Method and system for monitoring wear of a machining tool based on deep learning

Abstract

This application relates to a method and system for monitoring wear of machining tools based on deep learning. The method includes: acquiring multi-source sensor data of the machining tool and generating a real-time energy flow feature sequence through physical formula conversion; inputting the real-time energy flow feature sequence into a trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; calculating the irreversibility confidence level based on the instantaneous power contribution sequence through a time-inversion consistency test, and multiplying it point-by-point with the power contribution sequence to obtain an irreversible damage power sequence; integrating and summing the irreversible damage power sequence from the start of machining to the current moment to generate a cumulative irreversible damage energy value; and identifying the real-time wear state of the tool through a predefined mapping table based on the cumulative irreversible damage energy value, and obtaining a wear prediction value through wear amount regression prediction. This method enables integrated monitoring of wear on machining tools.

Description

Technical Field

[0001] This invention belongs to the field of CNC machining technology, and in particular relates to a method and system for monitoring tool wear in machining based on deep learning. Background Technology

[0002] With the development of CNC machining technology, deep learning technology is increasingly widely used in the field of industrial monitoring, resulting in intelligent monitoring technologies based on multi-source sensor signals such as vibration, cutting force, acoustic emission, and current.

[0003] Intelligent monitoring technology, characterized by end-to-end learning, can directly establish a mapping relationship between sensor data and tool wear amount or wear state, and has become the mainstream technology direction in the field of tool wear monitoring. Intelligent monitoring technology typically involves directly inputting the time-series signals collected by sensors into convolutional neural networks, recurrent neural networks (RNNs), or hybrid models (such as CNN-LSTM). Through the model's autonomous learning, it mines the correlated features in the data, thereby achieving the judgment and prediction of tool wear state or wear amount, without the need for complex feature engineering by humans.

[0004] However, the current monitoring method has obvious deep-seated defects: First, it is essentially a complex "correlation" learning process, and the model decision-making process lacks physical interpretability. When the machining conditions change, the signal feature distribution will drift accordingly, causing the generalization ability of the trained model to drop sharply. It is often necessary to collect a large amount of data again for the new conditions and make fine adjustments, which is costly. Second, it directly correlates the original signal with the wear geometry, ignoring the inherent physical nature of tool wear - the irreversible entropy increase process driven by energy. During the cutting process, some energy will be dissipated in the form of friction and plastic deformation, and these irreversible energy dissipations are the core reasons for tool material loss and performance degradation. Existing methods have failed to explore and utilize this fundamental physical law. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and system for monitoring tool wear in machining based on deep learning to address the aforementioned technical problems.

[0006] Firstly, this application provides a deep learning-based method for monitoring tool wear in machining processes, including:

[0007] S101. Acquire multi-source sensor data of machining tools, convert the multi-source sensor data through physical formulas, and generate a real-time energy flow characteristic sequence.

[0008] S102. Input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; wherein, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time.

[0009] S103. Based on the instantaneous power contribution sequence of each energy channel, the irreversibility confidence of each energy channel at each time step is calculated by the time inversion consistency test method, and the instantaneous power contribution sequence and the corresponding irreversibility confidence are multiplied point by point to obtain the irreversible damage power sequence of each energy channel.

[0010] S104. Integrate and sum the power sequences of each irreversible damage from the start of processing to the current time to generate the cumulative irreversible damage energy value.

[0011] S105. Based on the cumulative irreversible damage energy value, the real-time wear state of the machining tool is obtained by mapping and identification through a predefined cumulative irreversible damage energy-wear state mapping table, and the wear amount prediction value of the machining tool is obtained by wear amount regression prediction calculation.

[0012] In one embodiment, the multi-source sensor data includes cutting force signals, electrical power signals, vibration signals, and acoustic emission signals;

[0013] Acquire multi-source sensor data from machining tools, transform the multi-source sensor data using physical formulas, and generate a real-time energy flow characteristic sequence, including:

[0014] S201. Obtain the orthogonal accelerations of the machining tool and workpiece in orthogonal directions, and calculate the instantaneous cutting power value at each moment based on the cutting force signal through vector dot product to generate an instantaneous cutting power time series; wherein, the instantaneous cutting power time series is used to characterize the rate at which mechanical energy is used.

[0015] S202. Filter the electrical power signal to obtain the net input electrical power time series; wherein, the net input electrical power time series is used to characterize the core fundamental intensity of energy input;

[0016] S203. The vibration signal is numerically integrated to obtain the vibration velocity. Based on the vibration velocity and the preset equivalent mass of the tool-workpiece system, the instantaneous vibration energy value at each moment is calculated using the kinetic energy formula to generate an instantaneous vibration energy time series. The instantaneous vibration energy time series is used to characterize the instantaneous kinetic energy change of the machining tool due to vibration.

[0017] S204. Bandpass filtering of a specific frequency band is performed on the acoustic emission signal to obtain the filtered acoustic emission signal. The filtered acoustic emission signal is then divided into frames with a preset duration window, and the root mean square value of the filtered acoustic emission signal within each window is calculated. The root mean square values ​​are arranged in chronological order to obtain the acoustic emission energy flux density time series. The acoustic emission energy flux density time series is used to characterize the instantaneous energy release intensity of machining tool damage.

[0018] S205. Obtain the cutting parameter vector of the real-time machining state, and align and splice the instantaneous cutting power time series, net input electric power time series, instantaneous vibration energy time series, acoustic emission energy flux density time series and cutting parameter vector in the time dimension to generate a real-time energy flow feature sequence.

[0019] In one embodiment, the vibration signal is numerically integrated to obtain the vibration velocity. Based on the vibration velocity and a preset equivalent mass of the tool-workpiece system, the instantaneous vibration energy value at each moment is calculated using the kinetic energy formula to generate an instantaneous vibration energy time series, including:

[0020] S301. Perform high-pass filtering on the vibration signal to obtain a pure vibration acceleration signal;

[0021] S302. Use the trapezoidal integral method to numerically integrate the pure vibration acceleration signal to obtain a preliminary vibration velocity sequence;

[0022] S303. Perform high-pass filtering on the preliminary vibration velocity sequence to obtain the vibration velocity time sequence;

[0023] S304. Calculate the instantaneous vibration energy value at each moment using the kinetic energy formula;

[0024] S305. Remove outliers from the instantaneous vibration energy values ​​and arrange them in chronological order to generate an instantaneous vibration energy time series.

