference is large under complex working conditions, and the single signal processing still has defects in measurement accuracy; although intelligent data processing methods can achieve real-time and comprehensive monitoring to a certain extent, most of the prediction methods in the existing technology lack more detailed monitoring of different tool types, different processing materials, different processing conditions, etc., resulting in the prediction accuracy still needs to be improved, and detailed monitoring often increases a lot of calculations. The lightweight model built into the CNC machine tool cannot meet the computing power, so there are still defects in achieving timely and accurate monitoring.

[0005] In summary, there is a need to provide a tool wear monitoring system that can achieve efficient, accurate and real-time tool wear monitoring, especially tool wear monitoring system that has adaptability and reliability under complex working conditions. Summary of the invention

[0006] In order to achieve efficient, accurate and real-time tool wear monitoring under complex working conditions and timely replace tools, the present application provides a CNC lathe tool wear monitoring method and system.

[0007] In a first aspect, the present application provides a method for monitoring tool wear of a CNC lathe, comprising: Real-time monitoring of CNC lathe tool operation process data, including: tool type, processing material, processing conditions and operating parameters; According to the tool type, processing material, and processing conditions, the preset operating parameter threshold is matched. When the real-time operating parameter is detected to exceed the preset operating parameter threshold, the current CNC lathe tool operation process data is transmitted to the edge computing node; Utilize edge computing nodes to match the corresponding preset operating parameter combinations according to tool type, processing material, and processing conditions, and complete operating parameter screening according to the matched preset operating parameter combinations; based on the screened operating parameters, use feature selection algorithms to select and extract key features, obtain the optimal feature set, and upload it to the cloud platform with the corresponding tool type, processing material, and processing conditions; On the cloud platform, the pre-built tool wear prediction model is matched according to the tool type, processing material, and processing conditions, and the optimal feature set is input into the matching tool wear prediction model to obtain the wear value, wear degree, and life prediction results; Based on the acquired wear values, wear degree and life prediction results, early warning decisions are made and early warning prompts are generated.

[0008] By adopting the above solution, the tool operating parameters under complex working conditions are monitored in real time and dynamically matched with the preset operating parameter thresholds, abnormal wear conditions are effectively identified and relevant data is transmitted to the edge computing node; the edge computing node is used to screen and select features of the operating parameters, obtain the optimal feature set and upload it to the cloud platform, to ensure data quality, reduce data transmission volume, and improve response speed; and on the cloud platform, through adaptive matching of the tool wear prediction model, the wear degree and life of the tool are accurately predicted, and the prediction accuracy is improved, so that effective early warning decisions can be made based on the prediction results, assisting in timely replacement of tools, and improving production efficiency and equipment utilization.

[0009] Preferably, the matching of the preset operating parameter thresholds according to the tool type, processing material, and processing conditions includes: Collect the operating parameters and tool wear values ​​corresponding to the tool operation process under different tool types, different processing materials, and different processing conditions, and use the deep learning algorithm to obtain the mapping relationship between the operating parameters and wear values ​​of the tool during operation under the conditions of a single tool, a single processing material, and a single or combined processing condition; According to the constructed mapping relationship, the threshold range of the operating parameters in different wear stages under the known tool type, known processing material and known processing conditions is analyzed and obtained; the different wear stages are determined according to the correlation curve generated by calculating the operation time and the wear value during the operation of the tool; the different wear stages include light wear stage, medium wear stage and heavy wear stage; Receive the user's demand for tool wear monitoring, and match the preset threshold range of the operating parameters in the specific wear stage according to the user's demand for tool wear monitoring; determine the preset operating parameter threshold value according to the matched threshold range of the operating parameters in the specific wear stage, so as to complete the matching of the preset operating parameter threshold value according to the tool type, processing material and processing conditions collected in real time.

[0010] By adopting the above solution, historical data is collected and a deep learning algorithm is used to establish a mapping relationship between operating parameters and tool wear values. Then, the preset operating parameter thresholds are determined based on the operating parameter threshold ranges at different wear stages and the user's monitoring needs. This avoids false alarms or omissions in wear monitoring due to fixed thresholds and improves user experience.

[0011] Preferably, the matching of corresponding preset operating parameter combinations according to tool type, processing material, and processing conditions includes: Traverse different tool types, processing materials and processing conditions, respectively count the operating parameters corresponding to the tool operation process under the historical single tool type, single processing material and single or combined processing conditions, and set the preset operating parameter combination generated based on different operating parameters; According to each set of preset operating parameter combinations, historical operating parameters and corresponding wear values ​​are screened out and divided into training sets and test sets. According to the test set, the accuracy of the prediction results of the tool wear prediction model generated by training the training set is obtained, and the preset operating parameter combinations are arranged in order of accuracy to generate a ranking; The preset operating parameter combinations generated under the same combination conditions according to the tool type, processing material, and processing conditions are sorted, and a group of preset operating parameter combinations corresponding to the tool wear prediction model with the highest accuracy are used as the matching preset operating parameter combinations according to the preset operating parameter combinations sorting. When the actual operating parameters are selected, if some operating parameters in the currently matched preset operating parameter combination are missing in the real-time operating parameters, the preset operating parameter combinations are sorted, and the preset operating parameter combination with the next priority and an accuracy greater than the preset accuracy is selected as the re-matched preset operating parameter combination. If the preset operating parameter combination with an accuracy greater than the preset accuracy cannot be queried according to the priority, the preset operating parameter combination with the highest comprehensive score is selected from all the preset operating parameter combinations with an accuracy greater than the preset accuracy as the final matching preset operating parameter combination, thereby completing the matching of the corresponding preset operating parameter combinations according to the tool type, processing material, and processing conditions and selecting the corresponding operating parameters; the comprehensive score is obtained by weighted calculation based on the similarity score with the real-time operating parameters and the accuracy score of completing the real-time operating parameter mapping supplement based on the missing operating parameters.

[0012] By adopting the above scheme, a comprehensive analysis of historical data is carried out, training and testing are completed, and the best preset operating parameter combination is determined to maximize the reflection of tool performance under actual working conditions. Taking into account the situation that the actually collected data may not be able to adapt to the best preset operating parameter combination, the preset operating parameter combination that meets the accuracy requirements is selected in sequence, and when the preset operating parameter combination that meets the requirements cannot be obtained, the accuracy of the missing data and the similarity score with the actual operating parameters are selected to comprehensively judge and obtain the preset operating parameter combination with the highest score, and the best operating parameters that can reflect tool wear under the current working conditions are adaptively screened out, thereby improving the accuracy of subsequent predictions.

[0013] Preferably, the screening-based operating parameters use a feature selection algorithm to select and extract key features to obtain an optimal feature set, including: When the preset operating parameter threshold is determined according to the threshold range of the operating parameter in the light wear stage or the moderate wear stage, the corresponding feature selection algorithm adopts an integrated feature selection algorithm; the integrated feature selection algorithm includes: preliminarily using a variance threshold or a mutual information algorithm to complete a preliminary screening of target features, and then using a random forest feature importance ranking technique to complete a secondary screening of features; after using the integrated feature selection algorithm to select and extract key features, according to the first preset number of features matching the light wear stage or the moderate wear stage, retaining the same number of features according to the sorting of the screened features; When the preset operating parameter threshold is determined according to the threshold range of the operating parameter in the severe wear stage, the corresponding feature selection algorithm adopts a single feature selection algorithm; the single feature selection algorithm includes: using a variance threshold or a mutual information algorithm to screen and sort the importance of the target feature; after using the feature selection algorithm to select key features, according to the second preset number of features matching the severe wear stage, the same number of features are retained according to the sorting of the screened features.

