Numerical control machine tool wear automatic detection and compensation method based on artificial intelligence

Abstract

The invention provides a numerical control machine tool wear automatic detection and compensation method based on artificial intelligence, and the method comprises the steps: collecting the cutting force data of a high-curvature region in real time through multi-sensor fusion, and obtaining the cutting force fluctuation characteristics; cutting temperature data of the high-curvature area are monitored and obtained, the cutting temperature change rate is extracted, whether the temperature exceeds a preset threshold value or not is judged, and if yes, an alarm mechanism is triggered, and cutting parameters are adjusted; predicting the tool wear rate in combination with the co-evolution relationship between wear and temperature, the online monitoring data and the processed time, and generating a wear prediction curve in a preset time period; and performing trend analysis and feature extraction on the wear prediction curve to obtain wear parameter changes of the cutter in a preset time, and if the prediction curve shows that the wear parameter changes at a certain time point in the future exceed a preset critical value, adjusting the cutting parameters and generating a target cutting parameter combination.

Description

Technical Field

[0001] The present invention relates to the technical field of numerical control machine tools, and particularly to an automatic detection and compensation method for tool wear of a numerical control machine tool based on artificial intelligence. Background Art

[0002] In the machining process of a numerical control machine tool for a new type of CVT transmission thrust wheel, due to the complex groove-shaped surface of the thrust wheel and large differences in cutting parameters at different positions, the tool wear shows a special problem of local rapid loss. When continuous cutting is carried out in a high-curvature area, the tool faces a complex stress state and thermodynamics environment, which is prone to accelerating wear. In order to accurately control the machining quality, it is necessary to monitor the cutting process in real time and dynamically adjust the machining parameters. However, due to the geometric characteristics of the high-curvature area, it is difficult for traditional single sensors to accurately capture the changes in the cutting state. Although multi-sensor fusion technology can provide more comprehensive data, how to effectively integrate and analyze these data to achieve real-time evaluation of the tool state is still a challenge. In addition, even if the tool wear trend can be accurately predicted, how to adjust the cutting parameters while ensuring the machining accuracy is also a complex problem. The adjustment of the feed rate and cutting depth needs to consider multiple factors such as the workpiece material properties, the current state of the tool, and the machining requirements, and these factors will change dynamically as the cutting process progresses. Therefore, how to establish an intelligent machining system that can comprehensively consider multi-source information, respond in real time, and adaptively adjust to maximize the tool life and ensure the stability of machining quality in the high-curvature area has become a technical problem to be solved urgently. Summary of the Invention

[0003] The present invention provides an automatic detection and compensation method for tool wear of a numerical control machine tool based on artificial intelligence, mainly including:

[0004] Collect the cutting force data in the high-curvature area in real time through multi-sensor fusion to obtain the cutting force fluctuation characteristics;

[0005] Monitor and obtain the cutting temperature data in the high-curvature area, extract the cutting temperature change rate, and determine whether the temperature exceeds a preset threshold. If so, trigger an alarm mechanism and adjust the cutting parameters, where the cutting parameters include the tool feed rate and the cutting depth;

[0006] Collect the vibration signal in the cutting process in real time, and perform spectrum analysis to obtain the vibration frequency and amplitude changes of the tool. Combine the cutting force fluctuation characteristics and the cutting temperature data, and use the finite element analysis method to obtain the stress distribution, establish a relationship model between the cutting force, the cutting temperature, the tool vibration signal and the stress distribution, and determine the stress concentration position in the high-curvature area;

[0007] If continuous cutting is performed at the stress concentration position for a long time, evaluate the local tool wear rate and the change of wear morphology by combining the cutting force and vibration signal data of off-line and on-line monitoring, and analyze the co-evolution relationship between wear and temperature based on the historical cutting force, vibration signal and current cutting temperature data;

[0008] Combine the co-evolution relationship between wear and temperature, on-line monitoring data and the processed time to predict the tool wear rate, and generate a wear prediction curve for a preset time period;

[0009] Conduct trend analysis and feature extraction on the wear prediction curve to obtain the change of tool wear parameters within the preset time. If the prediction curve shows that the change of wear parameters at a certain future time point will exceed the preset critical value, adjust the cutting parameters to generate a target cutting parameter combination;

[0010] According to the target cutting parameter combination, identify the wear compensation area at the set control period, evaluate the tool wear state and the machining quality of the workpiece, calculate the compensation amount of the cutting parameters, and adjust the tool motion trajectory of the wear compensation area in real time according to the compensation amount.

[0011] Furthermore, the cutting force data of the high curvature area is collected in real time through multi-sensor fusion to obtain the cutting force fluctuation characteristics, including: obtaining the sensor coordinate matrix according to the position relationship of the multi-sensor array on the cutting surface, extracting the force signal sensitivity coefficient from the sensor calibration data record, generating the sensor data acquisition timestamp sequence for the sampling frequency, and performing spatial mapping calibration on the sensor raw data using the coordinate transformation matrix. Divide the data into blocks according to the sensor acquisition period and data cache threshold, extract the force signal timestamp identifier from the data block, use the sliding mean filter to eliminate the amplitude noise of the force signal, and perform time series alignment on multiple groups of sensing data based on the timestamp identifier. Set the curvature threshold for the cutting surface curvature gradient value, divide the cutting surface into high and low curvature areas using the curvature threshold, extract the corresponding sensor data sequence from the divided area, and calculate the wavelet coefficient of the force signal amplitude sequence using wavelet transform. Determine the data fusion weight coefficient according to the sensor layout position, use the weight coefficient to perform weighted summation on the wavelet coefficients of multiple sensors, and reconstruct the fused force signal sequence from the weighted result. Calculate the local variance value of the force signal sequence using a sliding window to obtain the cutting force volatility curve, set the clustering center based on the mean and standard deviation of the volatility curve, perform segmental clustering on the volatility data through Euclidean distance measurement, and extract the cutting force fluctuation feature vector from the clustering result.

[0012] Further, monitor and obtain the cutting temperature data in the high-curvature area, extract the change rate of the cutting temperature, and determine whether the temperature exceeds the preset threshold. If so, trigger the alarm mechanism and adjust the cutting parameters. The cutting parameters include the tool feed rate and the cutting depth, and the method includes: evenly dividing the machining area according to the curvature gradient value of the cutting surface, arranging a thermocouple sensor array at the grid nodes to collect temperature data, using cubic spline interpolation to generate the grid temperature field distribution map, and calculating the temperature field time series through the temperature sampling rate setting. Use the Bayesian filtering state equation to eliminate the noise of the temperature field data, use the temperature sensor calibration curve to compensate for the measurement error, calculate the temperature field gradient distribution through central difference, and judge the boundary of the high-curvature area based on the gradient threshold. Calculate the temperature change rate between adjacent sampling points according to the temperature field time series, use the sliding variance window to detect the temperature mutation point, obtain the temperature change rate curve from the mutation point, and extract the temperature fluctuation characteristics for the change rate curve. Compare the temperature fluctuation characteristics with the preset temperature threshold, locate the abnormal point coordinates when the temperature fluctuation amplitude exceeds the threshold range, and determine the temperature overlimit area range based on the abnormal point coordinates. Divide the feed rate adjustment interval according to the temperature overlimit area range, establish the cutting depth constraint condition using the temperature change rate, and calculate the optimal parameter combination based on the linear programming solver. Use the parameter combination to adjust the feed rate in segments, limit the feed amount through the cutting depth constraint, and obtain the tool motion trajectory from the adjusted parameter sequence.