[0025] In one embodiment, the expression for the cumulative irreversible damage energy value is:

[0026]

[0027] in, Indicates the end time The cumulative irreversible damage energy value, Indicates the total number of energy channels. Indicates the first Each energy channel at time The irreversible confidence level, Indicates the first Each energy channel at time The instantaneous power contribution sequence.

[0028] In one embodiment, the process for constructing the cumulative irreversible damage energy-wear state mapping table is as follows:

[0029] S501. Obtain the historical cumulative irreversible damage energy value data and the actual wear status label corresponding to the cumulative irreversible damage energy value data during the entire life cycle of machining tool wear.

[0030] S502. Classify the historical cumulative irreversible damage energy value data according to the actual wear status label, and generate the energy distribution range corresponding to each wear status.

[0031] S503. Based on the boundary values ​​of each energy distribution range, set a critical energy threshold; wherein, the critical energy threshold is used to distinguish different wear states of machining tools.

[0032] S504. Based on the critical energy threshold, define the energy boundary corresponding to the wear state of different machining tools, construct the correlation between the energy value range and the wear state label of different machining tools, and store the correlation in the form of a lookup table to obtain the cumulative irreversible damage energy-wear state mapping table.

[0033] Secondly, this application also provides a deep learning-based machining tool wear monitoring system, comprising:

[0034] The feature conversion module is used to acquire multi-source sensor data of machining tools, convert the multi-source sensor data through physical formulas, and generate real-time energy flow feature sequences.

[0035] The decoupling network module is used to input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; wherein, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time.

[0036] The damage power generation module is used to calculate the irreversible confidence of each energy channel at each time step based on the instantaneous power contribution sequence of each energy channel through the time inversion consistency test method, and multiply the instantaneous power contribution sequence and the corresponding irreversible confidence of the instantaneous power contribution sequence point by point to obtain the irreversible damage power sequence of each energy channel.

[0037] The damage energy calculation module is used to integrate and sum the power sequences of each irreversible damage from the start of processing to the current time to generate the cumulative irreversible damage energy value.

[0038] The state decision and prediction module is used to obtain the real-time wear state of the machining tool by mapping and identifying the cumulative irreversible damage energy value through a predefined cumulative irreversible damage energy-wear state mapping table, and to obtain the predicted wear value of the machining tool by wear amount regression prediction calculation.

[0039] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0041] The aforementioned deep learning-based methods, systems, equipment, and media for monitoring tool wear in machining achieve physical interpretability of data by converting multi-source sensor data into a unified physical meaning energy flow feature sequence and unifying the dimensions of multi-source signals, thus providing inputs with clear physical meaning for subsequent models. Furthermore, by using an energy dissipation decoupling network model to decouple the instantaneous power contribution sequence of each channel from the energy flow features, it reveals the dynamic proportion and contribution law of different energy forms in the total dissipation, enhancing the interpretability and generalization ability of the energy dissipation decoupling network model. Finally, by calculating the irreversibility confidence of each channel through time-reversal consistency checks and comparing it with the power contribution... The irreversible damage power sequence is obtained by multiplying the power values ​​point by point. Reversible energy fluctuations are then removed, focusing on the irreversible energy dissipation process leading to material damage. The irreversible damage power sequence is integrated and summed in the time domain to generate a cumulative irreversible damage energy value. This value quantifies the total energy driving irreversible damage to the tool material from an energy accumulation perspective, establishing an energy accumulation characterization of the wear process. A pre-defined cumulative irreversible damage energy-wear state mapping table enables rapid and reliable identification of the wear state. Combined with wear amount regression prediction, a quantitative wear value is output, achieving real-time, accurate, and interpretable integrated monitoring of tool wear state and wear amount. This method enables integrated monitoring of wear on machining tools. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1A schematic flowchart illustrating a deep learning-based method for monitoring tool wear in machining, provided as an exemplary embodiment of this application;

[0044] Figure 2 This is a schematic diagram of a deep learning-based machining tool wear monitoring system provided as an exemplary embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0046] In one embodiment, such as Figure 1 As shown, a deep learning-based method for monitoring tool wear in machining is provided. This embodiment illustrates the application of this method to a monitoring terminal. It is understood that this method can also be applied to a monitoring server, or to a system including both a monitoring terminal and a monitoring server, and is implemented through interaction between the monitoring terminal and the monitoring server. In this embodiment, the method includes the following steps:

[0047] S101. Acquire multi-source sensor data of machining tools, convert the multi-source sensor data through physical formulas, and generate a real-time energy flow characteristic sequence.

[0048] Optionally, the multi-source sensor data can be collected by various sensors deployed at the machining site, and can be multi-dimensional time-series signals used to characterize the energy conversion and dissipation of the machining process using machining tools.

[0049] Optionally, the real-time energy flow feature sequence can be used to characterize a composite feature vector that has a unified energy physics meaning, formed by aligning and splicing multiple physical quantities in the time dimension.

[0050] Specifically, the monitoring terminal can establish real-time communication connections with various sensors deployed at the machining site through a data acquisition interface to acquire multi-source sensor data from machining tools. Furthermore, the monitoring terminal can use physical formulas to transform the multi-source sensor data, mapping the raw physical signals to energy-related features to obtain various energy-related time-series sequences. Even further, the monitoring terminal can align these energy-related time-series sequences with the real-time acquired cutting parameter vectors in the time dimension and sequentially concatenate them to generate a time-aligned real-time energy flow feature sequence.

[0051] S102. Input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel.

[0052] Optionally, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time.

[0053] Optionally, the energy dissipation decoupling network model can be a model built on a deep learning architecture.

[0054] Alternatively, the training process of the energy dissipation decoupling network model can be carried out in the following ways:

[0055] S1021. Obtain the energy flow characteristic sequence and labeled power contribution data of each energy channel collected throughout the entire life cycle of the tool as a sample set. Among them, the energy flow characteristic sequence can be a composite feature vector obtained by physical formula transformation of the corresponding machining tool from the new state to the failure state throughout the entire machining stage. The power contribution data of each energy channel can be labeled by combining physical modeling analysis and experimental calibration to characterize the actual contribution ratio of each energy channel to the total energy dissipation at different machining moments.

