A numerical control machining tool wear intelligent prediction and compensation method

Abstract

The application provides a numerical control machining tool wear intelligent prediction and compensation method, and belongs to the technical field of industrial control. The method comprises the following steps: collecting cutting process data of a numerical control machining tool, and performing preprocessing to obtain standardized data; performing symbol conversion on the standardized data to obtain symbol sequence data; obtaining a tool wear state prediction value based on the symbol sequence through a tool wear state prediction model; determining a wear compensation strategy according to the tool wear state prediction value; and taking the wear compensation strategy as a control instruction and executing it through a numerical control system. Symbol dynamic systems and an adaptive symbol mapping mechanism are introduced, the tool wear process is regarded as a symbol dynamic system, and a symbol sequence is used to represent the evolution of the wear state, so that the problems of low prediction accuracy, poor model generalization ability and lack of flexibility of the compensation strategy in the existing numerical control machining tool wear prediction and compensation method are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial control, in particular to a numerical control machining tool wear intelligent prediction and compensation method. BACKGROUND

[0002] In the process of numerical control machining, tool wear is a complex and dynamic process, which is affected by multiple factors such as cutting force, temperature, vibration, etc. The interaction and change between these factors have high nonlinearity and uncertainty, which makes it difficult to accurately predict and timely compensate the tool wear state.

[0003] Traditional methods for tool wear prediction often use a single signal feature to predict the tool wear state based on an empirical model; however, relying solely on a single signal feature for wear prediction, such as determining the wear degree only through the change of cutting force signal. This method cannot fully capture the complex dynamic behavior in the wear process, and easily ignores the influence of other important factors such as temperature and vibration signal changes, resulting in insufficient reliability of the prediction results. And the prediction effect of the model based on experience is acceptable under certain conditions, but when the machining conditions change, such as replacing workpieces of different materials or adjusting cutting parameters, the generalization ability of the model is poor, and it cannot adapt to new wear laws. For example, the wear prediction model established when machining aluminum alloy may be completely invalid when machining stainless steel, and a new model needs to be established, increasing the workload and time cost. Traditional compensation strategies are often fixed, such as replacing or sharpening the tool only when the wear reaches a certain degree. This strategy cannot be flexibly adjusted according to the real-time changes of the wear state, and is prone to cause insufficient compensation or excessive compensation. For example, the cutting parameters are not adjusted in time to slow down the wear rate in the early stage of wear, and the tool is replaced in a hurry when the wear is serious, affecting the machining efficiency and product quality.

[0004] Therefore, a numerical control machining tool wear intelligent prediction and compensation method is needed. SUMMARY

[0005] Therefore, the present application provides a numerical control machining tool wear intelligent prediction and compensation method, which introduces a symbolic dynamic system and an adaptive symbol mapping mechanism, regards the tool wear process as a symbolic dynamic system, and uses a symbol sequence to represent the evolution of the wear state, solving the problems of low prediction accuracy, poor model generalization ability, and lack of flexibility in compensation strategy in the existing numerical control machining tool wear prediction and compensation method.

[0006] To this end, the present application provides the following technical solutions:

[0007] A numerical control machining tool wear intelligent prediction and compensation method, comprising:

[0008] Collect cutting process data of CNC machining tools and perform preprocessing to obtain standardized data;

[0009] Performing symbol conversion on the standardized data to obtain symbol sequence data;

[0010] The tool wear state prediction model is used to obtain the tool wear state prediction value based on the symbol sequence data;

[0011] Determine the wear compensation strategy based on the tool wear state prediction value;

[0012] The wear compensation strategy is used as a control instruction and executed by the CNC system.

[0013] Furthermore, it also includes:

[0014] The future wear trend of the tool is characterized by the symbolic transition probability matrix;

[0015] The trend compensation strategy is obtained by combining the tool wear state prediction value and future wear trend.

[0016] Furthermore, the cutting process data of the CNC machining tool includes:

[0017] Cutting force signal, temperature signal and vibration signal.

[0018] Furthermore, the performing symbol conversion on the standardized data to obtain symbol sequence data includes:

[0019] Calculate the mean and standard deviation of the standardized data as statistical features;

[0020] The symbol intervals are divided based on statistical characteristics, and symbols are assigned to the standardized data.

