Tool wear monitoring method and system for heavy truck transmission system component machining machine tool

Abstract

The application relates to the technical field of machine tool cutter monitoring, in particular to a cutter wear monitoring method and system for a heavy truck transmission system part machining machine tool, and specifically comprises the following steps: collecting various working condition parameters during part machining, performing time-frequency domain analysis on the various working condition parameters, combining data change correlation between the working condition parameters, and constructing time-frequency fluctuation coefficients; performing cluster analysis on historical samples according to the working condition parameters, constructing each cluster, and constructing a prediction model about the time-frequency fluctuation coefficients, wherein the noise variance of the sample is adaptively determined through the time-frequency fluctuation coefficient of the sample; the prediction model is used to obtain a time-frequency fluctuation coefficient prediction value of a current machining sample, and the time-frequency fluctuation coefficient prediction value is compared with an actual value to judge the cutter wear state; the accuracy of cutter wear state recognition is improved, the influence of load fluctuation on cutter wear recognition is reduced, the monitoring error caused by dynamic load fluctuation is reduced, and the reliability of cutter monitoring is improved.

Description

Technical Field

[0001] This application relates to the field of machine tool monitoring technology, specifically to a method and system for monitoring tool wear on machine tools used in the processing of heavy-duty truck transmission system components. Background Technology

[0002] As the core hub of vehicle power transmission, the transmission system of heavy-duty trucks directly determines the load-bearing capacity and service life of the entire vehicle through the machining quality of its key components (such as gear shafts, differential housings, and drive shafts). Similarly, the cutting tool, as the core component of a machine tool performing cutting tasks, requires accurate monitoring of its wear state to ensure machining stability and quality. Therefore, building a high-precision, real-time responsive tool wear monitoring system has become crucial for improving the manufacturing level of heavy-duty truck transmission components, reducing tool consumption, and ensuring equipment safety.

[0003] However, existing technologies have significant limitations. Specifically, the working condition of cutting tools is heavily influenced by multiple factors, including the evolution of machining stages, fluctuations in machining conditions, and batch-to-batch material differences. For example, changes in depth of cut and feed rate during machining stage transitions, differences in hardness and internal structure of different batches of blank materials, and changes in cutting conditions all cause load fluctuations during tool operation. Most existing technologies employ the idealized assumption of "steady-state load" and have not yet established a systematic compensation mechanism for dynamic load fluctuations, relying solely on monitoring changes in signal amplitude to infer the degree of wear. This tool monitoring method is severely disconnected from actual working conditions, preventing the system from accurately decoupling signal drift caused by load fluctuations from the characteristic evolution resulting from tool wear, thus affecting the accuracy of tool wear condition identification. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for monitoring tool wear on machine tools used for processing heavy-duty truck transmission system components. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for monitoring tool wear on machine tools used for processing heavy-duty truck transmission system components. This method includes the following steps: Collect various operating parameters of the machine tool when processing each sample at different times, and construct a data sequence of various operating parameters for each sample; A time-domain fluctuation index is constructed for each sample based on the trend change characteristics and the degree of disorder in the data distribution of the data sequence; a frequency-domain fluctuation index is constructed for each sample based on the frequency energy distribution in the spectrum of the data sequence and the correlation between the spectrum of each data sequence and the spectrum of historical data of the same type; a coupling degradation index is constructed for each sample based on the degree of correlation between different types of data sequences; and a time-frequency fluctuation coefficient for each sample is constructed by combining the time-domain fluctuation index and the frequency-domain fluctuation index. All samples are clustered based on their operating parameters; the noise variance of each sample is determined based on the time-frequency fluctuation coefficient; all operating parameters, time-frequency fluctuation coefficients, and noise variances of all samples in each cluster are input into the regression model for training. All working parameters of the current machining sample are input into the trained regression model for prediction to obtain the predicted value of the time-frequency fluctuation coefficient of the current machining sample. The predicted value is then compared with the actual value of the time-frequency fluctuation coefficient of the current machining sample to determine the wear state of the tool.

[0005] In one embodiment, the process of obtaining the time-domain fluctuation index is as follows: The trend sequence of each data sequence is obtained by using a time series decomposition algorithm, and the trend sequence is fitted by a linear fitting algorithm to obtain the slope of the fitted line. A temporal volatility index is constructed for each sample based on the slope and the degree of distribution disorder of all data sequences. The temporal volatility index is positively correlated with the slope and the degree of distribution disorder, respectively.