[0056] S1022. Perform preprocessing operations on the sample set, divide the sample set into training set, validation set and test set according to a preset ratio, perform data normalization processing on the energy flow feature sequence to eliminate the dimensional differences between different feature dimensions, complete the sequence with missing length, and smooth the feature value with abnormal fluctuation.

[0057] S1023. Construct a deep learning network architecture, which may include, but is not limited to, a feature extraction layer, a temporal modeling layer, and an energy channel decoupling layer.

[0058] Preferably, the feature extraction layer can adopt a convolutional neural network structure, which can be used to mine local correlation features in the energy flow feature sequence; the temporal modeling layer can adopt a long short-term memory network or a Transformer structure to capture the dynamic dependency relationship of energy flow features over time; the energy channel decoupling layer can adopt a design that combines branch architecture and attention mechanism to decouple the fused features into power contribution features corresponding to each energy channel.

[0059] S1024. Set the hyperparameters for model training, select the mean squared error loss function as the optimization objective of the energy dissipation decoupling network model, the mean squared error loss function is used to calculate the deviation between the power contribution value of each energy channel predicted by the model and the labeled value, select the adaptive moment estimation optimizer to update the parameters iteratively, configure the preset initial value of learning rate and learning rate decay strategy, and adopt the early stopping mechanism to prevent the model from overfitting.

[0060] S1025. Input the preprocessed training set samples into the constructed deep learning network architecture, and calculate the predicted value of the power contribution of each energy channel through forward propagation. Calculate the loss value between the predicted value and the labeled value based on the loss function. Further, update the weight parameters of the deep learning network architecture layer by layer according to the loss value through the backpropagation algorithm. After each iteration, input the validation set samples into the model for performance evaluation. When the validation set loss value does not show a decreasing trend for several consecutive iterations, trigger the early stopping mechanism to terminate training.

[0061] S1026. After training, the test set samples are input into the energy dissipation decoupling network model for generalization capability verification. The prediction accuracy of the model is evaluated by indicators such as the coefficient of determination (R²) and root mean square error (RMSE). When the energy dissipation decoupling network model meets the preset accuracy requirements, the network structure and weight parameters of the energy dissipation decoupling network model are fixed and stored, and the trained energy dissipation decoupling network model is obtained.

[0062] For example, the monitoring terminal can perform data normalization, outlier removal, and sequence length matching on the real-time energy flow feature sequence to meet the input format requirements of the energy dissipation decoupling network model. Furthermore, the monitoring terminal can input the preprocessed real-time energy flow feature sequence into the energy dissipation decoupling network model. The energy dissipation decoupling network model uses forward propagation calculations through a multi-layer neural network to extract features layer by layer and complete the decoupling operation of energy channels, outputting the instantaneous power contribution sequence of each energy channel.

[0063] S103. Based on the instantaneous power contribution sequence of each energy channel, the irreversibility confidence of each energy channel at each time step is calculated by the time inversion consistency test method. The instantaneous power contribution sequence and the corresponding irreversibility confidence are multiplied point by point to obtain the irreversible damage power sequence of each energy channel.

[0064] Optionally, the irreversibility confidence level can be used to quantify the irreversibility of the energy dissipation process at each moment. The value ranges from 0 to 1, and the closer the value is to 1, the stronger the irreversibility of the energy dissipation at that moment.

[0065] Optionally, the irreversible damage power sequence can be used to characterize the true power components of irreversible damage caused to the machining tool material by each energy channel at different times.

[0066] Specifically, the monitoring terminal can perform time-inversion consistency checks on the instantaneous power contribution sequence. By performing time-inversion transformation on the instantaneous power contribution sequence at each moment, the similarity between the inverted sequence and the original sequence is calculated, and the irreversibility confidence of each energy channel at each moment is determined based on the similarity value.

[0067] Furthermore, after completing the irreversible confidence calculation, the monitoring terminal can perform point-by-point multiplication of the instantaneous power contribution sequence of each energy channel and the irreversible confidence corresponding to the instantaneous power contribution sequence, remove the components of reversible energy dissipation in the instantaneous power contribution sequence, and retain the power part generated only by the irreversible process to obtain the irreversible damage power sequence of each energy channel.

[0068] S104. Integrate and sum the power sequences of each irreversible damage from the start of processing to the current time to generate the cumulative irreversible damage energy value.

[0069] Optionally, the cumulative irreversible damage energy value can be used to characterize the total damage energy accumulated by the machining tool from the start of machining to the current moment, which is the sum of the irreversible damage power of each energy channel.

[0070] Specifically, the monitoring terminal can determine the time range of the integration operation. The start time can be the time when the machining tool begins to execute the machining task, and the end time can be the current machining time.

[0071] Furthermore, the monitoring terminal can divide the continuous time interval into several small time steps using the trapezoidal integral method, approximate the irreversible damage power within each time step as a constant value, calculate the power-time area corresponding to each time step and sum them up to obtain the cumulative irreversible damage energy value of a single energy channel within that time range.

[0072] Furthermore, the monitoring terminal can sum the cumulative irreversible damage energy values ​​of all energy channels after completing the integral calculation of a single energy channel, and generate the cumulative irreversible damage energy value up to the current moment by adding up the damage energy values ​​of each energy channel.

[0073] S105. Based on the cumulative irreversible damage energy value, the real-time wear state of the machining tool is obtained by mapping and identification through a predefined cumulative irreversible damage energy-wear state mapping table, and the wear amount prediction value of the machining tool is obtained by wear amount regression prediction calculation.

[0074] Optionally, wear amount regression prediction can construct a quantitative regression relationship for wear amount based on historical correlation data between cumulative irreversible damage energy value and actual tool wear amount. The quantitative regression relationship for wear amount can be established through statistical analysis methods such as multinomial fitting, linear regression and nonlinear fitting, which can realize the quantitative mapping from cumulative irreversible damage energy value to tool wear amount.

[0075] Specifically, the monitoring terminal can retrieve the pre-stored cumulative irreversible damage energy-wear state mapping table, match the calculated cumulative irreversible damage energy value with the energy value range in the cumulative irreversible damage energy-wear state mapping table, and obtain the matching result.