[0013] Further, collect the vibration signal during the cutting process in real time, and perform spectrum analysis to obtain the vibration frequency and amplitude change of the tool. Combine the cutting force fluctuation characteristics and the cutting temperature data, use the finite element analysis method to obtain the stress distribution, and establish the relationship model between the cutting force, the cutting temperature, the tool vibration signal and the stress distribution, and determine the stress concentration position in the high-curvature area, including: collect the acceleration sensor data according to the vibration sampling frequency, obtain the vibration signal envelope through the Hilbert transform, calculate the vibration spectrum using the fast Fourier transform, extract the vibration frequency and amplitude peak value from the spectrum, and decompose the vibration signal into a low-frequency base band and a high-frequency harmonic band using wavelet packet decomposition. Use the force signal peak point in the cutting force fluctuation characteristic curve and the extreme value point of the temperature field change rate to construct the boundary condition, construct tetrahedral mesh elements for the cutting area according to the grid size parameter, assign the stress tensor boundary value to the grid nodes, and calculate the unit stress component based on the linear elastic constitutive equation. Calculate the root mean square value of the vibration signal according to the vibration signal envelope curve, calculate the standard deviation of the force signal through the cutting force fluctuation characteristic curve, calculate the temperature gradient vector using the temperature field change rate data, and construct the stress influence factor matrix using the three groups of data. Use the neural network to perform feature mapping on the stress influence factor matrix, train and establish the stress distribution predictor, obtain the stress field distribution function from the output of the predictor, and determine the stress concentration position using the stress field gradient.

[0014] Furthermore, if continuous cutting is performed at the stress concentration position for a long time, evaluate the local tool wear rate and the change in wear morphology by combining the cutting force and vibration signal data of off-line and on-line monitoring, and analyze the co-evolution relationship between wear and temperature based on the historical cutting force and vibration signals and the current cutting temperature data, including: setting a fixed acquisition period according to the cutting duration record, obtaining the time series of force signals and vibration signals during the historical cutting process from the off-line database, collecting the real-time cutting force and vibration signals from the on-line sensors, and using wavelet multi-scale decomposition to remove high-frequency noise. Construct a grid monitoring area for the stress concentration position, calibrate the initial wear position using the peak point of the cutting force fluctuation degree, identify the wear contour boundary through the vibration amplitude change curve, and calculate the wear area value according to the boundary contour. Calculate the local wear rate using the wavelet energy coefficient, analyze the wear morphology characteristics based on the vibration signal envelope spectrum, calculate the wear evolution speed using the cumulative wear amount, and obtain the wear development trend from the wear speed curve. Perform feature decomposition on the historical cutting force signal sequence using a long short-term memory network, extract the time series features of the vibration signal through the self-attention mechanism, and construct a temperature change function using the temperature acquisition time series. Construct a feature space according to the cutting parameter matrix, establish a mapping relationship between the wear rate and the temperature change function using a support vector regressor, and predict the coupled change of wear and temperature through a recurrent neural network. Establish a state transition matrix based on the coupled change law of wear and temperature, describe the co-evolution process of wear and temperature using a Markov chain, and obtain the evolution model parameters from the state probability distribution.

[0015] Further, obtain historical cutting force and vibration signal data, extract historical cutting force values and vibration values, collect the cutting temperature of the current monitoring system, record the current temperature value, and use the regression analysis method to establish a relationship model between the cutting force, vibration value, and temperature value. If the fitting degree of the relationship model meets the preset threshold, determine the co-evolution relationship between the wear amount and the temperature value, including: dividing the sampling time period according to the historical data cycle, extracting the peak-to-valley ratio, standard deviation of the force value, and root mean square feature of the force value from the cutting force sequence, calculating the amplitude mean, amplitude variance, and amplitude frequency feature through the vibration value sequence, and performing time alignment on the feature data using linear interpolation. Collect real-time temperature values for the temperature monitoring points, set the data recording period based on the temperature acquisition frequency, calculate the temperature average, temperature fluctuation amplitude, and temperature change slope from the temperature data sequence, and denoise the temperature data using the db4 wavelet basis function. Use the least squares method to establish a multiple linear equation set for the cutting force eigenvalue, vibration amplitude feature, and temperature feature, calculate the goodness of fit of the equation through the R-squared value, and construct a feature correlation matrix using the Pearson correlation coefficient. Screen the feature pairs with a significance coefficient greater than the preset value according to the feature correlation matrix, establish a feature mapping function using polynomial regression, determine the polynomial order through significance testing, and obtain the variable weight coefficient from the mapping function. Calculate the linear combination of the wear amount and the temperature value based on the variable weight coefficient, establish a time series prediction function through a recursive neural network. If the R-squared value of the prediction function exceeds the preset threshold, obtain the neural network weight matrix. Use the neural network weight matrix to construct a state transition equation, generate an evolution function according to the co-variation law of the wear amount and the temperature value, and obtain the coupling coefficient matrix from the evolution function.

[0016] Further, in combination with the co-evolution relationship between wear and temperature, online monitoring data, and the processed time, predict the tool wear rate and generate a wear prediction curve for a preset time period, including: segmenting the online collected data in time series according to the monitoring data period, statistically counting the historical processing amount from the processing time period, performing block processing on the data sequence with a fixed sampling interval, and performing linear interpolation on the temperature data to obtain an equally spaced sampling sequence. Discretize the prediction time span with a uniform time step, extract the root mean square value of the cutting force, vibration amplitude, and temperature gradient features for the discretized time points, and perform standardization processing on the feature sequence based on the long short-term memory network. Use the Pearson correlation coefficient to screen the correlation of the feature sequence, calculate the change rate of the feature sequence through central difference, calculate the predicted wear rate value according to the co-evolution model parameters, and extract the trend features from the wear rate sequence. Among them, the feature change rate calculated by central difference reflects the dynamic characteristics of the wear process. Establish a cubic spline function for the wear rate trend feature, determine the spline coefficients by the least squares method, and generate a continuous wear prediction curve according to the spline function. Perform noise filtering on the wear prediction curve based on the recurrent neural network, calculate the average wear rate through a sliding mean window, and set an early warning judgment threshold according to the wear rate standard deviation. Calculate the prediction error statistic using historical wear data, establish a prediction confidence interval through the error distribution function, and obtain the wear prediction accuracy evaluation index.

[0017] Further, perform trend analysis and feature extraction on the wear prediction curve to obtain the change in the tool wear parameters within a preset time. If the prediction curve shows that the change in the wear parameters at a certain future time point will exceed the preset critical value, adjust the cutting parameters to generate a target cutting parameter combination, including: dividing the wear prediction curve segment according to the wear prediction duration, using the zero point of the second derivative to determine the position of the trend inflection point, calculating the piecewise linear regression coefficient by the least square method to obtain the wear slope value, calculating the variance of the wear amount through a fixed-length sliding window to obtain the wear fluctuation degree sequence. Perform Fourier transform on the wear fluctuation degree sequence to extract the main frequency component, calculate the average wear amount per unit time from the wear prediction curve segment, extract the time-frequency spectrum of the wear characteristics based on the convolutional neural network. If the peak value of the characteristic spectrum exceeds the preset critical threshold, it is determined as the critical wear point. In another implementation manner, the extracted wear characteristics may further include the change trend of the wear rate, the average wear rate, and the fluctuation of the wear rate. Use a decision tree to construct a three-layer parameter classification structure of the feed rate, spindle speed, and cutting depth, determine the parameter optimization order through the maximum information gain criterion, set and adjust the weight coefficient according to the parameter sensitivity, and calculate the parameter adjustment amount from the weight coefficient matrix. Use the cutting parameter constraint function to perform boundary limitation on the parameter adjustment amount, use the particle swarm algorithm to perform optimization calculation on the parameter combination, and obtain the parameter sequence that meets the constraint conditions from the optimization result. Construct a cutting process evaluation function based on the parameter sequence, calculate the fitness of the parameter combination through weighted summation, select the optimal parameter combination according to the fitness ranking, and generate the target cutting parameters from the optimal combination. Perform stability verification on the target cutting parameters, use the Monte Carlo method to simulate the parameter perturbation, and obtain the parameter stable interval from the simulation result.