[0006] In one embodiment, the process of obtaining the frequency domain fluctuation index is as follows: The energy in the frequency range where the frequency value in the spectrum is greater than the preset frequency threshold is taken as high-frequency energy, and the ratio of high-frequency energy to total energy in the spectrum is recorded as the high-frequency energy ratio. The correlation coefficient between the spectrogram of each data sequence of each sample and the spectrogram of the same type of data sequence of each sample processed in the past is calculated by the correlation algorithm. The frequency domain fluctuation index of each sample is calculated based on the high-frequency energy ratio and the correlation coefficient of all data sequence spectrograms for each sample. The frequency domain fluctuation index is positively correlated with the high-frequency energy ratio and negatively correlated with the correlation coefficient.

[0007] In one embodiment, the process of obtaining the coupling degradation index is as follows: Calculate the average correlation coefficient between any two types of data sequences for each sample, and use the inverse proportional mapping value of the average value as the coupling degradation index for each sample.

[0008] In one embodiment, the time-frequency fluctuation coefficient of each sample is the positive fusion value of the time-domain fluctuation index, the frequency-domain fluctuation index, and the coupling degradation index.

[0009] In one embodiment, the process of obtaining the noise variance of each sample is as follows: The time-frequency fluctuation coefficient of each sample is mapped to a positive number. The product of this positive number and the original fixed noise variance is used as the adjustment amount. The sum of the adjustment amount and the original fixed noise variance is used as the final noise variance of each sample.

[0010] In one embodiment, the process of obtaining the predicted value of the time-frequency fluctuation coefficient is as follows: All operating parameters of the current processing sample are input into the trained regression model corresponding to each cluster for prediction, and the prediction results of each regression model are obtained; the distance between the current processing sample and the cluster center of each cluster is determined based on all operating parameters of the current processing sample. The predicted value of the time-frequency fluctuation coefficient of the current processed sample is determined based on the prediction results and the distance.

[0011] In one embodiment, determining the predicted value of the time-frequency fluctuation coefficient of the current processed sample based on the prediction result and the distance specifically involves: The normalized value of the inverse proportional mapping of the distance is used as the weight of the prediction result corresponding to each cluster; the weighted sum of the prediction results corresponding to all clusters is used as the predicted value of the time-frequency fluctuation coefficient of the current processed sample.

[0012] In one embodiment, determining the wear state of the cutting tool specifically involves: If the difference between the predicted value of the time-frequency fluctuation coefficient of the current processing sample and its actual value is greater than a preset threshold, the tool is determined to be worn; if it is less than or equal to the preset threshold, the tool is determined not to be worn.

[0013] Secondly, embodiments of this application also provide a tool wear monitoring system for machine tools used in the processing of heavy truck transmission system components, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0014] The embodiments of this application have at least the following beneficial effects: This application collects various operating parameters of the machine tool during the processing of each sample at different times. By performing time-frequency domain analysis on these parameters and examining the correlation between their data changes, a time-frequency fluctuation coefficient is constructed to assess the impact of tool wear and load changes on the data. This enhances the differentiation of different machine tool operating states and improves the reliability of subsequent prediction models. Then, based on the operating parameters, historical samples are clustered, and prediction models for the time-frequency fluctuation coefficient are constructed for each cluster. The noise variance of the sample is adaptively determined using the sample's time-frequency fluctuation coefficient, which helps decouple tool wear and load changes, improves the accuracy of subsequent tool wear state identification, and reduces the impact of load fluctuations on tool wear identification. All operating parameters of the current processing sample are input into the trained regression model for prediction to obtain the predicted value of the time-frequency fluctuation coefficient for the current processing sample. This predicted value is compared with the actual value of the time-frequency fluctuation coefficient for the current processing sample to determine the tool wear state. Using the prediction model, the monitoring error caused by dynamic load fluctuations is reduced, the accuracy of tool wear state identification is improved, and the reliability of tool monitoring is enhanced. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the steps of a method for monitoring tool wear in a machine tool for processing heavy truck transmission system components according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of obtaining the frequency domain fluctuation index. Detailed Implementation

[0016] The following, in conjunction with the accompanying drawings, details the specific scheme of the tool wear monitoring method and system for processing heavy truck transmission system components provided in this application.

[0017] Please see Figure 1 The document illustrates a flowchart of a method for monitoring tool wear on a machine tool used for processing heavy-duty truck transmission system components, according to an embodiment of this application. The method includes the following steps: Step S1: Collect various working condition parameters of the machine tool when processing each sample at each time point, and construct a data sequence of various working condition parameters for each sample.

[0018] During the machining of each heavy-duty truck transmission system component, the machine tool collects vibration acceleration signals by installing a three-axis IEPE accelerometer at the front end of the spindle housing near the tool tip, with a data sampling rate of 30kHz, to capture high-frequency friction noise and micro-chipping impact caused by wear; a Hall power transmitter installed at the spindle driver input collects spindle power and current, with a data sampling rate of 1kHz; a broadband acoustic emission sensor is placed near the tool holder clamping area to collect acoustic emission signal data, with a data sampling rate of 500kHz; a piezoelectric rotary force gauge is installed between the workpiece table and the fixture (or using a force-measuring tool holder) to collect cutting force data, with a data sampling rate of 10kHz; cutting speed, feed rate, and depth of cut data are read in real time from the machine tool CNC system data interface, with a data refresh rate of 1Hz; all data are timestamped through a synchronous acquisition system to ensure accurate fusion of multi-source information.