[0021] Furthermore, the determination of the wear compensation strategy according to the tool wear state prediction value includes:

[0022] If the tool wear state prediction value is light wear, adjust the cutting parameters;

[0023] If the tool wear state prediction value is moderate wear, the tool is locally sharpened;

[0024] If the tool wear state prediction value is severe wear, replace the tool;

[0025] If the tool wear status prediction value is abnormal, the processing is stopped.

[0026] Furthermore, the trend compensation strategy obtained by combining the tool wear state prediction value and the future wear trend includes:

[0027] If the tool wear state prediction value is light wear and the future wear trend is moderate wear, increase the adjustment range of cutting parameters;

[0028] If the tool wear state prediction value is moderate wear and the future wear trend is severe wear, then replace the tool;

[0029] If the tool wear state prediction value is light wear and the future wear trend is abnormal, the equipment should be repaired;

[0030] If the tool wear state prediction value is abnormal and the future wear trend is abnormal, the process will be stopped.

[0031] Furthermore, obtaining the tool wear state prediction value based on the symbol sequence data by using the tool wear state prediction model includes:

[0032] Pre-training the tool wear state prediction model;

[0033] The pre-trained tool wear state prediction model is used to input symbol sequence data and output the tool wear state prediction value.

[0034] Furthermore, the tool wear state prediction model is pre-trained, including:

[0035] The tool wear state prediction model is an unsupervised learning model;

[0036] Perform cluster analysis on the symbol sequence; annotate the clustering results and map each cluster to the corresponding wear pattern;

[0037] Optimize model parameters using cross-validation;

[0038] Minimize the difference between the input symbol sequence and the reconstructed symbol sequence.

[0039] Furthermore, dividing the symbol intervals based on the statistical characteristics and assigning symbols to the standardized data includes:

[0040]

[0041] Among them, F represents the cutting force signal, T represents the temperature signal; V represents the vibration signal; are the mean values ​​of cutting force signal, temperature signal and vibration signal respectively, S F 、S T 、S V are the standard deviations of cutting force signal, temperature signal and vibration signal respectively.

[0042] Advantages and positive effects of the present invention:

[0043] This invention models the tool wear process as a symbolic dynamic system, converting continuous wear states into discrete symbol sequences, thereby accurately capturing nonlinear wear dynamics. The adaptive symbol mapping mechanism dynamically optimizes symbol segmentation rules by sensing data features in real time, enabling the symbolization process to adapt to different working conditions and individual differences, significantly improving the robustness of state representation. Combined with spatiotemporal correlation analysis of trend prediction databases, the multi-scale laws of wear evolution are mined from historical symbol sequences, enhancing the early perception of sudden wear while reducing the prediction delay of progressive wear. This hybrid architecture, which integrates dynamic symbolization and adaptive learning, enables compensation strategies to both quickly respond to sudden anomalies and perform forward-looking optimization based on long-term trends, ultimately achieving synergistic improvements in prediction accuracy, real-time performance, and generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 This is a flow chart of the intelligent prediction and compensation method for CNC machining tool wear in Example 1 of the present invention;

[0046] Figure 2 This is a flow chart of the intelligent prediction and compensation method for CNC machining tool wear combined with the symbolic transfer probability matrix in this embodiment 2;

[0047] Figure 3 This is a flowchart of the compensation strategy formulated in this embodiment 1. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0049] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] The present invention provides a method for intelligent prediction and compensation of CNC machining tool wear, which collects the cutting process data of CNC machining tools in real time, cleans and normalizes the collected cutting process data, and converts the processed normalized data into symbol sequence data according to the adaptive symbol mapping rule; uses the symbol sequence data and its corresponding tool wear pattern definition rule training model to form a trained wear state prediction model; inputs the symbol sequence data generated in real time into the trained wear state prediction model, and combines the trend prediction database of the symbol sequence data to predict future wear trends in real time and predict the current tool wear state; formulates corresponding compensation strategies based on the predicted tool wear state, combined with the processing technology requirements and tool characteristics. Through the above method, the accuracy and response speed of the wear state prediction are improved, and a more intelligent and flexible compensation strategy formulation is achieved. Specifically, the present invention includes the following steps:

[0051] S1, data collection;

[0052] Receive cutting process data of CNC machining tools, including cutting force signals, temperature signals and vibration signals;

[0053] S2, data preprocessing and symbol conversion;

[0054] 1) Preprocess the received cutting process data to obtain standardized data;

[0055] Specifically, the preprocessing includes: using a low-pass filter to filter out high-frequency noise; and unifying data of different dimensions into the same numerical range.