[0018] Further, calculate the current wear rate, adopt a time series analysis algorithm, construct a prediction curve model in combination with historical wear data, calculate the predicted value of the wear rate at a preset future time point. If the predicted value exceeds the preset critical value, call the parameter optimization algorithm to generate several sets of cutting parameter combinations, and search for the target cutting parameter combination within the preset parameter range, including: sampling the wear data with a fixed time window according to the historical data period, extracting the training data sequence from the historical data length, establishing a wear rate prediction model using a long short-term memory network, and calculating the prediction fitting accuracy through the root mean square error. Calculate the time domain statistics for the wear rate sequence, extract the frequency domain feature coefficients using wavelet transform, construct a time series state equation based on a recurrent neural network, and calculate the predicted value of the wear rate at the preset time point from the state equation. Calculate the warning level according to the difference between the wear rate predicted value and the critical wear value. If the warning level exceeds the preset threshold, trigger the parameter optimization link. Construct a feed rate interval, a cutting depth interval, and a spindle speed interval based on the preset parameter range, discretize the parameter interval using the uniform quantization method, and screen the parameter combinations according to the machining process constraints. Evaluate the parameter combinations using a cutting load constraint function, set the parameter upper limit according to the cutting temperature limit, determine the parameter lower limit through the machining accuracy requirements, and obtain the feasible parameter space from the constraint conditions. Use the genetic algorithm to search the feasible parameter space, control the parameter combination method through the crossover probability, adjust the parameter change range based on the mutation probability, and obtain the optimal cutting parameter combination from the search results.

[0019] Further, according to the target cutting parameter combination, at the set control period, identify the wear compensation area to be compensated, evaluate the wear state of the tool and the machining quality of the workpiece, so as to calculate the compensation amount of the cutting parameters, and adjust the tool motion trajectory of the wear compensation area in real time according to the compensation amount, including: execute the target cutting parameter combination in segments according to the control period length, arrange the wear detection array in a grid pattern, extract the wear profile features from the detection image using a convolutional neural network, and measure the surface roughness value of the machined surface with a laser profilometer. Extract a set of geometric parameters for the wear profile features, including the wear area, wear depth, and wear bandwidth, dynamically estimate the wear parameters using Kalman filtering, and construct a wear state vector from the estimation results. Calculate the parameter compensation increment based on the wear state vector, generate the feed rate correction amount and the cutting depth correction amount using a linear compensation function, and obtain the optimal compensation parameter combination through a quadratic programming solver. Smooth the compensation parameters using cubic spline interpolation, adjust the feed axis speed command according to the feed rate correction amount, and calculate the tool position compensation amount using the cutting depth correction amount. Establish a compensation trajectory equation for the tool position compensation amount, fit the trajectory curve using a fifth-order polynomial, and generate a position compensation instruction sequence from the fitted curve. Execute the compensation instruction sequence through a position feedforward controller, calculate the actual compensation error using the servo feedback signal, and adjust the compensation gain coefficient from the error signal.

[0020] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0021] The present invention discloses an artificial intelligence-based automatic detection and compensation method for tool wear in a numerically controlled machine tool. Aiming at the problem that it is difficult to accurately control tool wear during the cutting process in a high-curvature area, the present invention collects cutting force, temperature, and vibration data in real time through multi-sensor fusion, establishes a relationship model between cutting parameters and stress distribution, and locates the stress concentration position. Combining online monitoring data and historical data, a co-evolution model of wear and temperature is constructed to predict the tool wear rate and generate a wear prediction curve. Based on the prediction results, the present invention dynamically adjusts the cutting parameters and compensates the tool motion trajectory in real time to achieve intelligent control of tool wear in the high-curvature area. This method can effectively improve the machining accuracy in the high-curvature area, extend the tool life, reduce machining errors, and improve machining efficiency and product quality. Brief Description of the Drawings

[0022] Figure 1 It is a flowchart of an artificial intelligence-based automatic detection and compensation method for tool wear in a numerically controlled machine tool according to the present invention.

[0023] Figure 2 It is a schematic diagram of an artificial intelligence-based automatic detection and compensation method for tool wear in a numerically controlled machine tool according to the present invention. Detailed Embodiments

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0025] Such as Figure 1-2 , an artificial intelligence-based automatic detection and compensation method for tool wear in a numerically controlled machine tool in this embodiment may specifically include:

[0026] S101. Collect cutting force data in the high-curvature area in real time through multi-sensor fusion technology and extract the cutting force fluctuation characteristics. Obtain the coordinate matrix according to the spatial arrangement of the sensor array, perform spatial mapping calibration on the collected data using coordinate transformation to eliminate noise, and then divide the area based on the curvature threshold and reconstruct the force signal sequence through wavelet analysis and weighted fusion, and calculate the volatility curve to determine the cutting force fluctuation characteristics.

[0027] S1011. During the cutting process in the high-curvature region, use a multi-sensor array to collect cutting force data and generate a sensor coordinate matrix, where the sensors are evenly distributed along the edge of the cutting surface to form an annular monitoring network. Extract the sensitivity coefficient of each sensor and record the spatial position according to the calibration test. Map the original data to a unified coordinate system through coordinate transformation. Subsequently, use a moving average filter to eliminate noise from the mapped data and complete the temporal alignment of multiple groups of data according to the time stamp.

[0028] S1012. Set a threshold according to the curvature gradient of the cutting surface and divide it into high- and low-curvature regions. Extract the force signal sequence from the high-curvature region and calculate the wavelet coefficients using wavelet transform. Subsequently, determine the fusion weight coefficients according to the sensor positions and perform weighted summation on the wavelet coefficients to reconstruct the force signal sequence. Calculate the local variance through a sliding window to generate the cutting force volatility curve. Set the clustering centers based on the statistical characteristics of the volatility curve and extract the cutting force fluctuation feature vector through Euclidean distance measurement.

[0029] In the embodiment of the present invention, the change of the cutting force is monitored in real time through a multi-sensor array to ensure the comprehensiveness of the data in the high-curvature region. The sensor array usually consists of 8 piezoelectric force sensors, which are distributed at a 45-degree angle along the cutting surface. The sampling frequency is set to 1000 Hz, and each data block contains 512 sampling points. The sensitivity coefficient ranges from 0.95 to 1.05 in the calibration. The filtering uses a 15-point moving average process, and the alignment accuracy is better than 0.1 millisecond. The wavelet transform uses the db4 basis function for 4-layer decomposition. The weight coefficients are assigned according to the distance between the sensor and the high-curvature region, with a maximum value of 0.3 and a minimum value of 0.05. Further, 3 clustering centers are set according to the mean and standard deviation of the volatility curve, which respectively represent stable, transitional, and unstable cutting states. The feature vector is extracted through cluster analysis. Experiments show that the volatility in the high-curvature region is significantly higher than that in the low-curvature region, and the features are repetitive, reflecting the influence of curvature on cutting stability, providing a reliable data basis for subsequent wear assessment. The implementation method of S101 is not limited to the specific parameters described above and can be adjusted according to the actual machining scenario, such as the number of sensors, sampling frequency, or filtering window size, etc., to meet the requirements of different working conditions.