[0019] The various working parameters collected during the processing of each sample are subjected to maximum-minimum normalization. The normalized data are then used to construct corresponding data sequences according to the collection order, resulting in acceleration, power, current, acoustic emission, cutting force, cutting speed, feed rate, and depth of cut sequences for each sample in three directions.

[0020] Step S2: Construct a time-domain fluctuation index for each sample based on the trend change characteristics and the degree of disorder in the data distribution of the data sequence; construct a frequency-domain fluctuation index for each sample based on the frequency energy distribution in the spectrum of the data sequence and the correlation between the spectrum of each data sequence and the spectrum of historical data of the same type; construct a coupling degradation index for each sample based on the degree of correlation between different types of data sequences; and construct the time-frequency fluctuation coefficient for each sample by combining the time-domain fluctuation index and the frequency-domain fluctuation index.

[0021] The quality of components in a heavy-duty truck's transmission system directly impacts the vehicle's safety and performance, thus requiring strict control over their machining quality. During this process, the condition of the machine tool is crucial for ensuring component quality, necessitating real-time and accurate monitoring of tool wear. Current tool wear monitoring methods largely rely on the idealized assumption of "steady-state load." However, due to the complexity of the machining process and the high precision requirements for heavy-duty truck transmission system components, this monitoring method has significant limitations. It cannot compensate for dynamic load fluctuations during machining, leading to errors in tool wear monitoring and consequently affecting component machining quality. Therefore, a more accurate tool wear monitoring method is needed.

[0022] During the operation of a machine tool, the wear condition of the cutting tool affects the physical signals during the cutting process, such as the cutting force, vibration, acoustic emission signals, and spindle electrical parameters, which fluctuate. Furthermore, the fluctuation of the dynamic load during the machining process also causes these parameters to change. Therefore, in actual machining, it is necessary to quantitatively analyze the fluctuation of various monitoring signals during tool operation and combine the impact of tool wear and load changes on various physical signals to improve the accuracy of tool wear identification.

[0023] First, in terms of the time-domain amplitude distribution, the volatility of the signal needs to be examined. When the cutting process parameters are stable and the tool is normal, the time-domain signals of various monitoring signals fluctuate smoothly and randomly around the mean. However, when the tool wears, the increased friction on the flank face leads to instability in the cutting process and irregular chip breakage, resulting in high-frequency random pulses in the signal. This causes the overall fluctuation of the time-domain signal to exhibit an irreversible monotonic change trend. When the dynamic load fluctuates due to changes in the cutting process parameters, it will cause instantaneous step jumps in the time-domain signals of various monitoring signals, but overall, this fluctuation does not exhibit a monotonic change trend.

[0024] Secondly, in the frequency domain, when the cutting process parameters are stable and the tool is normal, the signal energy is highly concentrated in the low-frequency bands such as the spindle speed and its harmonics, with extremely low high-frequency noise and a stable spectral structure. However, when the tool wears, the spectrum shows a clear trend of "high-frequencyization," that is, the proportion of high-frequency energy increases significantly, while the spectrum diffuses or drifts, causing the proportion of main frequency energy to decrease. When the dynamic load fluctuates, although there is an instantaneous step in the time domain signal, since the load change (such as the cutting depth) is essentially a change in the amplitude of the input force, according to the linear system theory, the shape of its output response spectrum is determined by the system transfer function (remains unchanged), and the amplitude changes proportionally with the input force. This makes the frequency domain exhibit an overall unchanged spectrum shape, but the amplitude of each frequency component scales proportionally.

[0025] Finally, under normal conditions or when the load changes, different signals maintain a high and stable correlation or linear relationship, as well as between signals and process parameters. However, when the tool wears, this coupling relationship is broken, which manifests as a "force-vibration decoupling" phenomenon (such as power slowly increasing due to friction, while high-frequency vibration fluctuates violently due to instability), causing the correlation between multiple signals to show a downward trend.

[0026] Based on the above analysis, each data sequence is first used as input, and STL time series decomposition is employed to obtain the trend sequence of each data sequence. Then, linear fitting is performed on each trend sequence to output the fitted line equation. In this embodiment, the least squares method is used for linear fitting. Finally, a discrete Fourier transform is performed on each data sequence to obtain its spectrum. STL time series decomposition, least squares linear fitting, and discrete Fourier transform are all well-known techniques, and their specific processes will not be elaborated further.