[0056] 2) Converting the standardized data into symbol sequence data according to the adaptive symbol mapping rule, specifically,

[0057] For the standardized cutting force signal, temperature signal and vibration signal, their mean and standard deviation are calculated as statistical features.

[0058] Dynamically divide symbol intervals: Dynamically divide the symbol intervals of the cutting force signal, temperature signal, and vibration signal according to statistical characteristics. The cutting force signal is divided into three symbol intervals of "low", "medium", and "high", represented by symbols "0", "1", and "2" respectively. The temperature signal is divided into three symbol intervals of "low temperature", "medium temperature", and "high temperature", represented by symbols "3", "4", and "5" respectively. The vibration signal is divided into three symbol intervals of "low frequency", "medium frequency", and "high frequency", represented by symbols "6", "7", and "8" respectively.

[0059] Generate symbol sequence data: convert into symbol sequence data according to the adaptive symbol mapping rule.

[0060] The adaptive symbol mapping rule automatically adjusts the division of symbol intervals according to the changes in real-time data.

[0061] S3, obtaining a tool wear state prediction value based on the symbol sequence data using a tool wear state prediction model;

[0062] 1) Using symbol sequence data and its corresponding tool wear pattern to define rules and train the model, specifically,

[0063] Data preparation: Use unsupervised learning methods to perform cluster analysis on the symbol sequences, perform expert annotation on the clustering results, and map each cluster to a specific wear pattern;

[0064] Model selection and training: Select a suitable unsupervised learning model and use cross-validation to optimize model parameters to ensure the generalization ability of the model;

[0065] Training process: Minimize the input symbol sequence x and the reconstructed symbol sequence the differences between;

[0066] Model evaluation: Use the retained test set to evaluate the model performance to ensure that it can accurately predict the wear state corresponding to new symbol sequences.

[0067] 2) Inputting the real-time generated symbol sequence data into the trained wear state prediction model to predict the tool wear state; and using the symbol transition probability matrix to represent the future wear trend of the tool;

[0068] S4. Develop compensation strategies;

[0069] According to the predicted tool wear status and future wear trend combined with the machining process requirements and tool characteristics, the corresponding compensation strategy is formulated.

[0070] The method of the present invention is further described with specific examples:

[0071] S1. Data collection:

[0072] In this embodiment, a variety of high-precision sensors (cutting force, temperature, vibration) are installed at key locations of the CNC machining equipment to collect data in real time at a sampling frequency of 100 Hz.

[0073] A cutting force sensor is installed in the tool clamping area to monitor force changes during the cutting process in real time. A temperature sensor is placed near the tool to measure the temperature in the cutting zone. A vibration sensor is fixed near the tool spindle to capture tool vibration signals. The sensors are connected to the CNC system via a data acquisition card, collecting data in real time at a sampling rate of 100Hz. Ideally, when machining a complex part, the sensors continuously collect various signal data to form a complete data set for subsequent analysis.

[0074] S2. Data processing:

[0075] S21. Filter and denoise the data using a low-pass filter to remove high-frequency noise while retaining the main features of the cutting force, temperature, and vibration signals. Specifically, for the cutting force signal, high-frequency noise caused by equipment vibration is filtered out while retaining low-frequency components related to tool wear.

[0076] S22. Unify data of different dimensions into the same numerical range;

[0077] Normalization is performed on the data, bringing data of varying dimensions into the same numerical range to facilitate subsequent symbolic mapping and analysis. Specifically, data such as cutting force are normalized to the range [0, 1] to make them comparable with temperature and vibration data. This preprocessed data forms standardized data that more accurately reflects the actual changes during tool wear and provides high-quality input for generating symbolic sequence data.

[0078] S23. Calculate the mean and standard deviation of the standardized data as statistical features:

[0079] Cutting force:

[0080] temperature:

[0081] vibration:

[0082] Among them, F represents the cutting force signal, T represents the temperature signal; V represents the vibration signal; are the mean values ​​of cutting force signal, temperature signal and vibration signal respectively, S F 、S T 、S V are the standard deviations of cutting force signal, temperature signal and vibration signal respectively.

[0083] Step 2.4: Convert the preprocessed cutting force, temperature and vibration signals into symbol sequence data according to the adaptive symbol mapping rule.