[0076] Furthermore, the monitoring terminal can identify the real-time wear status of the machining tool based on the matching results. The real-time wear status may include, but is not limited to, three stages: initial wear, normal wear, and severe wear.

[0077] Furthermore, the monitoring terminal can call the quantitative regression relationship of wear amount, substitute the current cumulative irreversible damage energy value into the quantitative regression relationship of wear amount, and calculate and output the corresponding predicted value of machining tool wear amount through corresponding mathematical operations.

[0078] Preferably, the quantitative regression relationship of wear amount is taken as an example by polynomial fitting, and the expression can be:

[0079]

[0080] In the formula, This represents the predicted wear amount of machining tools. Indicates the end time The cumulative irreversible damage energy value, Represents the coefficient of the constant term. Indicates the first The regression coefficients of the order polynomial, the first The values ​​of the regression coefficients of the order polynomial are determined by fitting historical correlation data. The order of the polynomial can be selected based on the actual fitting accuracy requirements and should be a positive integer.

[0081] In the aforementioned deep learning-based method for monitoring tool wear in machining, the monitoring terminal converts multi-source sensor data into a unified physical meaning energy flow feature sequence, unifying the dimensions of multi-source signals and enabling physical interpretability of the data, providing inputs with clear physical meaning for subsequent models. Through an energy dissipation decoupling network model, the instantaneous power contribution sequence of each channel is decoupled from the energy flow features, revealing the dynamic proportion and contribution law of different energy forms in the total dissipation, thus enhancing the interpretability and generalization ability of the energy dissipation decoupling network model. Finally, the irreversibility confidence of each channel is calculated through a time-reversal consistency test and compared with the power contribution. Dot multiplication yields an irreversible damage power sequence, stripping away reversible energy fluctuations and focusing on the irreversible energy dissipation process leading to material damage. Time-domain integration and summation of the irreversible damage power sequence generate a cumulative irreversible damage energy value. This value quantifies the total energy driving irreversible damage to the tool material from an energy accumulation perspective, establishing an energy accumulation characterization of the wear process. A pre-defined cumulative irreversible damage energy-wear state mapping table enables rapid and reliable identification of the wear state. Combined with wear amount regression prediction, a quantitative wear value is output, achieving real-time, accurate, and interpretable integrated monitoring of tool wear state and wear amount. This method enables integrated monitoring of wear on machining tools.

[0082] In one embodiment, acquiring multi-source sensor data of a machining tool and converting the multi-source sensor data using a physical formula to generate a real-time energy flow feature sequence may include the following steps:

[0083] S201. Obtain the orthogonal acceleration of the machining tool and the workpiece in the orthogonal direction, and calculate the instantaneous cutting power value at each moment based on the cutting force signal by vector dot product, and generate an instantaneous cutting power time series.

[0084] Optionally, multi-source sensor data may include, but is not limited to, cutting force signals, electrical power signals, vibration signals, and acoustic emission signals.

[0085] Optionally, the instantaneous cutting power time series can be used to characterize the rate at which mechanical energy is done.

[0086] For example, the monitoring terminal can establish real-time data interaction with the triaxial accelerometer deployed at the tool holder and workpiece fixture and the cutting force sensor installed at the machine tool tool connection part through the industrial communication link, and obtain orthogonal acceleration data and cutting force signals. The cutting force signal collected by the cutting force sensor can be a multi-directional composite signal.

[0087] Furthermore, the monitoring terminal can perform orthogonal decomposition on the cutting force signal to obtain mutually perpendicular orthogonal cutting force components. Secondly, the monitoring terminal can perform time-stamp alignment, outlier identification and removal, and noise reduction filtering preprocessing on the orthogonal cutting force component data and the corresponding directional acceleration data.

[0088] Furthermore, the monitoring terminal can perform vector dot product calculations on the preprocessed orthogonal cutting force components and the corresponding directional acceleration components of the orthogonal cutting force components at each moment, and convert the calculation results into instantaneous cutting power values ​​in combination with the basic physical laws of cutting motion. The instantaneous cutting power values ​​at each moment are arranged in order according to time sequence to generate an instantaneous cutting power time series.

[0089] S202. Filter the power signal to obtain the net input power time series.

[0090] Optionally, the net input electrical power time series can be used to characterize the core fundamental strength of the energy input.

[0091] For example, the monitoring terminal can acquire the power signal collected by the power sensor, which may include the total input power information for driving the spindle to drive the tool to complete the cutting operation.

[0092] Furthermore, the monitoring terminal can employ an adaptive digital filtering algorithm to process the electrical power signal. This algorithm presets filtering parameters based on the energy consumption characteristics of the cutting process, identifying and filtering out invalid power components unrelated to cutting energy, retaining only the effective power portion directly related to the tool's cutting load. Finally, the monitoring terminal can arrange the filtered electrical power data in chronological order to generate a net input electrical power time series.

[0093] S203. Perform numerical integration on the vibration signal to obtain the vibration velocity, and based on the vibration velocity and the preset equivalent mass of the tool-workpiece system, calculate the instantaneous vibration energy value at each moment using the kinetic energy formula to generate an instantaneous vibration energy time series.

[0094] Optionally, the instantaneous vibration energy time series can be used to characterize the instantaneous kinetic energy change of machining tools caused by vibration.

[0095] Optionally, the equivalent mass of the tool-workpiece system can be used to physically correlate the measured vibration velocity and vibration kinetic energy, and is a physical constant for realizing the conversion from vibration signal to vibration energy.

[0096] For example, the monitoring terminal can use the trapezoidal integral method to perform numerical integration on the vibration signal to obtain a preliminary vibration velocity sequence. To eliminate the trend term error introduced by the integration operation, a high-pass filter is applied to the preliminary vibration velocity sequence to obtain a vibration velocity time sequence. Further, the monitoring terminal can retrieve the equivalent mass parameters of the tool-workpiece system and calculate the instantaneous vibration energy value moment-by-moment using the kinetic energy formula. Secondly, the monitoring terminal can identify and remove outliers from the instantaneous vibration energy values ​​and arrange them in chronological order to generate an instantaneous vibration energy time sequence.