[0014] Through the above scheme, the appropriate feature selection algorithm is selected according to the threshold range of the operating parameters in different wear stages, and the operating parameters selected by the operating parameter threshold in the light wear stage or the medium wear stage are screened multiple times to retain only a certain number of the most important features, thereby improving the selection accuracy of key features and improving the accuracy and robustness of the model; the operating parameters selected by the operating parameter threshold in the heavy wear stage are screened once to retain a considerable number of features, thereby ensuring the effectiveness of the selected key features and the prediction performance of the model under heavy wear conditions.

[0015] Preferably, the pre-built tool wear prediction model corresponding to the tool type, processing material, and processing condition matching includes: The tool wear prediction model includes multiple prediction models, each of which includes: a multi-branch feature input layer, a multi-branch feature fusion layer and a multi-branch prediction result output layer, wherein the output corresponding to the branch outputting the wear value and the branch outputting the wear degree is set as the input of the branch outputting the life prediction result, and is generated through a set of historical optimal feature sets, tool wear values, tool wear degrees and tool remaining life training under a single tool type, a single processing material and a single or combined processing condition.

[0016] Through the above scheme, by utilizing the design of multi-branch feature input layer, multi-branch feature fusion layer and multi-branch prediction result output layer, the model can more comprehensively capture and fuse different types of features, and use the output of wear degree as the feature input of life prediction result, so as to more effectively transmit feature information and ensure the validity of output results.

[0017] Preferably, making an early warning decision based on the acquired wear value, wear degree and life prediction result includes: Different levels of warning rules are set, and each level of warning rule corresponds to the wear value reaching a preset wear value, the tool wear degree reaching a preset degree, or the tool remaining life being within a preset remaining life threshold range; The urgency of the production plan of the CNC lathe is obtained, and the set warning rules are adjusted accordingly according to different urgency levels. The adjusted warning rules include: among the warning rules of the same level, the higher the urgency of the production plan, the higher the tool wear degree reaches a preset degree or the tool wear value reaches a preset wear value or the preset threshold value of the tool remaining life is lower; According to the early warning rules, the early warning level corresponding to the acquired wear value, wear degree and life prediction result is determined to realize early warning decision.

[0018] Through the above scheme, the warning threshold is dynamically adjusted according to the urgency of the CNC lathe production plan, and then the warning rules are refined and the warning decision is completed according to the refined warning rules.

[0019] Preferably, it also includes: While real-time monitoring of CNC lathe tool operation process data, synchronously collect CNC lathe tool operation process images; The acquired wear value, wear degree and life prediction results are statistically obtained, and the wear slope-operating time curve or life change rate-operating time curve corresponding to different wear stages are drawn in real time in chronological order; Real-time query whether the wear slope-operation time curve does not conform to the change trend of the wear stage or whether the life change rate-operation time curve does not conform to the change trend of the wear stage. If not, the synchronously acquired CNC lathe tool operation process image is transmitted to the cloud platform, and by querying the tool operation process data or analyzing the tool operation image, it is determined whether a preset event occurs, and the preset event includes a tool maintenance event or a processing material no-load event; if it exists, no processing is performed; if it does not exist, it is determined that the prediction result of the tool wear prediction model is to be verified; The image recognition technology of the cloud platform is used to analyze the images of the CNC lathe tool operation process to determine the tool wear value, tool wear degree and tool remaining life to verify the accuracy of the prediction results of the tool wear prediction model.

[0020] Through the above scheme, a time series analysis of wear changes and life changes is carried out, and abnormal trends are discovered in a timely manner in combination with the investigation of preset events. Based on the images of the tool operation process and using image recognition technology, the actual wear condition of the current tool is confirmed, thereby improving the accuracy and credibility of the prediction results.

[0021] Preferably, it also includes: Real-time monitoring and acquisition of running process data of tools from the same batch, where tools from the same batch refer to tools that use the same tool type to process the same material under the same processing conditions; Obtain the tool wear prediction output corresponding to each tool in the same batch of tools, and obtain the user's satisfaction with the tool wear prediction output corresponding to each tool in the same batch of tools, screen and retain the prediction outputs of several tools whose prediction output satisfaction is greater than the preset satisfaction; obtain the tool wear prediction output of each tool in the retained tools in the same batch respectively and construct the corresponding wear curve, and use the dynamic time warping algorithm to calculate the similarity of the wear curves of the prediction outputs of the tools in the same batch; The features corresponding to the tools in the same batch whose wear curve similarity is greater than the preset similarity of the wear curve are extracted to construct a feature matrix, calculate the covariance matrix, and screen the principal component features whose variance contribution rate is greater than the preset contribution rate through eigenvalue decomposition. The weight of each eigenvalue is dynamically allocated according to the contribution rate and fed back to the tool wear prediction model to adjust the feature weight ratio in the tool wear prediction model at the next moment.

[0022] Through the above scheme, a batch tool association feedback mechanism and a user satisfaction feedback mechanism are introduced to complete the dynamic adjustment of each feature weight parameter in the tool wear prediction model, thereby assisting in obtaining a prediction output with high prediction accuracy and meeting user needs.

[0023] Preferably, the process of generating a pre-built tool wear prediction model that matches the tool type, processing material, and processing conditions includes: According to a single tool type, a single processing material, a single or combined processing condition, a number of tool wear prediction models to be trained are designed accordingly, and the number of tool wear prediction models to be trained is determined to be N+1 according to the number of tools in the same batch N, and the feature weight parameters in each tool wear prediction model to be trained are distinguished in advance; One of the tool wear prediction models to be trained is used as a global model, and each of the remaining tool wear prediction models to be trained is used as a sub-model and pre-matched with a tool in the same batch of tools; The historical optimal feature set extracted from the historical operation data of each tool in the same batch of tools is used as the corresponding input sub-model to complete the matching sub-model training; after the training is completed, the feature weights and gradients of each sub-model are uploaded to the global model, and combined with the federated learning algorithm, the feature weights of each sub-model are aggregated to obtain the aggregated feature weights, and the aggregated feature weights are sent to each sub-model for iterative updates until the global model converges, and the converged global model is obtained as the trained tool wear prediction model to generate the corresponding pre-built tool wear prediction model under the conditions of a single tool type, a single processing material, and a single or combined processing condition.

[0024] Through the above scheme, in order to more accurately obtain the tool wear prediction results under the conditions of a single tool type, a single processing material, and a single or combined processing condition, combined with the federated learning algorithm, multiple models with different feature weight parameters are designed to match the same batch of tools to complete the acquisition of the global optimal tool prediction model and improve the prediction accuracy.