[0084] Dynamically divide the symbol intervals: Dynamically divide the symbol intervals of the cutting force signal, temperature signal and vibration signal according to the statistical characteristics. The cutting force signal is divided into three symbol intervals of "low", "medium" and "high", which are represented by the symbols "0", "1" and "2" respectively. The temperature signal is divided into three symbol intervals of "low temperature", "medium temperature" and "high temperature", which are represented by the symbols "3", "4" and "5" respectively. The vibration signal is divided into three symbol intervals of "low frequency", "medium frequency" and "high frequency", which are represented by the symbols "6", "7" and "8" respectively.

[0085]

[0086] Moreover, the adaptive symbol mapping rule can automatically adjust the division of symbol intervals according to the changes in real-time data, so that the symbol sequence can more accurately reflect the actual changes in the wear state.

[0087] S25. Generate symbol sequence data: Convert to symbol sequence data according to the above-mentioned adaptive symbol mapping rules. For example, at a certain time point, if: the cutting force signal falls in the "low" symbol interval, it is marked as "0", the temperature signal falls in the "medium temperature" symbol interval, it is marked as "4", and the vibration signal falls in the "high frequency" symbol interval, it is marked as "8", then the symbol sequence of this time point is "048", that is, each time point consists of three symbols (cutting force, temperature, vibration). This symbolic representation not only simplifies the data structure, but also facilitates subsequent pattern recognition and trend prediction. The symbols of multiple time points are combined into a complete symbol sequence, such as "048047138...". The symbol sequence can effectively capture the dynamic change law in the wear process, has high information compression ability and anti-noise interference ability, and provides a basis for subsequent wear state prediction and compensation strategy formulation.

[0088] S3. Build a tool wear state prediction model and perform pre-training;

[0089] S31. Data preparation:

[0090] Use unsupervised learning methods to perform cluster analysis on symbol sequences and label the clustering results; define wear pattern classification, including: slight wear, moderate wear, severe wear, and abnormal conditions.

[0091] In combination with tool wear pattern definition rules, specific wear patterns are defined based on multiple signal combinations to ensure that all possible situations are covered, including:

[0092] Slight wear: The symbol combination is "046" and "147", characterized by low cutting force and relatively stable temperature and vibration.

[0093] Moderate wear: The symbol combination is "157", characterized by moderate cutting forces, increased temperature and vibration.

[0094] Severe wear: The symbol combination is "258", characterized by higher cutting forces, significantly increased temperature and vibration.

[0095] Abnormal status: The symbol combination is "168" or "249", characterized by abnormally high temperature or severe vibration, which may indicate tool failure or other problems.

[0096] Historical data is used during the process for pattern validation to ensure that the defined wear patterns accurately reflect actual wear conditions.

[0097] Unsupervised learning methods are used to cluster symbol sequences. Training is performed by minimizing the difference between the input symbol sequence and the reconstructed symbol sequence, optimizing model parameters. This ensures that the model can accurately predict tool wear status without explicit labels.

[0098] Symbol sequence clustering: In the absence of wear labels, the symbol sequence needs to be clustered first. The purpose of this step is to find natural groupings or patterns in the data through unsupervised learning methods, providing a basis for subsequent wear pattern annotation. The original symbol sequence is a symbol sequence collected by the sensor and generated through preprocessing and symbol mapping rules. The symbol sequence at each time point consists of three symbols (cutting force, temperature, vibration), such as "046", "147", etc. Suppose there is a symbol sequence: "046147258046", which represents the state change of the tool at different time points.

[0099] In this embodiment, the goal of K-means clustering is to assign symbol sequences to k clusters so that the symbol sequences within each cluster are as similar as possible, while the differences between different clusters are as large as possible. The specific K-means clustering process is as follows:

[0100] Initialization: Randomly select k initial center points.

[0101] Assignment: Assign each symbol sequence to the cluster with the nearest center point.

[0102] Update: Recalculate the center point of each cluster.

[0103] Repeat: Repeat the assignment and update steps until convergence, i.e. the clusters no longer change.

[0104] Labeling: After clustering, you'll find that certain symbol combinations frequently appear in the same cluster. This is because these symbol combinations share similar characteristics and likely correspond to similar wear states. For example, if multiple symbol sequences all contain "046" or "147" and they frequently appear in the same cluster, these symbol combinations will be labeled as slightly worn based on their characteristics. This is because, based on historical experience and professional knowledge, these symbol combinations often correspond to slightly worn states.