[0030] S102. Monitor and obtain the cutting temperature data in the high-curvature region and extract the temperature change rate feature. Judge whether the temperature exceeds the preset threshold and trigger an alarm and parameter adjustment when it exceeds the limit. Collect the temperature data through grid division and a thermocouple array, generate the temperature field distribution and perform noise processing. Subsequently, calculate the temperature gradient and mutation feature, and optimize the cutting parameters based on linear programming to control the tool feed speed and cutting depth.

[0031] S1021. Uniformly divide the machining area into grids according to the cutting surface curvature gradient value, arrange a thermocouple sensor array at the nodes to collect temperature data, use the cubic spline interpolation technique to perform spatial reconstruction on the discrete temperature data to generate a continuous grid temperature field distribution map, calculate the temperature field time series by setting the sampling frequency, and use the Bayesian filtering state equation to eliminate measurement noise. At the same time, combine the sensor calibration curve to compensate for errors to improve data accuracy.

[0032] S1022. Calculate the temperature gradient distribution through the central difference method for the grid temperature field distribution map and determine the boundary of the high curvature area. Extract the temperature change rate of adjacent sampling points from the temperature field time series and use a sliding variance window to detect temperature mutation points. Generate a temperature change rate curve according to the mutation points and extract the fluctuation characteristics. Then compare the fluctuation characteristics with the preset temperature threshold to locate the temperature overrun area and divide the feed speed adjustment interval.

[0033] S1023. Establish cutting depth constraint conditions based on the range of the temperature overrun area and combine the temperature change rate characteristics. Use a linear programming solver to calculate the optimal cutting parameter combination and perform segmented adjustment of the feed speed. At the same time, limit the feed amount through the cutting depth constraint and generate an adjusted tool motion trajectory.

[0034] In the embodiment of the present invention, providing a basis for process parameter adjustment by real-time monitoring of cutting temperature data. For the complex thermodynamic environment in the high curvature area, a grid division scheme with a 10-mm spacing is adopted, and a K-type thermocouple sensor is arranged at each node. The temperature measurement range covers 0 to 800 degrees Celsius, and the sampling frequency is set to 100 Hz. The temperature field is reconstructed using cubic spline interpolation. Experimental verification shows that the interpolation error is less than 2 degrees Celsius, ensuring the reliability of the distribution map. The Bayesian filtering describes the temperature evolution through the state equation and performs filtering in combination with an observation model with a noise standard deviation of 0.5 degrees Celsius. The calibration curve is obtained by piecewise fitting in a constant temperature oil bath, and the error compensation accuracy is better than 1 degree Celsius in the full temperature range.

[0035] Furthermore, the 5-point central difference is adopted to calculate the temperature gradient, and the threshold is set at 50 degrees Celsius per millimeter to identify high-curvature boundaries. The temperature change rate is calculated by the difference between adjacent points in the time series. A 32-point sliding variance window is used to detect mutation points. When the variance exceeds 10, it is determined as abnormal. The generated change rate curve shows that the maximum value in the high-curvature area can reach 120 degrees Celsius per second, which is much higher than 40 degrees Celsius per second in the flat area. If the amplitude of temperature fluctuation exceeds the preset threshold, the abnormal coordinates are located and the over-limit area is divided. The feed speed adjustment range is set at 500 to 2000 millimeters per minute, and the cutting depth is restricted to 0.1 to 2 millimeters. The linear programming solver aims at temperature stability and calculates the parameter combination in combination with the speed continuity constraint. In the area where the curvature radius is less than 5 millimeters, experiments show that reducing the feed speed to 40% of the initial value and adjusting the cutting depth to less than 0.3 millimeters can effectively inhibit the temperature rise. The adjustment process adopts a piecewise smoothing strategy, and the speed change rate is limited within 200 millimeters per minute to avoid processing instability caused by parameter mutation. The cutting depth is adjusted in real time through the compensation amount positively correlated with the temperature change rate. The optimized tool path significantly reduces the heat accumulation in the high-curvature area, improving the processing quality and tool life. For the real-time processing of temperature overrun, the alarm mechanism is triggered immediately after detecting the mutation point, and the optimized parameters are transmitted to the numerical control system through the preset control cycle to ensure timely adjustment of the processing process. Experimental data show that the temperature control effect in the high-curvature area is better than the traditional static parameter scheme, and the temperature distribution uniformity is improved by about 15%, providing reliable support for subsequent wear prediction and compensation.

[0036] S103. Vibration signals during the cutting process are collected in real time, and the tool vibration frequency and amplitude change characteristics are extracted through spectrum analysis. Combining the cutting force fluctuation data and cutting temperature information, a stress distribution model is constructed using finite element analysis. The relationships between cutting force, temperature, vibration and stress are established, and the stress concentration positions in the high-curvature area are determined. The stress field calculation is optimized through signal processing and deep learning techniques to improve the positioning accuracy.

[0037] In the embodiment of the present invention, a piezoelectric acceleration sensor is used to collect vibration signals during the cutting process in real time at a sampling frequency of 20,000 Hz. The envelope curve of the vibration signal is generated through Hilbert transform, and the spectrum data is calculated using fast Fourier transform. The frequency and amplitude characteristics with a fundamental frequency lower than 500 Hz and high-frequency harmonics in the range of 2000 to 5000 Hz are extracted from it. Subsequently, the signal is divided into two frequency bands, namely low-frequency fundamental wave and high-frequency harmonic wave, through wavelet packet decomposition to enhance the feature resolution. The cutting force fluctuation data provides periodic peak information through the force signal sequence obtained in the previous step, and the temperature data reflects the heat source distribution in the form of a gradient vector. The three together serve as the input basis for stress analysis.

[0038] S1031. Calculate the root mean square value according to the vibration signal envelope curve, extract the standard deviation from the cutting force fluctuation characteristic curve, and generate the temperature gradient vector by using the temperature field time series. Construct the stress influence factor matrix with these three groups of data and standardize it to a unified dimension. Subsequently, assign boundary conditions to the tetrahedral mesh elements based on the linear elastic constitutive equation and calculate the element stress components to generate the initial stress distribution data.

[0039] S1032. Input the stress influence factor matrix into the neural network model based on deep learning. Generate the stress field distribution function through feature mapping and determine the stress concentration positions in the high curvature regions. Calculate the magnitude and direction of the principal stress by using the stress field gradient and evaluate the concentration degree in combination with the stress intensity criterion. Subsequently, optimize the positioning accuracy of the stress singular points through the mesh encryption technology and construct the stress hot spot regions to analyze the crack propagation trend.

[0040] In the embodiment of the present invention, the calculation reliability of the stress distribution is improved through multi-source data fusion. The root mean square value of the vibration signal characterizes the overall strength and maintains between 0.5 and 1.5 g during stable cutting. The standard deviation of the cutting force reflects the discreteness and ranges from 50 to 80 N. The temperature gradient vector changes with the cutting direction, and the included angle with the cutting direction is close to 90 degrees in the high curvature regions. After these data are integrated in matrix form, they are input into the finite element model as boundary conditions. The tetrahedral mesh adopts an adaptive division strategy, and the size is reduced to 0.1 mm in the stress concentration regions to ensure the accuracy. The element stress components are calculated by the linear elastic equation, considering the material stiffness and deformation characteristics.