[0025] In a second aspect, the present application provides a tool wear monitoring system for a CNC lathe, comprising: Operation data acquisition module, used to monitor the operation process data of CNC lathe tools in real time; The operation data processing module is used to match the preset operation parameter threshold according to the tool type, processing material, and processing conditions. When the real-time operation parameter is detected to exceed the preset operation parameter threshold, the operation process data of the CNC lathe tool at the current moment is transmitted to the edge computing node; The optimal feature set acquisition module is used to use the edge computing node to match the corresponding preset operating parameter combination according to the tool type, processing material, and processing condition, and complete the operating parameter screening according to the matched preset operating parameter combination; based on the screened operating parameters, the feature selection algorithm is used to select and extract key features, obtain the optimal feature set and upload it to the cloud platform together with the corresponding tool type, processing material, and processing condition; The wear and life prediction module is used on the cloud platform to match the corresponding pre-built tool wear prediction model according to the tool type, processing material, and processing conditions, input the optimal feature set into the matching tool wear prediction model, and obtain the wear value, wear degree, and life prediction results; The early warning decision and prompt generation module is used to combine the production plan, make early warning decisions and generate early warning prompts based on the acquired wear values, wear degrees and life prediction results.

[0026] By adopting the above solution, combined with edge computing technology and cloud platform technology, adaptive screening of operating parameters under different working conditions and acquisition of the optimal feature set can be completed, thus realizing real-time and accurate prediction of the tool wear status of CNC lathes.

[0027] In summary, this application has the following beneficial effects: 1. Real-time collection of tool operation process data and intelligent judgment based on preset operation parameter thresholds ensures rapid response even under complex working conditions; by combining edge computing nodes with cloud platforms, adaptive screening from operation parameters to operation parameter features is achieved to obtain the optimal feature set, and adaptively match the tool wear prediction model to complete real-time, efficient and accurate prediction; 2. Through a multi-level early warning mechanism, the early warning rules are dynamically adjusted according to the urgency of the production plan, and early warning decisions are generated based on the predicted wear value, wear degree and remaining life, thus improving the stability and safety of production; 3. Through time series analysis, the changing trend of tool wear is observed. When identifying abnormal trends, images of the CNC lathe tool operation process are collected synchronously. The accuracy of the tool wear prediction model is verified by analyzing the images, which enhances the credibility and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of the method for monitoring tool wear of a CNC lathe described in a specific embodiment; Figure 2 It is a structural schematic diagram of the tool wear monitoring system for a CNC lathe in a specific embodiment. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0030] like Figure 1 As shown, the embodiment of the present application discloses a method for monitoring tool wear of a CNC lathe, the main steps of which include: S1. Real-time monitoring of CNC lathe tool operation process data.

[0031] Specifically, the CNC lathe tool operation process data is monitored in real time, including: tool type and usage time, processing materials, processing conditions and operating parameters, etc.

[0032] Among them, the tool type during the operation process can be determined by the recorded tool model and material, such as: drill, turning tool and milling cutter, etc., and the processing material during the operation process can be determined by the recorded processing material, such as: high-speed steel, cemented carbide and ceramic materials, etc.; the processing conditions of the lathe tools can be collected in real time through the installed sensor network, such as: finishing, semi-finishing and roughing divided according to the processing parameters such as the feed speed and cutting depth of the tool; the operating parameters of the lathe tools can be collected in real time through the installed sensor network, including: current data, temperature data, vibration signals, sound signals, cutting force signals and other parameters that reflect tool wear.

[0033] In addition, in order to further ensure the accuracy of the monitoring results, while real-time monitoring of the CNC lathe tool operation process data, a high-precision camera device is used to simultaneously capture the CNC lathe tool operation process images.

[0034] S2. Matching preset operation parameter thresholds according to real-time monitoring of CNC lathe tool operation process data.

[0035] Considering that the simultaneous real-time monitoring of the CNC lathe tool operation process data for processing, wear degree and life monitoring requires huge computing power, while the computing resources on the CNC lathe end are limited, therefore, through actual production experience and actual data, under the combination of specific tool materials, specific processing materials, specific processing conditions and other conditions, the operating parameters that may lead to abnormal conditions such as more serious or even excessive wear are obtained, and the corresponding preset operating parameter thresholds are set to judge the operating conditions with more serious or even excessive wear risks.

[0036] For example, for a milling cutter, when processing aluminum alloy materials, if the cutting force exceeds 100N, it is considered that there may be a risk of severe wear, and a preset threshold value of the cutting force is set accordingly.

[0037] S3. Based on the matched preset operating parameter threshold, the operating process data corresponding to the operating parameters exceeding the preset operating parameter threshold is transmitted to the edge computing node.

[0038] Specifically, when it is monitored that the real-time operating parameters exceed the preset operating parameter threshold, that is, any type of operating parameter in the real-time operating parameters exceeds the preset operating parameter threshold of the corresponding type, such as temperature or vibration frequency or cutting force, the CNC lathe tool operation process data at the current moment will be transmitted to the edge computing node.

[0039] S4. Use the edge computing node to filter the acquired CNC lathe tool operation process data, obtain the optimal feature set and upload it to the cloud platform.

[0040] Specifically, in order to reduce the burden of data transmission, edge computing nodes are used to preliminarily process and filter the operation process data so that only key data points are transmitted, improving data processing efficiency and achieving timely prediction; the specific steps include: Considering different types of tools, cutting different processing materials under different processing conditions, and other specific tool operation scenarios, different operating parameters have different degrees of importance in determining the wear state. For example, the cutting force and vibration data of the tool in the rough processing stage are often more important, while the temperature and sound signals of the tool in the corresponding fine processing stage are more concerned. Therefore, based on the actual expert production experience and actual data, the best operating parameter combination that can reflect the tool wear condition in different tool operation scenarios can be obtained, and then the preset operating parameter combination suitable for the combination of specific tool materials, specific processing materials, and specific processing conditions can be set and stored in each edge computing node. Using the edge computing node, the corresponding preset operating parameter combination is matched according to the tool type, processing material, and processing condition, and the operating parameter screening is completed according to the matched preset operating parameter combination.

[0041] After screening and obtaining the operating parameters, data processing can be performed on the determined operating parameters, including data synchronization, data filtering, data cleaning, and data compensation, etc., and the same sampling interpolation or dimensionality reduction processing can be performed on heterogeneous data.

[0042] After data processing of the operating parameters, in order to reduce the burden of data transmission and ensure the subsequent accurate and timely completion of model prediction, the edge computing nodes are continuously used on the basis of the screened and processed operating data to extract physical features that are strongly correlated with tool wear and construct a high-information data set, i.e., the optimal feature set. Specifically, the feature selection algorithm can be used to select and extract key features, including: feature extraction of operating parameters, including: extraction of time domain features (RMS, peak value, pulse factor, etc.), frequency domain features (FFT spectrum centroid, wavelet packet energy entropy, etc.) and time-frequency analysis (time-frequency spectrum diagram generated by STFT, etc.) of vibration signals / acoustic emission signals; extraction of statistical features (such as mean, variance, peak-to-peak value, etc.) of current signals and cutting force signals; use of principal component analysis, variance threshold algorithm, mutual information algorithm, or random forest feature importance ranking and other feature selection algorithms to screen features, retain key features, and generate the optimal feature set.

[0043] Finally, the optimal feature set and its corresponding tool type, processing material, processing conditions and other operating process data are uploaded to the cloud platform.

[0044] S5. On the cloud platform, the pre-built tool wear prediction model is matched according to the tool type, processing material, and processing conditions, and the optimal feature set is input into the matching tool wear prediction model to obtain the wear value, wear degree, and life prediction results.