[0105] In this example, "046" indicates low cutting force, moderate temperature, and minimal vibration, which is typically indicative of minor wear. In this way, unlabeled data can be converted into labeled data, providing a foundation for subsequent supervised learning.

[0106] S32, constructing a tool wear state prediction model and performing pre-training;

[0107] Tool wear state prediction models include: deep learning-based autoencoders (Autoencoder), variational autoencoders (VAE), or clustering-based models.

[0108] Cross-validation was used to optimize model parameters and ensure the generalization ability of the model.

[0109] In this embodiment, the tool wear state prediction model is an autoencoder; an autoencoder is a neural network designed to reconstruct input data and consists of an encoder and a decoder, where the encoder maps the input symbol sequence x to a low-dimensional representation (hidden layer) z, and the decoder reconstructs the low-dimensional representation z back to the original symbol sequence.

[0110] Training process: The goal of the autoencoder is to minimize the input symbol sequence x and the reconstructed symbol sequence The difference between them. The loss function usually uses the mean square error (MSE):

[0111]

[0112] Among them, x i is the original symbol sequence, is the reconstructed symbol sequence, and N is the number of samples.

[0113] For example, suppose there is an original symbol sequence "046". After encoding and decoding by the autoencoder, the reconstructed symbol sequence is "046". If the reconstruction result is very close to the original sequence, it means that the model is able to capture the characteristics of the symbol sequence well.

[0114] To ensure the generalization ability of the model, cross-validation is used to optimize model parameters. The dataset is divided into a training set and a validation set. The model is trained and evaluated repeatedly to ultimately select the model configuration with the best performance.

[0115] Use the reserved test set to evaluate the model's performance and ensure it can accurately predict the wear status of new symbol sequences. Evaluation metrics include precision, recall, and F1 score. Assume there are 100 symbol sequences in the test set, and 50 of them are correctly classified as light wear, the model's accuracy is 50%.

[0116] If the model performance is not ideal, you can consider introducing a semi-supervised learning method to use a small amount of labeled data to guide the model's learning process: fine-tune the model based on the labeled data to improve the accuracy of the model. You can also use historical experience to make manual adjustments, especially for those cases where the model is difficult to distinguish, as historical experience can provide valuable supplementary information. Suppose there is a series of symbol sequences: "046147258046". After K-means clustering, it is found that some symbol combinations (such as "046" and "147") frequently appear in the same cluster. Based on the characteristics of these symbol combinations, they are labeled as slight wear. Then, the autoencoder is used to train the model to minimize the difference between the original symbol sequence and the reconstructed symbol sequence. The model parameters are optimized through cross-validation, and the model performance is evaluated on the test set, ultimately obtaining a model that can accurately predict the wear status.

[0117] S4, obtaining a tool wear state prediction value based on the symbol sequence data by using the pre-trained tool wear state prediction model;

[0118] S5. According to the prediction value output by the tool wear state prediction model, combined with the processing requirements and tool characteristics, formulate corresponding compensation strategies, such as Figure 3 shown.

[0119] If the tool wear prediction value is slight wear (such as the symbol combination "046" and "147"), the cutting parameters are adjusted; in this embodiment, preferably, the cutting speed is reduced by 5% and the feed rate is increased by 10% to slow down the wear rate.

[0120] For tool wear prediction values ​​of moderate wear (such as symbol combination "157"), the tool is partially ground and the tool position is recalibrated; in this embodiment, preferably, the grinding amount is 0.1 mm.

[0121] For tool wear prediction values ​​that indicate severe wear (such as symbol combination "258"): replace the tool with a new one immediately, and reset the tool and lathe parameters according to the characteristics of the new tool.

[0122] If the tool wear prediction value is abnormal (such as the symbol combination "168" or "249"): stop processing immediately, check the tool and other possible problems to avoid further damage.

[0123] The compensation strategy is converted into specific CNC instructions, which are then executed by the CNC system. Specifically, instructions for adjusting cutting parameters can be sent directly to the CNC system's parameter setting module, enabling real-time parameter updates. Instructions for tool grinding and replacement must be processed through the CNC system's tool management module to ensure accurate execution of the compensation strategy.