[0041] Furthermore, use a 5-layer convolutional neural network to extract features from the matrix, and generate a stress distribution predictor through the training of 3000 groups of experimental data. The stress field distribution function output by the predictor shows that the included angle between the principal stress direction and the cutting direction in the high curvature regions is between 75 and 85 degrees, and the principal stress value is significantly higher than that in the plane regions, about 2 to 3 times higher. The stress intensity factor is close to 80% of the material fracture toughness at the singular points, indicating a relatively high risk of crack propagation. Therefore, the regions with the curvature change rate greater than a certain threshold are encrypted with the mesh, the element size is reduced to 0.05 mm, the coordinates of the singular points are accurately located to the micron level through the stress field gradient, and the crack propagation rate is calculated based on fracture mechanics to evaluate the regional stability.

[0042] In practical applications, it is found that the coupling effect between the temperature gradient and the stress direction significantly affects the stress distribution in the high-curvature region, especially when the radius of curvature is less than 5 mm. This coupling effect is quantified through the construction of stress hot spots, and the stress concentration degree in the hot spot region is closely related to the cutting conditions. To improve the analysis accuracy, the wavelet packet decomposition adopts a 4-layer structure, divides the signal into 16 frequency bands, and the extracted harmonic features further enrich the expression ability of vibration data. Experiments show that the stress model combining multi-source data can better reflect the changes in the machining state than single-signal analysis.

[0043] S104. If the stress concentration position in the high-curvature region undergoes long-term continuous cutting, evaluate the local wear rate and morphological changes of the tool by combining the historical cutting force and vibration signals in the offline database and the real-time data monitored online. Construct a co-evolution model of wear and temperature through multi-scale signal processing and feature extraction to reveal its coupling law, and use a neural network to predict the wear trend to provide a basis for subsequent compensation.

[0044] In the embodiment of the present invention, the force signal and vibration signal during the cutting process are collected at fixed intervals. The offline data comes from a database storing more than 500 sets of machining records, and the real-time data is collected by an online sensor at a frequency of 1000 Hz. For the high-frequency interference in the signal, 4-layer wavelet multi-scale decomposition is used to remove the noise above 2000 Hz and retain the main frequency bands reflecting the cutting state. Subsequently, a grid monitoring area with a spacing of 0.1 mm is constructed at the stress concentration position. The initial wear point is calibrated using the peak value fluctuation of the cutting force, and the wear boundary is identified through the vibration amplitude change curve. The wear area is calculated and its evolution trend is analyzed. In the experiment, the peak value of the cutting force fluctuates between 350 and 400 N, the vibration amplitude increases from 0.5 g to 2.5 g, and the wear area can reach 0.8 square millimeters.

[0045] S1041. Extract feature data from the time series of the denoised cutting force and vibration signals. Use indicators such as the peak-to-valley ratio, standard deviation, and average amplitude of the force value to characterize the signal characteristics. Achieve time alignment through linear interpolation. Subsequently, use the db4 wavelet basis function for further denoising and calculate the temperature change slope and average value. Establish a multiple linear regression model based on the least squares method and optimize the feature mapping function through significance testing to improve the fitting accuracy.

[0046] S1042. Calculate the local wear rate for the wear boundary points and extract the morphological features through the envelope spectrum of the vibration signal. Quantify the change in the frequency band distribution using the wavelet energy coefficient, and combine the long short-term memory network and self-attention mechanism to decompose the temporal features of the historical signal. Subsequently, establish a mapping relationship between the wear rate and temperature based on the support vector regressor and the recurrent neural network and predict the coupling change trend.

[0047] S1043. Construct a state transition matrix based on the prediction results and use the Markov chain to describe the co-evolution process of wear and temperature, quantify the mutual influence between wear evolution and temperature accumulation, and provide a dynamic basis for tool condition monitoring.

[0048] In the embodiment of the present invention, cutting force features are extracted from historical data. For example, the peak-to-valley ratio of the force value remains between 1.2 and 1.5 in the stable state, and the standard deviation fluctuates between 50 and 80 N. Combining the average amplitude of the vibration signal rising from 0.5 g to 2.0 g and the peak change of the frequency feature in the range of 2000 to 3000 Hz, the time axes of multi-source data are aligned by linear interpolation. The temperature data is collected at a frequency of 5 Hz, and the average value rises from 200 degrees Celsius to 550 degrees Celsius, and the change slope shows a phased jump. The multiple linear regression model contains 9 variables, the R-squared value reaches 0.92, and the Pearson correlation coefficient shows that the correlation between cutting force and temperature is as high as 0.85. The significance threshold is set to 0.75, and 5 groups of key feature pairs are selected. The mapping function uses a third-order polynomial regression, and the weight coefficient shows that the influence proportion of the cutting force reaches 0.45, reflecting its dominant role in wear.

[0049] Furthermore, the sideband modulation in the frequency band of 2000 to 3000 Hz is analyzed using the vibration envelope spectrum. The modulation degree increases with the deepening of wear, and the high-frequency energy proportion rises from 15% to 35%. The initial local wear rate is 0.02 mm / min, and it shows exponential growth later, reaching 0.08 mm / min. The long short-term memory network configures 128 hidden layer nodes, combines the self-attention mechanism to extract long-range dependence features, the support vector regressor maps the relationship between wear and temperature with a radial basis kernel function, and the recursive neural network predicts that the wear rate increases by 0.015 mm / min for every 100 degrees Celsius increase in temperature. The state transition matrix defines four states: normal, accelerated, severe, and failure. The calculation of the Markov chain shows that the transition probability is positively correlated with the temperature change rate. Experimental verification shows that when the temperature exceeds 550 degrees Celsius, the tool enters the severe wear stage, revealing the co-evolution mechanism of wear and temperature in the high-curvature region. The long-time cutting experiment shows that when the machining exceeds 30 minutes, the energy proportion in the vibration frequency band exceeds 40%, the peak cutting force reaches 450 N, and the remaining life is less than 5 minutes, and the prediction error is controlled within 10%. This dynamic modeling method can better adapt to complex working conditions compared with traditional static analysis, laying a foundation for the real-time monitoring and optimization adjustment of tool wear.

[0050] In practical applications, the acquisition period and model parameters can be adjusted according to the machining material and cutting conditions. For example, for carbide machining, the period can be shortened to 5 minutes, or the number of hidden layer nodes can be increased to improve the prediction accuracy to ensure the adaptability of the model to different scenarios. The construction of the evolution model can also provide feedback for tool design, optimizing the cutting edge geometry to slow down the wear rate.

[0051] S105. Input the online monitoring data and the processed time into the co-evolution model of wear and temperature to predict the tool wear rate and generate a wear prediction curve for a preset time period. Extract the feature sequence through time-series segmentation processing and optimize the prediction result using a neural network.