[0045] Specifically, considering that there are differences in the wear state of the tool under specific operating scenarios such as specific tool materials, specific processing materials, specific processing conditions and other combination conditions, in order to more accurately predict the adaptive wear results for different operating scenarios, a tool wear prediction model that matches the combination conditions such as specific tool materials, specific processing materials, specific processing conditions and other combination conditions can be pre-constructed and built into the cloud platform, so as to utilize the cloud platform to support the storage and application of multiple different tool wear prediction models; among them, each pre-constructed tool wear prediction model can adopt a variety of machine learning algorithms, such as support vector machines (SVM), neural networks, etc., and is generated by training the optimal feature set obtained through a large amount of historical data as well as historical actual wear values, wear degree, and remaining life.

[0046] On the cloud platform, the pre-built tool wear prediction model is matched according to the tool type, processing material, and processing conditions. The optimal feature set is input into the matching tool wear prediction model to obtain the wear value, wear degree, and life prediction results.

[0047] S6. Make early warning decisions and generate early warning prompts based on the acquired wear values, wear degrees and life prediction results.

[0048] Specifically, based on the acquired wear degree and life prediction results, it can be determined whether the tool needs to be replaced in order to issue a corresponding warning and generate warning prompt information; in order to realize multiple protection mechanisms, different levels of warning rules are set, such as moderate wear and severe wear, which correspond to different warning levels, and reasonable warning decisions are set according to different warning levels, such as: different color warning signs, which are transmitted and displayed on the display terminal of the CNC machine tool.

[0049] By adopting the method described in the above embodiment, the operation process data of the CNC lathe tool is collected in real time, and efficient, accurate and real-time tool wear monitoring is achieved through preset operation parameter threshold matching, edge computing node processing and cloud platform prediction.

[0050] In a specific embodiment, in order to more accurately obtain the preset operating parameter thresholds under complex working conditions and ensure accurate wear prediction, a deep learning algorithm is used to establish a mapping relationship between operating parameters and tool wear values ​​to form an empirical model, thereby better assisting in matching and obtaining the preset operating parameter thresholds. The method specifically includes: matching the corresponding preset operating parameter thresholds according to tool type, processing material, and processing conditions: Collect historical operating parameters and tool wear values ​​corresponding to the tool operation process under different tool types, different processing materials, and different processing conditions, such as the historical operating parameters and tool wear values ​​of the milling cutter during the finishing operation of cemented carbide materials; cluster the collected historical data to construct a class of training data based on historical data such as operating parameters and tool wear values ​​corresponding to specific processing conditions such as a single tool, a single processing material, and a single or combined processing condition, and use a deep learning algorithm to obtain the mapping relationship between the operating parameters and wear values ​​of the tool during operation under the conditions of a single tool, a single processing material, and a single or combined processing condition, that is, obtain the mapping relationship between the operating parameters and the wear values ​​as the tool operation time increases through training.

[0051] According to the constructed mapping relationship, the threshold ranges of the operating parameters in different wear stages under known tool types, known processing materials and known processing conditions are analyzed and obtained; wherein, the different wear stages are mainly determined by fitting the operating time and wear value of the tool during operation under the historical single tool, single processing material and single or combined processing conditions, respectively, and calculating and generating the correlation curve generated by the operating time and wear value of the tool during operation, such as: the wear value within the first preset wear value range is identified as a light wear stage, the wear value within the second preset wear value range is identified as a moderate wear stage, and the wear value within the third preset wear value range is identified as a heavy wear stage.

[0052] The preset operating parameter thresholds are determined according to the needs of the user. When the user's demand for tool wear monitoring is the first demand, that is, the user requires high operating performance for the worn tool, the operating parameter threshold range of the light wear stage is matched, and the preset operating parameter threshold is determined according to the threshold range of the operating parameters in the light wear stage; when the user's demand for tool wear monitoring is the second demand, that is, the user requires general operating performance for the worn tool, the operating parameter threshold range of the moderate wear stage is matched, and the preset operating parameter threshold is determined according to the threshold range of the operating parameters in the moderate wear stage; when the user's demand for tool wear monitoring is the third demand, that is, the user requires non-strict operating performance for the worn tool, the operating parameter threshold range of the severe wear stage is matched, and the preset operating parameter threshold is determined according to the threshold range of the operating parameters in the severe wear stage.

[0053] In addition, it is necessary to continuously conduct reinforcement learning on the determination of preset operating parameter thresholds to optimize the data transmission during the operation process. Specifically, the operating parameters and tool wear values ​​corresponding to specific processing conditions such as a single tool, a single processing material, and a single or combined processing condition are regularly supplemented as a supplementary set of training data, so as to train and optimize the mapping relationship between operating parameters and wear values ​​during the tool operation process corresponding to a single tool, a single processing material, and a single or combined processing condition. According to the optimized mapping relationship, the threshold range of the operating parameters at different wear stages under the conditions of known tool types, known processing materials, and known processing conditions is re-analyzed.

[0054] In a specific embodiment, in order to more accurately obtain the preset operating parameter combination under complex working conditions and ensure accurate wear prediction, a deep learning algorithm is used to obtain the most accurate corresponding operating parameter combination for predicting tool wear conditions under complex working conditions. The method specifically includes: matching the corresponding preset operating parameter combination according to tool type, processing material, and processing conditions. Traverse different tool types, processing materials and processing conditions, respectively count the operating parameters corresponding to the tool operation process under the historical single tool type, single processing material and single or combined processing conditions, and set the preset operating parameter combination generated based on different operating parameters; Combined with a single tool type, a single processing material, and a single or combined processing condition, the historical operating parameters and corresponding wear values ​​are screened out according to each set of preset operating parameter combinations, and the training set and the test set are divided. The accuracy of the prediction results of the tool wear prediction model generated by training the training set is obtained according to the test set, and the preset operating parameter combinations under the corresponding combination conditions are arranged in order of accuracy.

[0055] Sort the preset operating parameter combinations generated under the same combination conditions according to the tool type, processing material, and processing conditions, and select a set of preset operating parameter combinations corresponding to the tool wear prediction model with the highest accuracy as the matching preset operating parameter combination according to the preset operating parameter combination sorting; for example, the preset operating parameter combinations under the combination conditions of milling cutter-hard alloy-finishing are sorted as: (A, B, D)\(A, B), (B, C, D)\(B, D), and select (A, B, D) as the matching preset operating parameter combination; When selecting actual operating parameters, there are some operating parameters in the currently matched preset operating parameter combination that are missing in the real-time operating parameters. For example, if the actual operating parameters are (B, D), the preset operating parameter combinations are sorted, and the preset operating parameter combination with the next priority and a precision greater than the preset precision is selected as the re-matched preset operating parameter combination; for example, if the preset operating parameter combination (A, B) with the next priority that meets the precision requirement is continued to be selected, and operating parameters are still missing, the preset operating parameter combination (B, D) that meets the actual operating parameters is continued to be queried, but the precision of this combination is lower than the preset precision requirement, that is, if the preset operating parameter combination with a precision greater than the preset precision cannot be queried according to the priority, the preset operating parameter combination is determined by data compensation.