[0124] Example 2

[0125] Combine Figure 2 As shown, based on steps S1-S5 of Example 1, this embodiment introduces a symbolic transfer probability matrix to characterize the future wear trend of the tool; and further formulates a tool wear compensation strategy based on the future wear trend of the tool.

[0126] S6. Input the symbol sequence data generated in real time into the trained wear state prediction model, and combine it with the trend prediction database of the symbol sequence data to predict the future wear trend and the current tool wear state in real time.

[0127] A dynamic trend prediction database is formed using the symbolic transition probability matrix to predict trends. Real-time data is input into the tool wear state prediction model to generate a prediction. This data is then compared against the database formed by the symbolic transition probability matrix for further prediction and verification. This not only improves prediction accuracy but also captures potential trend changes and anomalies.

[0128] Specifically, based on the trained tool wear state prediction model, a preliminary prediction of the wear state is made on the real-time data. Combined with the symbolic transition probability matrix, the probability of transitioning from the current state to the future state is evaluated to provide a more detailed trend forecast. By combining the two, the system can more accurately judge the current and future wear states and take corresponding compensation strategies in a timely manner.

[0129] S61. The symbol transition probability matrix describes the transition relationship between different symbols and is used as a dynamic database for trend prediction. Its construction steps are as follows:

[0130] 1) Count the number of transfers: For each symbol category (cutting force, temperature, vibration), count the number of transfers N between its internal symbols ij , and i and j must be transferred adjacently within the same category and cannot cross categories. Specifically, the cutting force symbol "0" can only be transferred to "1" or "2", but cannot jump directly to the temperature symbol "4".

[0131] For the symbol sequence: "046147258046", the data at each time point after decomposition is shown in Table 1:

[0132] Table 1

[0133] Time point Cutting force (F) Temperature (T) Vibration (V) <![CDATA[t1]]> 0 4 6 <![CDATA[t2]]> 1 4 7 <![CDATA[t3]]> 2 5 8 <![CDATA[t4]]> 0 4 6

[0134] From Table 1 we can see that:

[0135] Cutting force symbol (0,1,2): The symbol "0" appears twice, the symbol "1" appears once, and the symbol "2" appears once.

[0136] Transfer from 0 to 1: 1 time, transfer from 1 to 2: 1 time, transfer from 2 to 0: 1 time.

[0137] Temperature symbols (3,4,5): The symbol "4" appears 3 times and the symbol "5" appears once.

[0138] Transfer from 4 to 4: 1 time, transfer from 4 to 5: 1 time, transfer from 5 to 4: 1 time.

[0139] Vibrating symbols (6,7,8): The symbol "6" appears twice, the symbol "7" appears once, and the symbol "8" appears once.

[0140] Transfer from 6 to 7: 1 time, transfer from 7 to 8: 1 time, transfer from 8 to 6: 1 time.

[0141] 2) According to the number of transitions, calculate the probability P of transitioning from symbol i to symbol j ij :

[0142]

[0143] Among them, N ij is the number of transfers from symbol i to symbol j, ∑ k N ik is the total number of times symbol i appears.

[0144] Cutting force sign transition probability matrix:

[0145] 0 appears 2 times and transitions to 1 once:

[0146]

[0147] 1 appears once and shifts to 2 once:

[0148]

[0149] 2 appears once, and shifts to 0 once:

[0150]

[0151] Therefore, the cutting force sign transition probability matrix is:

[0152]

[0153] Similarly, the temperature symbol transition probability matrix is:

[0154]

[0155] Similarly, the vibration symbol transfer probability matrix is:

[0156]

[0157] The symbol transfer probability matrix is ​​not only a static probability table, but also a dynamic database that is updated in real time and used to predict future wear trends.

[0158] S62. Using symbol transfer probability matrix to assist wear state prediction:

[0159] The real-time symbol sequence is "046", the tool wear state prediction model predicts slight wear, and the symbol transfer probability matrix shows that it is most likely to transfer to "147" in the future, that is, moderate wear. Based on the slight wear compensation strategy, the adjustment range of the cutting parameters can be increased; in this embodiment, it is preferred to reduce the cutting speed by 10% and increase the feed rate by 15% to slow down the wear speed in advance.

[0160] The real-time symbol sequence is "157", and the tool wear state prediction model predicts moderate wear. The symbol transition probability matrix shows that there is a high probability of transitioning to "258" in the future, which means severe wear. The tool can be replaced to avoid further damage.