[0052] In the embodiment of the present invention, the online monitoring data is collected at a cycle of 30 seconds, and each cycle contains 600 sampling points. The machining time statistics show that the cumulative cutting time of a single tool reaches 120 minutes, and the proportion of the high-curvature area is about 35%. Perform time-series segmentation processing on the data. Divide the data into blocks through a fixed sampling interval and resample the temperature sequence into 1Hz equally-spaced data using linear interpolation to ensure the consistency of the time axis. The prediction time span is set to 30 minutes, discretized with a 1-minute step size, and extract the root mean square value of the cutting force, vibration amplitude, and temperature gradient features, which respectively reflect the energy of the force signal, vibration intensity, and heat distribution characteristics. Subsequently, eliminate the dimension difference through the normalization processing of the long short-term memory network.

[0053] S1051. Extract the historical machining volume statistical value from the segmented data and calculate the feature sequences of the root mean square value of the cutting force, vibration amplitude, and temperature gradient. Analyze the coupling relationship between the features using the Pearson correlation coefficient and generate a change rate sequence through central difference. Subsequently, predict the wear rate based on the co-evolution model parameters and extract the trend features to quantify the wear evolution process.

[0054] S1052. Construct a cubic spline function for the trend feature of the wear rate and optimize the fitting coefficient through the least squares method to generate a continuous prediction curve. Subsequently, use a recursive neural network for filtering processing to suppress noise and calculate the average value sequence of the wear rate through a sliding mean window. At the same time, set the warning threshold in combination with the standard deviation to achieve real-time monitoring.

[0055] In the embodiment of the present invention, the extracted root mean square value of the cutting force remains between 300 and 400N during normal cutting, the vibration amplitude range is 0.5 to 2.0g, and the temperature gradient can reach 80 degrees Celsius per millimeter in the high-curvature area. The calculation of the Pearson correlation coefficient shows that the correlation between the cutting force and the wear rate is 0.85, the correlation of the vibration amplitude is 0.78, and the correlation of the temperature gradient is as high as 0.92, indicating that the temperature has a particularly significant impact on wear. The change rate calculated by central difference reveals the dynamic characteristics of the wear acceleration stage. For example, the temperature change rate increases from 0.5 degrees Celsius per minute to 2.0 degrees Celsius per minute, reflecting the catalytic effect of heat accumulation on wear.

[0056] The wear rate curve is fitted using a cubic spline function, with 20 control points set and the coefficients optimized by the least squares method to keep the fitting error within 5%. The predicted curve exhibits three-stage characteristics: the initial wear rate is stable at 0.02 mm / min, gradually rises to 0.05 mm / min in the middle stage, and shows an exponential growth trend in the later stage. The recurrent neural network filters with a 32-point sliding window to smooth high-frequency fluctuations, uses the average value sequence as the benchmark, and takes three times the standard deviation as the warning threshold. Experimental verification shows that when the wear rate exceeds the threshold, the warning signal can indicate the risk of severe wear 5 to 8 minutes in advance, covering 90% of the test cases.

[0057] Furthermore, based on the analysis of 200 sets of historical data for prediction errors, the error distribution is approximately normal, with a mean of 0 and a standard deviation of 0.008 mm / min. The 95% confidence interval is plus or minus 0.016 mm / min from the predicted value. The short-term prediction accuracy reaches 95%, 85% in the middle term, and remains above 75% in the long term. This multi-scale prediction ability adapts to the requirements of different processing stages and provides a reliable basis for process optimization. In practical applications, the sampling period and feature extraction strategy can be adjusted according to the tool type or workpiece material. For example, when machining hard materials, the step size can be shortened to 15 seconds to capture more subtle changes, or the number of control points can be increased to improve the curve accuracy, ensuring the robustness of the prediction model under complex working conditions. By dynamically predicting the wear trend, the tool life is effectively extended and the risk of processing interruption is reduced.

[0058] S106. Conduct trend analysis and feature extraction on the wear prediction curve to obtain the change trend of the tool wear parameters within a preset time. If the prediction shows that the wear parameters exceed the preset critical value at a certain future time point, optimize the cutting parameters through a multi-level algorithm and generate a target combination, thereby realizing the dynamic adjustment of the machining process.

[0059] In the embodiment of the present invention, the wear prediction curve is divided with a 30-minute prediction duration, and each 5 minutes is used as an analysis segment. The trend inflection point is located by calculating the zero points of the second derivative. For example, in the interval of 15 to 20 minutes, the wear slope rises from 0.02 mm / min to 0.05 mm / min, indicating the start of accelerated wear. The least squares method is used to calculate the linear regression coefficients of each segment to quantify the slope change, and the variance of the wear amount is calculated through a 32-point sliding window to generate a sequence reflecting the fluctuation situation. The Fourier transform extracts the main frequency components of the fluctuation sequence, which are concentrated in the range of 0.05 to 0.1 Hz, revealing the periodic characteristics of wear. Subsequently, a convolutional neural network is used to process the main frequency data to generate a time-frequency spectrum. When the proportion of high-frequency components increases from 15% to 35%, it is determined that the near-critical wear state is approaching.

[0060] S1061. Construct a decision tree classification structure based on the time-frequency spectrum and optimize the cutting parameter sequence through the maximum information gain criterion. The decision tree has three layers, namely feed rate, spindle speed, and cutting depth in sequence. Allocate weights using parameter sensitivity and calculate the adjustment amount in combination with process constraints. Subsequently, use the particle swarm optimization algorithm to find the optimal parameter combination that meets the conditions.

[0061] Extract the training sequence from historical data, construct a wear rate prediction model using a long short-term memory network, configure 32 hidden layer nodes, extract frequency domain features through wavelet transform, and calculate the wear rate for the next 15 minutes using a recursive neural network. The predicted value increases from 0.02 mm / min to 0.07 mm / min. The critical value is set at 0.08 mm / min, and optimization is triggered when the predicted value exceeds 80%. The optimization range includes a feed rate of 500 to 2000 mm / min, a cutting depth of 0.1 to 2 mm, and a spindle speed of 2000 to 8000 rpm. The parameter interval is quantified by 10 discrete points. Decision tree analysis shows that the weight of the feed rate is 0.5, the spindle speed is 0.3, and the cutting depth is 0.2. The constraint conditions ensure that the ratio of the feed to the speed is between 0.1 and 0.4, and the cutting load does not exceed the critical value. The particle swarm optimization algorithm iterates 100 generations with 50 particles to generate the optimal combination: a feed rate of 1200 mm / min, a spindle speed of 4000 rpm, and a cutting depth of 0.8 mm.

[0062] S1062. Construct a cutting process evaluation function based on the parameter sequence and calculate the fitness value through weighted summation. The evaluation indicators include surface roughness, machining efficiency, and tool life, with weights of 0.4, 0.3, and 0.3 respectively. Subsequently, use the Monte Carlo method to verify the parameter stability and extract the stable interval from the simulation results to ensure the optimization effect. In the embodiment of the present invention, the adaptive adjustment of parameters is achieved through multi-level analysis and optimization. The calculation of the process evaluation function shows that the fitness of the optimal combination is 0.85, which is significantly improved compared to the initial value of 0.65. The Monte Carlo simulation conducts 1000 perturbation tests, with a 10% perturbation of the feed rate, a 5% perturbation of the spindle speed, and a 15% perturbation of the cutting depth. The index fluctuation is controlled within 5%, verifying the robustness. Experiments show that the optimized parameters reduce the wear rate by 30%, reduce the fluctuation amplitude by 40%, extend the tool life by 25%, and maintain the high-frequency component ratio below 20%, ensuring machining stability.

[0063] S107. Identify the wear compensation area to be compensated and evaluate the tool wear state and workpiece machining quality according to the target cutting parameter combination at a fixed control cycle, and adjust the tool motion trajectory in real time by calculating the compensation amount, so as to ensure machining accuracy and stability.