[0056] Specifically, considering the accuracy of the prediction results, data compensation cannot be simply completed randomly. It is necessary to start from the perspective of actual operating parameters and complete the construction of the mapping relationship between different operating parameters according to the correlation between the operating parameters. For example, according to the mapping relationship between operating parameters B (cutting force) and A (acoustic emission signal), the compensation of operating parameter A (acoustic emission signal) can be completed, and the compensation of operating parameter C can be completed accordingly. The mapping supplement accuracy of the missing operating parameters of the real-time operating parameters is determined according to the strength of the correlation between the operating parameters, and the mapping supplement accuracy score of the missing operating parameters of the real-time operating parameters is obtained accordingly. The stronger the correlation, the stronger the accuracy; and obtain the similarity calculation between the preset operating parameter combination for completing the compensation of the missing operating parameters and the real-time operating parameters, match the similarity score according to the calculation result, and obtain the comprehensive score by weighted calculation (such as: (A, B, D) score is less than (B, C, D)), so as to determine the preset operating parameter combination with the highest score as the final matching preset operating parameter combination, so as to complete the matching of the corresponding preset operating parameter combination according to the tool type, processing material, and processing condition and select the corresponding operating parameters.

[0057] In a specific embodiment, in order to further improve the selection accuracy of key features, the method uses a feature selection algorithm to select and extract key features based on the screened operating parameters to obtain the optimal feature set, including: Considering that when the real-time operating parameters exceed the threshold range of the operating parameters in the light wear stage or the moderate wear stage, the preset operating parameter threshold is determined accordingly, it is highly likely that the current operating tool wear is in the light or moderate wear degree, and the possibility of replacing the tool is small. There is a certain prediction fault tolerance rate, and redundant and noise features are effectively removed, and only the most important features are retained. Specifically, when the preset operating parameter threshold is determined according to the threshold range of the operating parameters in the light wear stage or the moderate wear stage, the corresponding feature selection algorithm adopts an integrated feature selection algorithm; the integrated feature selection algorithm includes: pre-using a variance threshold or a mutual information algorithm to complete the preliminary screening of the target features, and then using a random forest feature importance ranking technology to complete the secondary screening of the features; after using the integrated feature selection algorithm to select and extract key features, according to the first preset number of features matching the light wear stage, the same number of features are retained according to the sorting of the screened features; Correspondingly, considering that the real-time operating parameters exceed the threshold range of the operating parameters in the severe wear stage and the corresponding preset operating parameter threshold is determined, it is highly likely that the current operating tool wear is in a severe wear degree, and the possibility of replacing the tool is high. The prediction fault tolerance is low, and an appropriate number of features need to be retained. Specifically, when the preset operating parameter threshold is determined according to the threshold range of the operating parameters in the severe wear stage, the corresponding feature selection algorithm adopts a single feature selection algorithm; the single feature selection algorithm includes: using a variance threshold or a mutual information algorithm to screen and sort the importance of the target features; after using the feature selection algorithm to select key features, according to the second preset number of features that match the severe wear stage, the same number of features are retained according to the sorting of the screened features.

[0058] The number of first preset features matching the light wear stage or the medium wear stage is less than the number of second preset features matching the heavy wear stage.

[0059] In a specific embodiment, in order to ensure that the designed tool wear prediction model can more comprehensively capture and integrate different types of features, thereby improving the prediction accuracy of tool wear degree and life, the method The tool wear prediction model pre-built according to the tool type, processing material, and processing conditions includes: The tool wear prediction model includes multiple prediction models, each of which corresponds to a single tool type, a single processing material, and a single or combined processing condition and is designed with: a multi-branch feature input layer, a multi-branch feature fusion layer, and a multi-branch prediction result output layer. In order to further ensure the consistency between the tool wear value, the tool wear degree, and the remaining life prediction and the accuracy of the remaining life prediction, a branch with a wear value output and an output corresponding to the wear degree branch are set as inputs to the branch with a life prediction result output, and each prediction model is generated through a set of historical optimal feature sets, tool wear values, tool wear degrees, and tool remaining life training under a single tool type, a single processing material, and a single or combined processing condition.

[0060] The multi-branch feature fusion layer is designed with multiple sub-branch feature fusion layers. The first sub-branch feature fusion layer performs weighted fusion on each feature input branch in the optimal feature set; wherein the weight quantity ratio is determined according to the importance ranking and quantity in the optimal feature set; the second sub-branch feature fusion layer is designed to respectively fuse each feature input branch in the optimal feature set with the wear value branch and the wear degree branch output, and then perform weighted fusion on the fused content.

[0061] In addition, in order to further optimize the tool wear prediction value, the method also includes: adjusting the multi-branch feature input layer to include the multi-branch feature input layer at the previous moment, the multi-branch feature input layer at the next moment, and the multi-branch feature input layer at the current moment; adjusting the first sub-branch feature fusion layer and the second sub-branch feature fusion layer in the multi-branch fusion layer, both of which complete the feature fusion by interval cross-fusion, specifically including: during actual feature fusion, the input features of the multi-branch feature input layer at the previous moment are fused with the same type of input features of the multi-branch feature input layer at the current moment to complete the feature fusion at a certain moment, and the multi-branch feature input layer at the next moment is fused with the same type of input features of the multi-branch feature input layer at the current moment to complete the feature fusion at the next moment.

[0062] The method also includes: obtaining the user's satisfaction with the wear degree and life prediction results, and if the user's satisfaction with the wear degree and life prediction results is lower than the preset satisfaction level, it is determined that the pre-constructed tool wear prediction model corresponding to the tool type, processing material, and processing condition matching needs to be optimized and trained, and an incremental learning method is used to perform incremental training on the pre-constructed tool wear prediction model corresponding to the tool type, processing material, and processing condition matching, and directly subsequently obtaining the user's satisfaction with the wear degree and life prediction results is lower than the preset satisfaction level.

[0063] In a specific embodiment, in order to further ensure the accuracy of the early warning, combined with the existing production plan, to better avoid premature replacement of tools with a remaining service life or processing accidents caused by failure to timely discover serious wear, the method of making early warning decisions based on the obtained wear degree and life prediction results also includes: For the set warning rules of different levels, each level of warning rule corresponds to the tool wear value reaching the preset value of the tool wear value or the tool wear reaching the preset degree or the remaining life of the tool being within the preset threshold range of the remaining life; such as: the first level warning corresponds to the tool wear value reaching the first preset value of the tool wear value, the tool wear degree reaching the first preset degree (mild wear), and the remaining life of the tool being lower than the first preset threshold of the remaining life (50% of the life) and greater than the second preset threshold of the remaining life (30% of the life); the second level warning corresponds to the tool wear value reaching the second preset value of the tool wear value, the tool wear degree reaching the second preset degree (moderate wear), and the remaining life of the tool being lower than the second preset threshold of the remaining life (30% of the life) and greater than the third preset threshold of the remaining life (10% of the life); the third level warning corresponds to the tool wear value reaching the third preset value of the tool wear value, the tool wear degree reaching the third preset degree (severe wear), and the remaining life of the tool being lower than the third preset threshold of the remaining life (10% of the life) but greater than 0.

[0064] The urgency of the CNC lathe production plan is obtained, and the warning rules set are adjusted according to different urgency levels. The adjusted warning rules include: in the warning rules of the same level, the higher the urgency of the production plan, the higher the corresponding tool wear value reaches the preset value of the tool wear value, the higher the tool wear degree reaches the preset degree, and the lower the preset threshold of the tool remaining life; for example: the general urgency level maintains the original warning rule setting, and the serious urgency level corresponds to the adjustment of the corresponding level of warning rules, such as: the first level warning corresponds to the tool wear value reaching the second preset value of the tool wear value, the tool wear degree reaches the second preset degree (medium wear), and the tool remaining life is low The second level warning corresponds to the tool wear value reaching the third preset value of the tool wear value, the tool wear degree reaches the third preset degree (severe wear), and the tool remaining life is lower than the third preset threshold of the remaining life (10% of the life) and greater than the fourth preset threshold of the remaining life (5% of the life); the third level warning corresponds to the tool wear value reaching the fourth preset value of the tool wear value, the tool wear degree reaches the fourth preset degree (super severe wear), and the tool remaining life is lower than the fourth preset threshold of the remaining life (5% of the life) but greater than 0.