[0097] S204. Perform bandpass filtering on the acoustic emission signal in a specific frequency band to obtain the filtered acoustic emission signal. Perform frame processing on the filtered acoustic emission signal in a window of preset duration, calculate the root mean square value of the filtered acoustic emission signal in each window, and arrange the root mean square values ​​in time order to obtain the acoustic emission energy flux density time series.

[0098] Optionally, the acoustic emission energy flux density time series can be used to characterize the instantaneous energy release intensity of machining tool damage.

[0099] Optionally, the filtered acoustic emission signal can be used to characterize the core acoustic emission waveform information within a specific physically sensitive frequency band, which is directly excited by microscopic damage events such as wear of machining tools and microcracks after filtering out environmental noise and irrelevant frequency band interference.

[0100] For example, the monitoring terminal can use a bandpass filtering algorithm to process the acoustic emission signal. The filtering frequency band can be set based on the characteristic frequency range of the acoustic emission signal generated by tool material damage, enabling the extraction of signal components directly related to tool damage to obtain the filtered acoustic emission signal. Furthermore, the monitoring terminal can perform frame processing on the filtered acoustic emission signal using a preset time window, the window length being determined according to the dynamic characteristics of the cutting process. Further, the monitoring terminal can calculate the root mean square (RMS) value of the filtered acoustic emission signal within each window, which quantifies the energy intensity of the acoustic emission signal within the window period. Finally, the monitoring terminal can arrange the RMS values ​​of each time window in chronological order to obtain a time series of acoustic emission energy flux density.

[0101] S205. Obtain the cutting parameter vector of the real-time machining state, and align and splice the instantaneous cutting power time series, net input electric power time series, instantaneous vibration energy time series, acoustic emission energy flux density time series and cutting parameter vector in the time dimension to generate a real-time energy flow feature sequence.

[0102] Optionally, the cutting parameter vector can be used to characterize the cutting conditions at the current moment, and may include, but is not limited to, key machining parameters such as cutting speed, feed rate, and depth of cut.

[0103] For example, the monitoring terminal can use a high-precision timestamp as a reference to align the instantaneous cutting power time series, net input electrical power time series, instantaneous vibration energy time series, and acoustic emission energy flux density time series in the time dimension to obtain an aligned energy-related time series. Furthermore, the monitoring terminal can integrate the aligned energy-related time series with the cutting parameter vector to generate a real-time energy flow feature sequence with unified dimensions and clear physical meaning.

[0104] In this embodiment, the monitoring terminal transforms multi-source heterogeneous sensor data into a feature sequence with clear energy physics meaning according to physical laws. This provides a standardized input that is physically interpretable, dimensionally unified, and strongly correlated with the nature of wear for subsequent models, fundamentally enhancing the monitoring system's ability to generalize operating conditions and the reliability of decision-making.

[0105] In one embodiment, the vibration signal is numerically integrated to obtain the vibration velocity, and based on the vibration velocity and a preset equivalent mass of the tool-workpiece system, the instantaneous vibration energy value at each moment is calculated using the kinetic energy formula to generate an instantaneous vibration energy time series, which may include the following steps:

[0106] S301. Perform high-pass filtering on the vibration signal to obtain a pure vibration acceleration signal.

[0107] Alternatively, the pure vibration acceleration signal can be used to characterize the actual vibration acceleration variation characteristics of the machining tool during the cutting process after eliminating low-frequency drift interference.

[0108] For example, the monitoring terminal can invoke a high-pass filtering algorithm suitable for timing signal processing, and the filter cutoff frequency can be set based on the low-frequency vibration characteristics of the machine tool. Furthermore, the monitoring terminal can use the high-pass filtering algorithm to filter the vibration signal, effectively removing low-frequency drift interference components from the signal and retaining only the high-frequency effective signal directly related to the cutting vibration of the tool, ultimately obtaining a pure vibration acceleration signal.

[0109] S302. Use the trapezoidal integral method to numerically integrate the pure vibration acceleration signal to obtain a preliminary vibration velocity sequence.

[0110] For example, the monitoring terminal can invoke the trapezoidal integral method, which can divide a continuous time interval into several equidistant small steps, and approximate the acceleration within each step as a constant value. The area of ​​the trapezoid formed by the acceleration values ​​at adjacent times represents the velocity change within that step, and finally, the vibration velocity at the corresponding time is obtained by accumulating the velocity changes of each step.

[0111] Furthermore, the monitoring terminal can divide the pure vibration acceleration signal into segments according to the time step, perform trapezoidal integration on each segment, convert the acceleration signal into vibration velocity data, arrange the vibration velocity values ​​calculated at each moment in chronological order, and generate a preliminary vibration velocity sequence.

[0112] S303. Perform high-pass filtering on the preliminary vibration velocity sequence to obtain the vibration velocity time sequence.

[0113] For example, the monitoring terminal can further process the integral trend term error present in the initial vibration velocity sequence by calling a high-pass filtering algorithm. This integral trend term error is generated by the accumulation of tiny acceleration baseline offsets during the integration process and is unrelated to the actual vibration velocity of the machining tool. Furthermore, the monitoring terminal can set filtering parameters based on the characteristics of the vibration velocity signal to ensure that the filtering process only removes the trend term error in the sequence and does not affect the dynamic velocity components related to the vibration of the machining tool.

[0114] Furthermore, the monitoring terminal can eliminate invalid biases in the preliminary vibration velocity sequence through filtering to obtain the vibration velocity time sequence.

[0115] S304. Calculate the instantaneous vibration energy value at each moment using the kinetic energy formula.

[0116] For example, the monitoring terminal can call the classical kinetic energy formula, substitute the vibration velocity value at each moment in the vibration velocity time series and the equivalent mass of the tool-workpiece system into the formula, perform squaring and multiplication operations at each moment, and finally obtain the instantaneous vibration energy value corresponding to each moment.

[0117] S305. Remove outliers from the instantaneous vibration energy values ​​and arrange them in chronological order to generate an instantaneous vibration energy time series.

[0118] For example, the monitoring terminal can adopt an outlier identification method based on statistical distribution. The outlier identification method can set a confidence interval based on the overall distribution characteristics of instantaneous vibration energy values, and determine values ​​that exceed the confidence interval as outliers. Such outliers are mostly caused by non-steady-state factors such as instantaneous interference of sensors and accidental impacts during the cutting process.