[0064] In the embodiment of the present invention, the control period is set to 1 second, the target cutting parameters are segmented into 50 segments, wear data is collected through a 5×5 grid detection array, and a 64×64 pixel image is generated for each point. The convolution neural network is used to process the image. The network is configured with 3 convolutional layers to extract the wear contour features and output the key point coordinates. Subsequently, the surface roughness is measured by a laser profiler with a resolution of 10 microns, and geometric parameters such as wear area, depth, and bandwidth are obtained. In the experiment, the wear area increases from 0.2 square millimeters to 0.8 square millimeters, the depth increases from 0.05 millimeters to 0.15 millimeters, and the bandwidth expands from 0.3 millimeters to 0.9 millimeters. The Kalman filter is used to dynamically estimate the wear parameters, a state vector is constructed, and the optimal compensation combination, including the feed rate and cutting depth correction amount, is calculated through a quadratic programming solver.

[0065] S1071. Extract the wear contour features from the detected image and calculate the set of geometric parameters. Use the Kalman filter to dynamically estimate the wear area and depth to generate a state vector. Subsequently, calculate the feed rate and cutting depth correction amounts based on the linear compensation model and the quadratic programming solver to optimize the parameter combination.

[0066] The wear data is processed by the Kalman filter. The second-order state equation describes the parameter changes. The standard deviation of the observation noise is 0.01 millimeters, and the state noise is 0.005 millimeters. The compensation increment calculation is based on a linear model. For example, when the wear area increases by 0.1 square millimeters, the feed rate is reduced by 50 millimeters per minute, and the cutting depth is reduced by 0.05 millimeters. The quadratic programming solver limits the feed rate correction range to plus or minus 200 millimeters per minute, and the cutting depth to plus or minus 0.2 millimeters. At the same time, considering the processing efficiency constraint, ensure that the parameter adjustment takes into account both quality and efficiency. Cubic spline interpolation is used to smooth the compensation parameters with 4 control points to ensure the continuity of the change.

[0067] S1072. Establish a compensation trajectory equation according to the correction amount and generate a position compensation instruction sequence through fifth-order polynomial fitting. Use the position feedforward controller to execute the instructions and adjust the compensation gain in combination with the servo feedback signal, so as to realize the real-time optimization of the tool path.

[0068] In the embodiment of the present invention, the tool movement is adjusted through the compensation trajectory equation. The fifth-order polynomial is fitted by the least squares method to ensure the continuity of the acceleration. The instruction sequence is output to the servo system at intervals of 5 milliseconds, and the feed axis speed response time is controlled within 50 milliseconds. The Z-axis positioning accuracy is better than 0.005 millimeters. In the transition region, the compensation amount changes gradually according to the sine function to avoid impact. The initial gain is set to 0.8 and adjusted online according to the error signal. The root mean square error is controlled within 0.01 millimeters. Experiments show that after compensation, the surface roughness is stabilized at 1.2 microns, the dimensional accuracy is improved by 40%, the wear rate is reduced by 35%, and the tool life is extended by 2 times.

[0069] As described above, this is only the specific implementation manner of this specification. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein. It should be understood that the protection scope of this specification is not limited thereto. Any person skilled in the art within the technical scope disclosed in this specification can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this specification.

Claims

1. An artificial intelligence-based automatic detection and compensation method for CNC machine tool wear, characterized in that: The method comprises: Through multi-sensor fusion, the cutting force data of high curvature area is collected in real time to obtain the cutting force fluctuation characteristics; Monitor and obtain cutting temperature data in high curvature areas, extract the cutting temperature change rate, and determine whether the temperature exceeds the preset threshold. If so, trigger an alarm mechanism and adjust cutting parameters, including tool feed speed and cutting depth; The vibration signal during the cutting process is collected in real time, and the spectrum analysis is performed to obtain the vibration frequency and amplitude changes of the tool. The stress distribution is obtained by combining the cutting force fluctuation characteristics and cutting temperature data using the finite element analysis method, and a relationship model between the cutting force, cutting temperature, tool vibration signal and stress distribution is established to determine the stress concentration position in the high curvature area. If the stress concentration position is cut continuously for a long time, the local tool wear rate and wear morphology changes are evaluated by combining the cutting force and vibration signal data monitored offline and online, and the co-evolution relationship between wear and temperature is analyzed based on the historical cutting force and vibration signals and the current cutting temperature data; The wear prediction curve for a preset time period is generated by combining the co-evolution relationship between wear and temperature, online monitoring data and the tool wear rate predicted by the processing time. Perform trend analysis and feature extraction on the wear prediction curve to obtain the wear parameter changes of the tool within the preset time. If the prediction curve shows that the wear parameter changes will exceed the preset critical value at a certain time point in the future, adjust the cutting parameters to generate the target cutting parameter combination; According to the target cutting parameter combination and the set control cycle, the wear area to be compensated is identified, the wear state of the tool and the machining quality of the workpiece are evaluated, and the compensation amount of the cutting parameters is calculated. According to the compensation amount, the tool motion trajectory of the wear area to be compensated is adjusted in real time.

2. The method according to claim 1, characterized in that: The cutting force data of the high curvature area is collected in real time by multi-sensor fusion to obtain the cutting force fluctuation characteristics, including: Acquire a sensor coordinate matrix according to the positional relationship of the sensor on the cutting surface, and perform spatial mapping calibration on the sensor collected data through coordinate transformation; The sliding mean filter is used to eliminate the noise of the sensor data after spatial mapping calibration, and the time sequence alignment of multiple groups of sensor data after filtering is performed through the timestamp mark; For the sensor data after time series alignment, the high and low curvature areas are divided according to the cutting surface curvature threshold, and the force signal amplitude sequence is extracted to obtain the wavelet coefficients. The data fusion weight coefficient is determined according to the sensor arrangement position, and the wavelet coefficient is weighted and reconstructed by using the weight coefficient to obtain the force signal sequence. The force signal sequence is calculated through a sliding window to obtain the cutting force fluctuation characteristics.

3. The method according to claim 1, characterized in that The monitoring and acquisition of cutting temperature data in the high curvature area, the extraction of the cutting temperature change rate, and the determination of whether the temperature exceeds a preset threshold value, if so, triggering an alarm mechanism and adjusting the cutting parameters, the cutting parameters including the tool feed speed and cutting depth, include: The processing area is gridded according to the curvature gradient value of the cutting surface, the temperature data collected by the thermocouple sensor array is obtained from the grid nodes, and the grid temperature field distribution map is obtained by cubic spline interpolation; A Bayesian filter state equation is used to eliminate noise for the grid temperature field distribution diagram, and a temperature field gradient distribution diagram is obtained by central difference calculation; Acquire temperature field time series data from the temperature field gradient distribution diagram, use a sliding variance window to detect temperature mutation points in the time series data, and extract temperature fluctuation features for the temperature mutation points; The temperature fluctuation characteristics are compared with the preset temperature threshold. If the temperature fluctuation amplitude exceeds the threshold range, the temperature limit range is determined, the feed speed adjustment interval is divided according to the temperature limit range, and the optimal parameter combination is calculated using a linear programming solver.