[0065] According to the early warning rules, the early warning level corresponding to the acquired wear value, wear degree and life prediction result is determined to realize early warning decision.

[0066] In a specific embodiment, in order to further ensure the accuracy of tool wear prediction, in addition to using the operating parameters collected by the sensor, the image data of the operating process is also used to analyze and verify the accuracy of tool wear. The method also includes: While real-time monitoring of CNC lathe tool operation process data, synchronously collect CNC lathe tool operation process images; The wear value, wear degree and life prediction results are obtained statistically, and the wear slope-operating time curve or life change rate-operating time curve corresponding to different wear stages are drawn in real time in chronological order.

[0067] According to the law of tool wear, as the running time increases, in different wear stages, the wear slope gradually decreases in the light wear stage, the wear slope changes to positive and negative values ​​in the moderate wear stage, and the wear slope gradually increases in the severe wear stage. The wear slope-running time curve is queried in real time. Once a change trend that does not conform to the wear stage is found, it is determined that there may be a wear prediction error, and the synchronously acquired CNC lathe tool running process image is transmitted to the cloud platform; similarly, according to the law of tool wear, the life of the tool is constantly decreasing. In different wear stages, the tool life change gradually decreases in the light wear stage, the tool life change gradually flattens in the moderate wear stage, and the tool life change gradually increases in the severe wear stage. The life change rate-running time curve is queried in real time. Once a change trend that does not conform to the wear stage is found, it is determined that there may be a wear prediction error; it is determined that there may be a wear prediction error, and the synchronously acquired CNC lathe tool running process image is transmitted to the cloud platform.

[0068] At the same time, considering that there may be some events that cause irregular changes, such as: maintenance of the tool, or operating errors in the tool production line that cause the processing material to be empty, resulting in the tool not actually cutting, or the tool operation parameter assignment error, etc., these phenomena are checked first, and these events are pre-set as preset events. By querying the tool operation process data and or analyzing the tool operation image data, it is determined whether the preset event occurs. If it does, it indicates that the abnormal trend is caused by the preset event and no processing is performed. If it does not exist, it is determined that the prediction result of the tool wear prediction model is to be verified; The image recognition technology of the cloud platform is used to analyze the images of the CNC lathe tool operation process, determine the tool wear degree and life prediction, and verify the accuracy of the prediction results of the tool wear prediction model.

[0069] In a specific embodiment, in order to further improve the accuracy of tool wear, a user feedback mechanism and a batch feedback mechanism are designed to optimize the weight parameters of each feature in the wear prediction model to improve the prediction accuracy. The method also includes: Considering the processing similarities of tools from the same batch and the differences of the processing sensor network in collecting the operation process data of tools from the same batch, a batch feedback mechanism for tools from the same batch is designed to adjust the feature weights, and monitor and obtain the operation process data of tools from the same batch in real time; wherein, tools from the same batch refer to tools that use the same tool type to process the same processing material under the same processing conditions.

[0070] Correspondingly, the above-mentioned optimal feature set of edge nodes is obtained and respectively input into the tool wear prediction model for prediction, and the corresponding tool wear prediction output for each tool in the same batch of tools is obtained; in addition to considering the feedback information of the same batch of tools, the user's satisfaction with the prediction results is also considered, and the tool prediction results with high user satisfaction are selected as much as possible for weight adjustment. Therefore, the user's satisfaction with the tool wear prediction output corresponding to each tool in the same batch of tools is obtained, and the prediction outputs of several tools whose prediction output satisfaction is greater than the preset satisfaction are screened and retained.

[0071] The tool wear prediction output of each tool in the retained tools of the same batch is obtained respectively and the wear curve is constructed accordingly. The dynamic time warping algorithm is used to calculate the similarity of the wear curves predicted by the tools of the same batch. The tool wear prediction output corresponding to the tools with lower similarity is eliminated by similarity, and the extracted features corresponding to the tools of the same batch whose similarity of wear curves is greater than the preset similarity of wear curves are obtained to construct a feature matrix, calculate the covariance matrix, and select the principal component features with a variance contribution rate greater than the preset contribution rate through eigenvalue decomposition, such as the principal component features with a contribution rate greater than 85%, and dynamically allocate the weight of each eigenvalue according to the contribution rate, and feed it back to the tool wear prediction model to adjust the weight ratio of each feature in the tool wear prediction model at the next moment; wherein, the dynamic weight allocation formula is: In the formula, is the eigenvalue weight of the i-th principal component, is the eigenvalue of the i-th principal component.

[0072] In addition, considering the similarity of processing conditions of tools in the same batch, and the differences in the processing sensor network in collecting the data of the same batch of tools, the wear prediction model suitable for all tools in the same batch is trained and constructed by using the difference data between different tools under the same processing conditions (maintaining the adaptability of processing conditions) (considering the possible differences in the collection of actual data). Specifically, the generation process of the pre-built tool wear prediction model corresponding to the tool type, processing material, and processing conditions includes: For a single tool type, a single processing material, and a single or combined processing condition, several tool wear prediction models to be trained are designed accordingly, and the number of tool wear prediction models to be trained is determined to be N+1 based on the number of tools in the same batch N, and the feature weight parameters in each tool wear prediction model to be trained are pre-distinguished; that is, for N tool wear prediction models designed with the same structure, different feature weight parameters presented in the model are adjusted respectively.

[0073] One of the tool wear prediction models to be trained is used as the global model, and each of the remaining tool wear prediction models to be trained is used as a sub-model and pre-matched with a tool in the same batch of tools, that is, each sub-model corresponds to a tool number.

[0074] The historical optimal feature set extracted from the historical operation data of each tool in the same batch of tools is input into the corresponding sub-model to complete the matching sub-model training; after the training is completed, the feature weight and gradient of each sub-model are uploaded to the global model, and combined with the federated learning algorithm, the feature weights of each sub-model are aggregated to obtain the aggregated feature weight. The specific aggregation formula includes: In the formula, is the aggregate feature weight, The aggregate feature weights of the previous iteration order, is the sub-model gradient.

[0075] The aggregated feature weights are sent down to each sub-model, and iterative updates are performed until the global model converges. The converged global model is obtained as the trained tool wear prediction model, and the corresponding pre-built tool wear prediction model is generated under the conditions of a single tool type, a single processing material, and a single or combined processing condition.