[0119] Furthermore, the monitoring terminal can individually analyze the calculated instantaneous vibration energy values ​​and eliminate any abnormal values. Secondly, the monitoring terminal can arrange the eliminated instantaneous vibration energy values ​​in chronological order to generate an instantaneous vibration energy time series.

[0120] In this embodiment, the monitoring terminal converts the original vibration signal into an accurate and stable instantaneous vibration energy time series through a multi-level signal processing link of multiple filtering and numerical integration. This effectively eliminates measurement noise and integration errors, ensuring the physical accuracy and data reliability of vibration energy characterization, and providing key and robust feature inputs for tool wear state analysis based on energy perspective.

[0121] In one embodiment, the expression for the cumulative irreversible damage energy value can be:

[0122]

[0123] in, Indicates the end time The cumulative irreversible damage energy value, Indicates the total number of energy channels. Indicates the first Each energy channel at time The irreversible confidence level, Indicates the first Each energy channel at time The instantaneous power contribution sequence.

[0124] For example, the monitoring terminal can retrieve the pre-stored cumulative irreversible damage energy value calculation expression and combine it with the parameters generated in the previous steps to complete the numerical calculation.

[0125] Preferably, This represents the total number of energy channels obtained after the energy dissipation decoupling network model decouples the real-time energy flow feature sequence. The total number of energy channels and the dimension of the energy-related time series included in the real-time energy flow feature sequence are mutually matched, and can adaptively match with the dimension of the input features. The output results can be derived from the energy dissipation decoupling network model trained in step S102, providing instantaneous power contribution data for each energy channel at different times during the processing. The irreversibility confidence level, which can be derived from the time-reversal consistency test method in step S103, can be used to quantify the irreversibility of energy dissipation at each time step. Secondly, regarding the upper and lower limits of integration, It can be the machining start time of the machining tool. For the current processing time, the monitoring terminal uses the trapezoidal integral method to... and The product terms are integrated in the time domain to obtain the cumulative irreversible damage energy value of a single energy channel, and then applied to all... The integral results of each energy channel are summed to generate the cutoff time. Cumulative irreversible damage energy value .

[0126] In this embodiment, the monitoring terminal introduces an integral expression that integrates irreversible confidence to accumulate and quantify the energy contribution of multiple channels and the irreversibility of damage in the time domain. This enables a physically quantifiable characterization of the total damage energy of the tool, providing a direct physical basis for accurate judgment of wear status.

[0127] In one embodiment, the process of constructing the cumulative irreversible damage energy-wear state mapping table may include the following steps:

[0128] S501. Obtain the historical cumulative irreversible damage energy value data and the actual wear status label corresponding to the cumulative irreversible damage energy value data during the entire life cycle of machining tool wear.

[0129] Optionally, historical cumulative irreversible damage energy data can be used to characterize the dynamic change history of cumulative irreversible damage energy at each moment from brand new state to failure state during the entire machining cycle of different types of machining tools.

[0130] Optionally, the actual wear status label can be used to characterize the actual wear stage of the machining tool at the time corresponding to the acquisition time of the historical cumulative irreversible damage energy value data.

[0131] For example, the monitoring terminal can retrieve a pre-stored tool life cycle experimental database through a data interface. The tool life cycle experimental database can include historical operating data of tools of various specifications under different processing conditions. Among them, the historical cumulative irreversible damage energy value data comes from the calculation results of the cumulative irreversible damage energy at each moment during the corresponding tool's life cycle processing. The actual wear status label can be obtained by offline detection of tool wear morphology using a high-precision optical measuring instrument and manual annotation in combination with industry standards.

[0132] Furthermore, the monitoring terminal can perform integrity checks on the retrieved data, removing any missing or erroneous data entries.

[0133] S502. Classify the historical cumulative irreversible damage energy value data according to the actual wear status label, and generate the energy distribution range corresponding to each wear status.

[0134] Optionally, the energy distribution range can be used to characterize the range of historical cumulative irreversible damage energy values ​​under the same wear condition, reflecting the correspondence between different wear stages and cumulative irreversible damage energy.

[0135] For example, the monitoring terminal can divide the historical cumulative irreversible damage energy value data into multiple data subsets according to the actual wear state label, with each data subset corresponding to a wear state. Secondly, for each data subset, the monitoring terminal can use statistical analysis methods to calculate the central tendency and dispersion indicators of the data, and determine the energy value range corresponding to each wear state by analyzing the data distribution characteristics, thus generating the energy distribution interval corresponding to each wear state.

[0136] S503. Set the critical energy threshold based on the boundary values ​​of each energy distribution interval.

[0137] Optionally, the critical energy threshold is used to distinguish different wear states of machining tools.

[0138] For example, the monitoring terminal can extract the boundary values ​​of each energy distribution interval. For energy distribution intervals corresponding to adjacent wear states, the characteristic values ​​at the boundary of the energy distribution intervals are selected as the reference basis for the critical energy threshold. Furthermore, the monitoring terminal can modify and optimize the reference basis for the critical energy threshold in combination with the safety and stability requirements of the processing process, ensuring that the set critical energy threshold can distinguish different wear states. The setting of the critical energy threshold needs to take into account both the statistical characteristics of the data distribution and the application requirements of actual processing.

[0139] Furthermore, the monitoring terminal can verify the wear state by inputting historical data from the entire tool life cycle, statistically analyze the accuracy, false positive rate, and false negative rate of wear state identification under different thresholds, and iteratively adjust the threshold value based on the verification results until the wear state differentiation results fully meet the preset judgment criteria, and finally determine the critical energy threshold that can distinguish different wear states.

[0140] S504. Based on the critical energy threshold, define the energy boundary corresponding to the wear state of different machining tools, construct the correlation between the energy value range and the wear state label of different machining tools, and store the correlation in the form of a lookup table to obtain the cumulative irreversible damage energy-wear state mapping table.

[0141] For example, the monitoring terminal can retrieve the critical energy threshold obtained in step S503 and use the critical energy threshold as the core basis for dividing the energy boundaries of different wear states. Secondly, the monitoring terminal can divide the overall range of cumulative irreversible damage energy into multiple continuous and non-overlapping energy value intervals according to the critical energy threshold from low to high, with each energy value interval uniquely corresponding to a tool wear state label.