4. The method according to claim 1, characterized in that: The real-time collection of vibration signals during the cutting process and the spectrum analysis to obtain the vibration frequency and amplitude changes of the tool are performed, the stress distribution is obtained by combining the cutting force fluctuation characteristics and the cutting temperature data, and a relationship model between the cutting force, cutting temperature, the tool vibration signal and the stress distribution is established to determine the stress concentration position in the high curvature area, including: Obtaining a vibration signal envelope curve through Hilbert transform according to the vibration sampling data, and obtaining vibration spectrum data through fast Fourier transform of the vibration signal envelope curve; Constructing boundary condition parameters according to the vibration spectrum data and the cutting force fluctuation characteristic curve, wherein the boundary condition parameters are calculated by a linear elastic constitutive equation to obtain unit stress components; The vibration signal envelope curve and the cutting force fluctuation characteristic curve are used to calculate a stress influence factor matrix, wherein the stress influence factor matrix includes a root mean square value of the vibration signal and a standard deviation of the force signal; The stress field distribution function is obtained according to the stress influence factor matrix through neural network feature mapping, and the stress field distribution function is output by the deep learning training model to determine the stress concentration position.

5. The method according to claim 1, characterized in that If the stress concentration position is cut continuously for a long time, the local tool wear rate and wear morphology change are evaluated by combining the cutting force and vibration signal data monitored offline and online, and the co-evolution relationship between wear and temperature is analyzed based on the historical cutting force and vibration signals and the current cutting temperature data, including: According to the historical cutting process records, the time series of cutting force signal and vibration signal are obtained, and the time series of cutting force signal and vibration signal are decomposed by wavelet multi-scale to obtain denoised signal; The wear initial position is calibrated by using the cutting force fluctuation peak point for the denoised signal, and the wear profile boundary point is obtained by using the vibration amplitude change curve for the wear initial position; The wear profile boundary points are used to calculate the local wear rate, and the local wear rate is used to obtain the wear morphology characteristic value through the vibration signal envelope spectrum; Performing feature decomposition on the historical cutting force signal sequence according to the wear morphology feature value, wherein the feature decomposition obtains the vibration signal time series feature through a self-attention mechanism; Analyzing and predicting wear-temperature coupling changes based on the vibration signal timing characteristics; According to the predicted wear-temperature coupled changes, the Markov chain is used to describe the co-evolution of wear and temperature.

6. The method according to claim 5, characterized in that Also includes: Obtain historical cutting force and vibration signal data, extract historical cutting force and vibration values, collect the cutting temperature of the current monitoring system, record the current temperature value, and use regression analysis to establish a relationship model between cutting force, vibration value and temperature value. If the relationship model fit meets the preset threshold, the co-evolution relationship between wear amount and temperature value is determined, including: The peak-to-valley ratio and standard deviation of the force value are obtained from the cutting force sequence, the amplitude mean and amplitude frequency characteristics are obtained from the vibration sequence, and the time-aligned feature data are obtained through linear interpolation. The time-aligned characteristic data are denoised using a wavelet basis function, the temperature average and the temperature change slope characteristics are calculated from the denoised data sequence, a multivariate linear equation system is established using the least squares method, and a characteristic correlation matrix is ​​obtained using the Pearson correlation coefficient; If the significance coefficient in the feature correlation matrix is ​​greater than a preset value, a feature mapping function is established using polynomial regression, a state prediction function is obtained through a recursive neural network, and a neural network weight matrix is ​​obtained from the prediction function.

7. The method according to claim 1, characterized in that The method combines the co-evolution relationship between wear and temperature, online monitoring data and the processing time to predict the tool wear rate and generate a wear prediction curve for a preset time period, including: Performing time-series segmentation processing on the online collected data according to the monitoring data cycle, and obtaining historical processing volume statistics from the time-series segmented data; Extracting the characteristic sequence of the root mean square value of cutting force, vibration amplitude and temperature gradient according to the historical machining volume statistics, and obtaining standardized characteristic data through a long short-term memory network; Calculating the Pearson correlation coefficient using the standardized characteristic data, and obtaining a characteristic change rate sequence from the correlation coefficient; If the characteristic change rate sequence meets the preset threshold requirement, the wear prediction curve is filtered using a recursive neural network to obtain the wear prediction curve for a preset time period.

8. The method according to claim 1, characterized in that The wear prediction curve is subjected to trend analysis and feature extraction to obtain the wear parameter change of the tool within a preset time. If the prediction curve shows that the wear parameter change will exceed a preset critical value at a certain time point in the future, the cutting parameters are adjusted to generate a target cutting parameter combination, including: Determine the trend inflection point position according to the wear prediction curve, wherein the trend inflection point position is calculated by the second-order derivative zero point; The least square method is used to calculate the piecewise linear regression coefficient for the trend inflection point position, and the wear slope value is obtained from the piecewise linear regression coefficient; Performing a Fourier transform operation on the wear slope value, extracting a main frequency component of a wear characteristic from the Fourier transform operation, and processing the main frequency component of the wear characteristic via a convolutional neural network to obtain a characteristic time-frequency spectrum; Constructing a decision tree classification structure according to the characteristic time-frequency spectrum, the decision tree classification structure comprising a feed speed layer, a spindle speed layer and a cutting depth layer, and obtaining a parameter optimization sequence from the decision tree classification structure by using a maximum information gain criterion; A cutting process evaluation function is constructed for the parameter optimization sequence. The cutting process evaluation function obtains a parameter combination fitness value through a weighted sum operation, and a target cutting parameter combination is generated from the parameter combination fitness value.

9. The method according to claim 8, characterized in that Also includes: Calculate the current wear rate, use the time series analysis algorithm, combine the historical wear data to build a prediction curve model, calculate the wear rate prediction value at a preset time point in the future, if the prediction value exceeds the preset critical value, call the parameter optimization algorithm to generate several sets of cutting parameter combinations, and search for the target cutting parameter combination within the preset parameter range, including: According to the wear data series, a fixed time window is used for sampling, and a long short-term memory network model is constructed from the sampled data to obtain the wear rate prediction value; Performing wavelet transform on the wear rate prediction value to obtain frequency domain characteristic coefficients, using the frequency domain characteristic coefficients to establish a recursive neural network state equation, and calculating the wear rate at a preset time point from the state equation; If the difference between the wear rate and the preset critical wear value exceeds the warning threshold, cutting parameter optimization is triggered, and a feed speed interval, a cutting depth interval and a spindle speed interval are established based on the cutting parameter optimization; The parameter interval is uniformly quantized to obtain discrete parameter combinations, and the feasible parameter space is obtained from the constraints. The feasible parameter space is searched using a genetic algorithm, and the optimal cutting parameter combination is obtained from the search results.

10. The method according to claim 1, characterized in that The method includes identifying the wear area to be compensated according to the target cutting parameter combination and the set control cycle, evaluating the wear state of the tool and the machining quality of the workpiece, thereby calculating the compensation amount of the cutting parameters, and adjusting the tool motion trajectory of the wear area to be compensated in real time according to the compensation amount, including: A convolutional neural network is used to obtain wear profile features from the detected image, and wear area and wear depth parameters are extracted according to the wear profile features; Performing Kalman filter operation on the wear area and wear depth parameters to obtain a wear state vector; Calculating an optimal compensation parameter combination through a quadratic programming solver according to the wear state vector, wherein the optimal compensation parameter combination includes a feed speed correction amount and a cutting depth correction amount; A compensation trajectory equation is established for the feed speed correction amount and the cutting depth correction amount, and a position compensation instruction sequence is obtained by fitting the compensation trajectory equation through a fifth-order polynomial. The position compensation instruction sequence is used to control tool position compensation.

Images