[0076] like Figure 2 As shown, the embodiment of the present application discloses a tool wear monitoring system for a CNC lathe, comprising: The operation data acquisition module 101 is used to monitor the operation process data of the CNC lathe tool in real time; The operation data processing module 102 is used to match the preset operation parameter threshold according to the tool type, processing material, and processing condition, and transmit the current CNC lathe tool operation process data to the edge computing node when the real-time operation parameter exceeds the preset operation parameter threshold; The optimal feature set acquisition module 103 is used to use the edge computing node to match the corresponding preset operating parameter combination according to the tool type, processing material, and processing condition, and complete the operation parameter screening according to the matched preset operating parameter combination; based on the screened operating parameters, the feature selection algorithm is used to select and extract key features, obtain the optimal feature set and upload it to the cloud platform together with the corresponding tool type, processing material, and processing condition; The wear and life prediction module 104 is used on the cloud platform to match the corresponding pre-built tool wear prediction model according to the tool type, processing material, and processing conditions, input the optimal feature set into the matching tool wear prediction model, and obtain the wear value, wear degree, and life prediction results; The warning decision and prompt generation module 105 is used to make warning decisions and generate warning prompts in combination with the production plan based on the acquired wear value, wear degree and life prediction results.

[0077] In a specific embodiment, the system further includes: The operation data acquisition module 101 is also used to synchronously acquire the operation process image of the CNC lathe tool while monitoring the operation process data of the CNC lathe tool in real time; The wear and life prediction verification module 106 is used to statistically obtain the wear value, wear degree and life prediction results, and to draw the wear slope-operation time curve or life change rate-operation time curve corresponding to different wear stages in real time in chronological order; to query in real time whether the wear slope-operation time curve does not conform to the change trend of the wear stage or the life change rate-operation time curve does not conform to the change trend of the wear stage; if not, the synchronously obtained CNC lathe tool operation process image is transmitted to the cloud platform, and by querying the tool operation process data or analyzing the tool operation image, it is determined whether a preset event occurs, and the preset event includes a tool maintenance event or a processing material no-load event; if it exists, it is not processed; if it does not exist, it is determined that the prediction result of the tool wear prediction model is to be verified; the image recognition technology of the cloud platform is used to analyze the CNC lathe tool operation process image to determine the tool wear value, tool wear degree and tool remaining life to verify the accuracy of the prediction result of the tool wear prediction model.

[0078] In a specific embodiment, the operation data acquisition module 101 in the system is also used to monitor and obtain the operation process data of the same batch of tools in real time, and the same batch of tools refers to tools that use the same tool type to process the same processing material under the same processing conditions; The wear and life prediction module 104 is also used to obtain the tool wear prediction output corresponding to each tool in the same batch of tools, and obtain the user's satisfaction with the tool wear prediction output corresponding to each tool in the same batch of tools, screen and retain the prediction output of several tools whose prediction output satisfaction is greater than a preset satisfaction; obtain the tool wear prediction output of each tool in the retained tools in the same batch respectively and construct a corresponding wear curve, and use a dynamic time warping algorithm to calculate the similarity of the wear curves of the prediction outputs of the tools in the same batch; obtain the extracted features corresponding to the tools in the same batch whose wear curve similarity is greater than the preset similarity of the wear curve to construct a feature matrix, calculate the covariance matrix, and screen the principal component features whose variance contribution rate is greater than the preset contribution rate through eigenvalue decomposition, and dynamically assign the weight of each eigenvalue according to the contribution rate and feed it back to the tool wear prediction model to adjust the feature weight ratio in the tool wear prediction model at the next moment.

[0079] The embodiment of the present application also discloses a computer-readable storage medium.

[0080] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed such as the above-mentioned CNC lathe tool wear monitoring method. The computer-readable storage medium includes, for example: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0081] The embodiment of the present application also discloses a computer device.

[0082] Specifically, the computer device includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute the above-mentioned CNC lathe tool wear monitoring method.

[0083] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. A method for monitoring tool wear of a CNC lathe, characterized in that: include: Real-time monitoring of CNC lathe tool operation process data, including: tool type, processing material, processing conditions and operating parameters; According to the tool type, processing material, and processing conditions, the preset operating parameter threshold is matched. When the real-time operating parameter is detected to exceed the preset operating parameter threshold, the current CNC lathe tool operation process data is transmitted to the edge computing node; Utilize edge computing nodes to match the corresponding preset operating parameter combinations according to tool type, processing material, and processing conditions, and complete operating parameter screening according to the matched preset operating parameter combinations; based on the screened operating parameters, use feature selection algorithms to select and extract key features, obtain the optimal feature set, and upload it to the cloud platform with the corresponding tool type, processing material, and processing conditions; On the cloud platform, the pre-built tool wear prediction model is matched according to the tool type, processing material, and processing conditions, and the optimal feature set is input into the matching tool wear prediction model to obtain the wear value, wear degree, and life prediction results; Based on the acquired wear values, wear degree and life prediction results, early warning decisions are made and early warning prompts are generated.

2. The CNC lathe tool wear monitoring method according to claim 1, characterized in that: The preset operating parameter thresholds corresponding to the tool type, processing material, and processing conditions are matched as follows: Collect the operating parameters and tool wear values ​​corresponding to the tool operation process under different tool types, different processing materials, and different processing conditions, and use the deep learning algorithm to obtain the mapping relationship between the operating parameters and wear values ​​of the tool during operation under the conditions of a single tool, a single processing material, and a single or combined processing condition; According to the constructed mapping relationship, the threshold range of the operating parameters in different wear stages under the known tool type, known processing material and known processing conditions is analyzed and obtained; the different wear stages are determined according to the correlation curve generated by calculating the operation time and the wear value during the operation of the tool; the different wear stages include light wear stage, medium wear stage and heavy wear stage; Receive the user's demand for tool wear monitoring, and match the preset threshold range of the operating parameters in the specific wear stage according to the user's demand for tool wear monitoring; determine the preset operating parameter threshold value according to the matched threshold range of the operating parameters in the specific wear stage, so as to complete the matching of the preset operating parameter threshold value according to the tool type, processing material and processing conditions collected in real time.

3. The CNC lathe tool wear monitoring method according to claim 2 is characterized in that: The preset operating parameter combinations corresponding to the tool type, processing material, and processing conditions include: Traverse different tool types, processing materials and processing conditions, respectively count the operating parameters corresponding to the tool operation process under the historical single tool type, single processing material and single or combined processing conditions, and set the preset operating parameter combination generated based on different operating parameters; According to each set of preset operating parameter combinations, historical operating parameters and corresponding wear values ​​are screened out and divided into training sets and test sets. According to the test set, the accuracy of the prediction results of the tool wear prediction model generated by training the training set is obtained, and the preset operating parameter combinations are arranged in order of accuracy to generate a ranking; The preset operating parameter combinations generated under the same combination conditions according to the tool type, processing material, and processing conditions are sorted, and a group of preset operating parameter combinations corresponding to the tool wear prediction model with the highest accuracy are used as the matching preset operating parameter combinations according to the preset operating parameter combinations sorting. When the actual operating parameters are selected, if some operating parameters in the currently matched preset operating parameter combination are missing in the real-time operating parameters, the preset operating parameter combinations are sorted, and the preset operating parameter combination with the next priority and an accuracy greater than the preset accuracy is selected as the re-matched preset operating parameter combination. If the preset operating parameter combination with an accuracy greater than the preset accuracy cannot be queried according to the priority, the preset operating parameter combination with the highest comprehensive score is selected from all the preset operating parameter combinations with an accuracy greater than the preset accuracy as the final matching preset operating parameter combination, thereby completing the matching of the corresponding preset operating parameter combinations according to the tool type, processing material, and processing conditions and selecting the corresponding operating parameters; the comprehensive score is obtained by weighted calculation based on the similarity score with the real-time operating parameters and the accuracy score of completing the real-time operating parameter mapping supplement based on the missing operating parameters.