[0161] The real-time symbol sequence is "046", the tool wear state prediction model predicts slight wear, and the symbol transition probability matrix shows that it may transition to an abnormal state (such as "168") in the future; this indicates that some unforeseen working condition changes have occurred, and the tool and other equipment should be checked immediately.

[0162] The real-time symbol sequence is "168", the tool wear state prediction model predicts an abnormal state, and the symbol transition probability matrix shows that it may continue to remain in the abnormal state or turn to a more serious state (such as "269") in the future; then stop processing immediately, check the tool and other equipment to prevent further damage.

[0163] S63. The symbol transition probability matrix can help explain why certain symbol combinations appear frequently, providing a more intuitive understanding. If the model predicts moderate wear, but the symbol transition probability matrix indicates a return to slight wear in the future, this can be interpreted as the tool being restored to good working condition after regrinding.

[0164] S64, using the symbol transfer probability matrix to monitor abnormal conditions;

[0165] If the symbol transition probability matrix shows a sudden shift from a "normal" symbol combination (such as "046") to an abnormal symbol combination (such as "168"), this may indicate an impending tool failure or other problem, triggering an alarm mechanism in a timely manner. When the symbol sequence remains in an abnormal state (such as "168") for a long time, the symbol transition probability matrix can help identify whether this anomaly is temporary or persistent, thereby determining whether the process needs to be stopped immediately.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent prediction and compensation of tool wear in CNC machining, characterized in that: include: Collect cutting process data of CNC machining tools and perform preprocessing to obtain standardized data; Perform symbol conversion on the standardized data to obtain symbol sequence data, including: Calculate the mean and standard deviation of the standardized data as statistical features; Divide the symbol interval based on statistical characteristics and assign symbols to the standardized data: Cutting force range division Temperature range division Vibration range division in, Indicates cutting force signal, Indicates temperature signal; Indicates vibration signal; 、 、 are the mean values ​​of cutting force signal, temperature signal and vibration signal respectively, 、 、 are the standard deviations of cutting force signal, temperature signal and vibration signal respectively; A tool wear state prediction model is used to obtain a tool wear state prediction value based on the symbol sequence data. The tool wear state prediction model is an unsupervised learning model. The symbol sequence is clustered and labeled, and each cluster is mapped to a corresponding wear pattern. The model parameters are optimized using cross-validation to minimize the difference between the input symbol sequence and the reconstructed symbol sequence. A wear compensation strategy is determined according to the tool wear state prediction value; the wear compensation strategy is used as a control instruction and executed through a numerical control system.

2. The method for intelligent prediction and compensation of CNC machining tool wear according to claim 1, characterized in that: Also includes: The future wear trend of the tool is characterized by the symbolic transition probability matrix; The trend compensation strategy is obtained by combining the tool wear state prediction value and future wear trend.

3. The method for intelligent prediction and compensation of CNC machining tool wear according to claim 1, characterized in that: The cutting process data of the CNC machining tool includes: Cutting force signal, temperature signal and vibration signal.

4. The method for intelligent prediction and compensation of CNC machining tool wear according to claim 1, characterized in that: The method of determining the wear compensation strategy according to the tool wear state prediction value includes: If the tool wear state prediction value is light wear, adjust the cutting parameters; If the tool wear state prediction value is moderate wear, the tool is locally sharpened; If the tool wear state prediction value is severe wear, replace the tool; If the tool wear status prediction value is abnormal, the processing is stopped.

5. The method for intelligent prediction and compensation of CNC machining tool wear according to claim 2, characterized in that: The trend compensation strategy obtained by combining the tool wear state prediction value and the future wear trend includes: If the tool wear state prediction value is light wear and the future wear trend is moderate wear, increase the adjustment range of cutting parameters; If the tool wear state prediction value is moderate wear and the future wear trend is severe wear, then replace the tool; If the tool wear state prediction value is light wear and the future wear trend is abnormal, the equipment should be repaired; If the tool wear state prediction value is abnormal and the future wear trend is abnormal, the process will be stopped.

6. The method for intelligent prediction and compensation of CNC machining tool wear according to claim 1, characterized in that: The method of obtaining a tool wear state prediction value based on the symbol sequence data by using the tool wear state prediction model includes: Pre-training the tool wear state prediction model; The pre-trained tool wear state prediction model is used to input symbol sequence data and output the tool wear state prediction value.

Images