[0142] Furthermore, the monitoring terminal can establish a one-to-one correspondence between energy value ranges and wear status labels. Through this correspondence, the specific wear stage corresponding to each energy value range can be clearly identified. Even further, the monitoring terminal can organize the established relationships according to a structured data format and store them in local memory as key-value pairs, thus creating a cumulative irreversible damage energy-wear status mapping table.

[0143] In this embodiment, the monitoring terminal constructs an energy-wear status lookup table based on critical thresholds by statistically analyzing the historical data of the entire lifespan of machining tools. This enables a rapid and accurate mapping from cumulative energy values ​​to wear status, significantly improving the real-time performance and decision reliability of wear status identification.

[0144] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0145] Based on the same inventive concept, this application also provides a system for implementing the deep learning-based machining tool wear monitoring system described above. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more deep learning-based machining tool wear monitoring system embodiments provided below can be found in the limitations of the deep learning-based machining tool wear monitoring method described above, and will not be repeated here.

[0146] In one exemplary embodiment, such as Figure 2 As shown, a deep learning-based machining tool wear monitoring system 600 is provided, comprising:

[0147] The feature conversion module 601 is used to acquire multi-source sensor data of machining tools, convert the multi-source sensor data through physical formulas, and generate a real-time energy flow feature sequence.

[0148] The decoupling network module 602 is used to input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; wherein, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time.

[0149] The damage power generation module 603 is used to calculate the irreversible confidence of each energy channel at each time step based on the instantaneous power contribution sequence of each energy channel through the time inversion consistency test method, and multiply the instantaneous power contribution sequence and the irreversible confidence corresponding to the instantaneous power contribution sequence point by point to obtain the irreversible damage power sequence of each energy channel.

[0150] Damage energy calculation module 604 is used to integrate and sum the irreversible damage power sequences from the start of processing to the current time to generate a cumulative irreversible damage energy value.

[0151] The state decision and prediction module 605 is used to obtain the real-time wear state of the machining tool by mapping and identifying it through a predefined cumulative irreversible damage energy-wear state mapping table based on the cumulative irreversible damage energy value, and to obtain the predicted wear value of the machining tool by wear amount regression prediction calculation.

[0152] In one embodiment, the feature conversion module includes:

[0153] The cutting power sequence generation unit is used to acquire the orthogonal acceleration of the machining tool and the workpiece in the orthogonal direction, and calculate the instantaneous cutting power value at each moment based on the cutting force signal through vector dot product to generate an instantaneous cutting power time sequence; wherein, the instantaneous cutting power time sequence is used to characterize the rate at which mechanical energy is done;

[0154] The electric power filtering processing unit is used to filter the electric power signal to obtain the net input electric power time series; wherein, the net input electric power time series is used to characterize the core basic intensity of energy input;

[0155] The vibration energy calculation unit is used to numerically integrate the vibration signal to obtain the vibration velocity, and based on the vibration velocity and the preset equivalent mass of the tool-workpiece system, calculates the instantaneous vibration energy value at each moment using the kinetic energy formula to generate an instantaneous vibration energy time series; wherein, the instantaneous vibration energy time series is used to characterize the instantaneous kinetic energy change of the machining tool due to vibration;

[0156] The acoustic emission energy extraction unit is used to perform bandpass filtering on the acoustic emission signal in a specific frequency band to obtain the filtered acoustic emission signal. The filtered acoustic emission signal is then segmented into frames with a preset duration window, and the root mean square value of the filtered acoustic emission signal within each window is calculated. The root mean square values ​​are then arranged in chronological order to obtain the acoustic emission energy flux density time series. The acoustic emission energy flux density time series is used to characterize the instantaneous energy release intensity of machining tool damage.

[0157] The energy feature construction unit is used to obtain the cutting parameter vector of the real-time machining state, and to align and splice the instantaneous cutting power time series, net input electric power time series, instantaneous vibration energy time series, acoustic emission energy flux density time series and cutting parameter vector in the time dimension to generate a real-time energy flow feature sequence.

[0158] In one embodiment, the vibration energy calculation unit includes:

[0159] The filtering and purification subunit is used to perform high-pass filtering on the vibration signal to obtain a pure vibration acceleration signal.

[0160] The integral conversion subunit is used to call the trapezoidal integration method to numerically integrate the pure vibration acceleration signal to obtain a preliminary vibration velocity sequence.

[0161] The signal detrending subunit is used to perform high-pass filtering on the preliminary vibration velocity sequence to obtain the vibration velocity time sequence.

[0162] The kinetic energy calculation subunit is used to calculate the instantaneous vibration energy value at each moment using the kinetic energy formula;

[0163] The energy sequence generation sub-unit is used to remove outliers from instantaneous vibration energy values ​​and arrange them in chronological order to generate an instantaneous vibration energy time series.

[0164] In one embodiment, the damage energy calculation module includes:

[0165] The expression for the cumulative irreversible damage energy value is:

[0166]

[0167] in, Indicates the end time The cumulative irreversible damage energy value, Indicates the total number of energy channels. Indicates the first Each energy channel at time The irreversible confidence level, Indicates the first Each energy channel at time The instantaneous power contribution sequence.

[0168] In one embodiment, the state decision and prediction module includes:

[0169] The historical data acquisition unit is used to acquire historical cumulative irreversible damage energy value data and the actual wear status label corresponding to the cumulative irreversible damage energy value data throughout the entire life cycle of machining tool wear.

[0170] The state classification and interval division unit is used to classify historical cumulative irreversible damage energy value data according to the actual wear state label and generate energy distribution intervals corresponding to each wear state.

[0171] The threshold setting unit is used to set the critical energy threshold based on the boundary values ​​of each energy distribution range; wherein, the critical energy threshold is used to distinguish different wear states of machining tools.

[0172] The mapping table construction unit is used to define the energy boundaries corresponding to different machining tool wear states based on the critical energy threshold, construct the association between energy value ranges and different machining tool wear state labels, and store the association in the form of a lookup table to obtain the cumulative irreversible damage energy-wear state mapping table.

[0173] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the deep learning-based machining tool wear monitoring method described above.