4. The CNC lathe tool wear monitoring method according to claim 2, characterized in that: The screening-based operating parameters use a feature selection algorithm to select and extract key features to obtain the optimal feature set, including: When the preset operating parameter threshold is determined according to the threshold range of the operating parameter in the light wear stage or the moderate wear stage, the corresponding feature selection algorithm adopts an integrated feature selection algorithm; the integrated feature selection algorithm includes: preliminarily using a variance threshold or a mutual information algorithm to complete a preliminary screening of target features, and then using a random forest feature importance ranking technique to complete a secondary screening of features; after using the integrated feature selection algorithm to select and extract key features, according to the first preset number of features matching the light wear stage or the moderate wear stage, retaining the same number of features according to the sorting of the screened features; When the preset operating parameter threshold is determined according to the threshold range of the operating parameter in the severe wear stage, the corresponding feature selection algorithm adopts a single feature selection algorithm; the single feature selection algorithm includes: using a variance threshold or a mutual information algorithm to screen and sort the importance of the target feature; after using the feature selection algorithm to select key features, according to the second preset number of features matching the severe wear stage, the same number of features are retained according to the sorting of the screened features.

5. The CNC lathe tool wear monitoring method according to claim 1, characterized in that: The tool wear prediction model pre-built according to the tool type, processing material, and processing conditions includes: The tool wear prediction model includes multiple prediction models, each of which includes: a multi-branch feature input layer, a multi-branch feature fusion layer and a multi-branch prediction result output layer, wherein the output corresponding to the branch outputting the wear value and the branch outputting the wear degree is set as the input of the branch outputting the life prediction result, and is generated through a set of historical optimal feature sets, tool wear values, tool wear degrees and tool remaining life training under a single tool type, a single processing material and a single or combined processing condition.

6. The CNC lathe tool wear monitoring method according to claim 1, characterized in that: The early warning decision making based on the acquired wear value, wear degree and life prediction result includes: Different levels of warning rules are set, and each level of warning rule corresponds to the wear value reaching a preset wear value, the tool wear degree reaching a preset degree, or the tool remaining life being within a preset remaining life threshold range; The urgency of the production plan of the CNC lathe is obtained, and the set warning rules are adjusted accordingly according to different urgency levels. The adjusted warning rules include: among the warning rules of the same level, the higher the urgency of the production plan, the higher the tool wear degree reaches a preset degree or the tool wear value reaches a preset wear value or the preset threshold value of the tool remaining life is lower; According to the early warning rules, the early warning level corresponding to the acquired wear value, wear degree and life prediction result is determined to realize early warning decision.

7. The method for monitoring tool wear of a CNC lathe according to claim 1, characterized in that: Also includes: While real-time monitoring of CNC lathe tool operation process data, synchronously collect CNC lathe tool operation process images; The acquired wear value, wear degree and life prediction results are statistically obtained, and the wear slope-operating time curve or life change rate-operating time curve corresponding to different wear stages are drawn in real time in chronological order; Real-time query whether the wear slope-operation time curve does not conform to the change trend of the wear stage or whether the life change rate-operation time curve does not conform to the change trend of the wear stage. If not, the synchronously acquired CNC lathe tool operation process image is transmitted to the cloud platform, and by querying the tool operation process data or analyzing the tool operation image, it is determined whether a preset event occurs, and the preset event includes a tool maintenance event or a processing material no-load event; if it exists, no processing is performed; if it does not exist, it is determined that the prediction result of the tool wear prediction model is to be verified; The image recognition technology of the cloud platform is used to analyze the images of the CNC lathe tool operation process to determine the tool wear value, tool wear degree and tool remaining life to verify the accuracy of the prediction results of the tool wear prediction model.

8. The method for monitoring tool wear of a CNC lathe according to claim 1, characterized in that: Also includes: Real-time monitoring and acquisition of operation process data of tools from the same batch, where tools from the same batch refer to tools that use the same tool type to process the same material under the same processing conditions; Obtain the tool wear prediction output corresponding to each tool in the same batch of tools, and obtain the user's satisfaction with the tool wear prediction output corresponding to each tool in the same batch of tools, screen and retain the prediction outputs of several tools whose prediction output satisfaction is greater than the preset satisfaction; obtain the tool wear prediction output of each tool in the retained tools in the same batch respectively and construct the corresponding wear curve, and use the dynamic time warping algorithm to calculate the similarity of the wear curves of the prediction outputs of the tools in the same batch; The features corresponding to the tools in the same batch whose wear curve similarity is greater than the preset similarity of the wear curve are extracted to construct a feature matrix, calculate the covariance matrix, and screen the principal component features whose variance contribution rate is greater than the preset contribution rate through eigenvalue decomposition. The weight of each eigenvalue is dynamically allocated according to the contribution rate and fed back to the tool wear prediction model to adjust the feature weight ratio in the tool wear prediction model at the next moment.

9. The method for monitoring tool wear of a CNC lathe according to claim 8, characterized in that: The process of generating a pre-built tool wear prediction model that matches tool type, processing material, and processing conditions includes: According to a single tool type, a single processing material, a single or combined processing condition, a number of tool wear prediction models to be trained are designed accordingly, and the number of tool wear prediction models to be trained is determined to be N+1 according to the number of tools in the same batch N, and the feature weight parameters in each tool wear prediction model to be trained are distinguished in advance; One of the tool wear prediction models to be trained is used as a global model, and each of the remaining tool wear prediction models to be trained is used as a sub-model and pre-matched with a tool in the same batch of tools; The historical optimal feature set extracted from the historical operation data of each tool in the same batch of tools is used as the corresponding input sub-model to complete the matching sub-model training; after the training is completed, the feature weights and gradients of each sub-model are uploaded to the global model, and combined with the federated learning algorithm, the feature weights of each sub-model are aggregated to obtain the aggregated feature weights, and the aggregated feature weights are sent to each sub-model for iterative updates until the global model converges, and the converged global model is obtained as the trained tool wear prediction model to generate the corresponding pre-built tool wear prediction model under the conditions of a single tool type, a single processing material, and a single or combined processing condition.

10. A CNC lathe tool wear monitoring system, characterized in that: include: Operation data acquisition module, used to monitor the operation process data of CNC lathe tools in real time; The operation data processing module is used to match the preset operation parameter threshold according to the tool type, processing material, and processing conditions. When the real-time operation parameter is detected to exceed the preset operation parameter threshold, the operation process data of the CNC lathe tool at the current moment is transmitted to the edge computing node; The optimal feature set acquisition module is used to use the edge computing node to match the corresponding preset operating parameter combination according to the tool type, processing material, and processing condition, and complete the operating parameter screening according to the matched preset operating parameter combination; based on the screened operating parameters, the feature selection algorithm is used to select and extract key features, obtain the optimal feature set and upload it to the cloud platform together with the corresponding tool type, processing material, and processing condition; The wear and life prediction module is used on the cloud platform to match the corresponding pre-built tool wear prediction model according to the tool type, processing material, and processing conditions, input the optimal feature set into the matching tool wear prediction model, and obtain the wear value, wear degree, and life prediction results; The early warning decision and prompt generation module is used to combine the production plan, make early warning decisions and generate early warning prompts based on the acquired wear values, wear degrees and life prediction results.

Images