[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0175] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0176] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for monitoring wear of machining tools based on deep learning, characterized in that, The method includes: S101. Acquire multi-source sensor data of machining tools, and convert the multi-source sensor data using physical formulas to generate a real-time energy flow feature sequence; S102. Input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; wherein, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time. S103. Based on the instantaneous power contribution sequence of each energy channel, the irreversibility confidence of each energy channel at each time moment is calculated by the time inversion consistency test method, and the instantaneous power contribution sequence and the irreversibility confidence corresponding to the instantaneous power contribution sequence are multiplied point by point to obtain the irreversible damage power sequence of each energy channel. S104. Integrate and sum the irreversible damage power sequences from the start of processing to the current time to generate the cumulative irreversible damage energy value. S105. Based on the cumulative irreversible damage energy value, the real-time wear state of the machining tool is obtained by mapping and identification through a predefined cumulative irreversible damage energy-wear state mapping table, and the wear amount prediction value of the machining tool is obtained by wear amount regression prediction calculation.

2. The method according to claim 1, characterized in that, The multi-source sensor data includes cutting force signals, electrical power signals, vibration signals, and acoustic emission signals; The process of acquiring multi-source sensor data from machining tools and converting the multi-source sensor data using physical formulas to generate a real-time energy flow feature sequence includes: S201. Obtain the orthogonal accelerations of the machining tool and the workpiece in the orthogonal direction, and based on the cutting force signal, calculate the instantaneous cutting power value at each moment through vector dot product to generate an instantaneous cutting power time series; wherein, the instantaneous cutting power time series is used to characterize the rate at which mechanical energy is used. S202. Filter the electrical power signal to obtain the net input electrical power time series; wherein, the net input electrical power time series is used to characterize the core basic intensity of energy input; S203. The vibration signal is numerically integrated to obtain the vibration velocity. Based on the vibration velocity and the preset equivalent mass of the tool-workpiece system, the instantaneous vibration energy value at each moment is calculated using the kinetic energy formula to generate an instantaneous vibration energy time series. The instantaneous vibration energy time series is used to characterize the instantaneous kinetic energy change of the machining tool due to vibration. S204. The acoustic emission signal is bandpass filtered in a specific frequency band to obtain a filtered acoustic emission signal. The filtered acoustic emission signal is then segmented into frames using a window of preset duration, and the root mean square (RMS) value of the filtered acoustic emission signal within each window is calculated. The RMS values ​​are then arranged in chronological order to obtain an acoustic emission energy flux density time series. The acoustic emission energy flux density time series is used to characterize the instantaneous energy release intensity of the machining tool damage. S205. Obtain the cutting parameter vector of the real-time machining state, and align and splice the instantaneous cutting power time sequence, the net input electrical power time sequence, the instantaneous vibration energy time sequence, the acoustic emission energy flow density time sequence, and the cutting parameter vector in the time dimension to generate the real-time energy flow feature sequence.

3. The method according to claim 2, characterized in that, The vibration signal is numerically integrated to obtain the vibration velocity. Based on the vibration velocity and a preset equivalent mass of the tool-workpiece system, the instantaneous vibration energy value at each moment is calculated using the kinetic energy formula to generate an instantaneous vibration energy time series, including: S301. Perform high-pass filtering on the vibration signal to obtain a pure vibration acceleration signal; S302. The trapezoidal integral method is used to numerically integrate the pure vibration acceleration signal to obtain a preliminary vibration velocity sequence. S303. Perform high-pass filtering on the preliminary vibration velocity sequence to obtain a vibration velocity time sequence; S304. Calculate the instantaneous vibration energy value at each of the stated moments using the kinetic energy formula; S305. The instantaneous vibration energy values ​​are removed from the outliers and arranged in chronological order to generate the instantaneous vibration energy time series.

4. The method according to claim 1, characterized in that, The expression for the cumulative irreversible damage energy value is: in, Indicates the end time The cumulative irreversible damage energy value, Indicates the total number of energy channels. Indicates the first Each energy channel at time The irreversible confidence level, Indicates the first Each energy channel at time The instantaneous power contribution sequence.

5. The method according to any one of claims 1 to 4, characterized in that, The construction process of the cumulative irreversible damage energy-wear state mapping table is as follows: S501. Obtain the historical cumulative irreversible damage energy value data and the actual wear state label corresponding to the cumulative irreversible damage energy value data during the entire life cycle of the machining tool wear. S502. Classify the historical cumulative irreversible damage energy value data according to the actual wear state label, and generate the energy distribution interval corresponding to each wear state. S503. Based on the boundary values ​​of each of the energy distribution intervals, a critical energy threshold is set; wherein, the critical energy threshold is used to distinguish different wear states of machining tools. S504. Based on the critical energy threshold, define the energy boundaries corresponding to different machining tool wear states, construct the correlation between energy value ranges and different machining tool wear state labels, and store the correlation in the form of a lookup table to obtain the cumulative irreversible damage energy-wear state mapping table.

6. A deep learning-based machining tool wear monitoring system, characterized in that, The system includes: The feature conversion module is used to acquire multi-source sensor data of machining tools, convert the multi-source sensor data through physical formulas, and generate a real-time energy flow feature sequence. The decoupling network module is used to input the real-time energy flow feature sequence into the trained energy dissipation decoupling network model to obtain the instantaneous power contribution sequence of each energy channel; wherein, the instantaneous power contribution sequence is used to characterize the power proportion and dynamic contribution law of each energy channel to the total energy dissipation of the machining tool at each time. The damage power generation module is used to calculate the irreversibility confidence of each energy channel at each time step based on the instantaneous power contribution sequence of each energy channel by using the time inversion consistency test method, and multiply the instantaneous power contribution sequence and the irreversibility confidence corresponding to the instantaneous power contribution sequence point by point to obtain the irreversible damage power sequence of each energy channel. The damage energy calculation module is used to integrate and sum the irreversible damage power sequences from the processing start time to the current time to generate a cumulative irreversible damage energy value. The state decision and prediction module is used to obtain the real-time wear state of the machining tool by mapping and identifying it through a predefined cumulative irreversible damage energy-wear state mapping table based on the cumulative irreversible damage energy value, and to obtain the predicted wear value of the machining tool by wear amount regression prediction calculation.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images