[0027] It should be noted that there are many existing time series decomposition algorithms, linear fitting algorithms, and time-frequency transformation algorithms. Implementers may also use other time series decomposition algorithms to obtain the trend term of the data sequence, use other linear fitting algorithms to perform linear fitting on the trend sequence, and use other time-frequency transformation algorithms to obtain the spectrum of the data sequence. This application does not impose any specific restrictions.

[0028] Furthermore, taking the current machining sample as an example, the time-frequency fluctuation coefficient of each sample is calculated to measure the fluctuation of the tool operation data. Preferably, in this embodiment, the expression for the time-frequency fluctuation coefficient is: In the formula, It is the time-frequency fluctuation coefficient of the currently processed sample; , , These are the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively. , , These are the time-domain fluctuation index, frequency-domain fluctuation index, and coupling degradation index of the current processed sample, respectively. This is the normalization function. Wherein, Implementers can set it themselves according to the actual situation. , , The value of is not specifically limited in this application; in this embodiment... , , Take values ​​of 0.33, 0.33, and 0.34 respectively.

[0029] It is the mean slope of the trend sequence fitted line of all data sequences of the current processed sample; It is the mean of the standard deviations of the elements in the sequence of all data sequences of the current processed sample; e is the natural constant; It is a normalization function. It should be noted that an exponential function is used to increase the influence of trend characteristics. The normalization function used in this embodiment is maximum-minimum normalization. Implementers in other embodiments of this application may use other normalization methods, and this application does not impose specific limitations.

[0030] Where n is the number of types of operating parameters, and in this embodiment n=10; , These are the high-frequency energy and total signal energy in the spectrum of the data sequence of the j-th type of working condition parameter of the current processing sample, respectively. In this embodiment, the high-frequency energy in the spectrum refers to the energy corresponding to the frequency range that exceeds five times the spindle speed. It is the high-frequency energy ratio of the spectrum of the data sequence of the j-th type of working condition parameter of the current processing sample. ,in, This is the spectrum of the j-th type of working condition parameter of the current processed sample, where m is the preset number of historical processed samples. In this embodiment, m is the sum of the number of samples in the previous batch and the number of processed samples in the current batch. This is a spectrum of parameters for the j-th type of operating condition in the i-th processing prior to the current processing. It is the absolute value of the Pearson correlation coefficient between the two spectrograms. It should be noted that, regarding the division of high-frequency energy in the spectrograms, in this embodiment, the spindle frequency under the current time window is dynamically calculated using real-time machine tool cutting speed data and a known transmission ratio, thereby determining the threshold boundary of the fifth harmonic of the spindle frequency. In other embodiments of this application, the implementer can set the threshold boundary for high-frequency energy division according to the actual situation.

[0031] Specifically: Calculate the absolute value of the Pearson correlation coefficient between any two data sequences of the current processed sample, denoted as the first correlation coefficient; calculate the mean of all the first correlation coefficients of the current processed sample; and take the difference between the natural number 1 and this mean as... It should be noted that before calculating the correlation coefficient between any two types of data sequences, the different sampling rates result in different sequence lengths. To facilitate subsequent processing, the data sequences of different dimensions are compressed to a uniform length. Specifically, a time sliding window is set, with the window length corresponding to the time interval of the data acquisition for the operating parameter with the lowest acquisition frequency. The sliding step size is the same as the window length. The root mean square (RMS) of all elements corresponding to all acquisition times within each time sliding window of the data sequence for each type of operating parameter is used as each new element in that data sequence, resulting in the compressed data sequence. This compresses all data sequences to a uniform length, and the correlation coefficient is then calculated using the compressed data sequences.

[0032] In other embodiments of this application, the time-frequency fluctuation coefficient may also be the normalized value of the product of the time-domain fluctuation index, the frequency-domain fluctuation index, and the coupling degradation index. In this embodiment, maximum-minimum normalization is used to normalize this product.

[0033] Tool wear causes significant fluctuations in time-domain signals, and these fluctuations exhibit an irreversible monotonic trend. Existing techniques use standard deviation as a statistical measure to assess data dispersion, i.e., data fluctuation, and the STL algorithm to obtain a trend sequence to measure data trend changes. This scheme similarly utilizes standard deviation and trend sequence to measure the fluctuations and monotonic trends of time-domain signals. Secondly, in the frequency domain, tool wear increases the proportion of high-frequency energy, causing spectral drift. Existing techniques use correlation coefficients to measure the correlation between two data sequences. This scheme also uses the correlation coefficient of the spectral vector to measure the stability of the spectral structure and combines it with the proportion of high-frequency energy to measure spectral drift. Finally, the correlation coefficient is used to measure the correlation between different signals.

[0034] Step S3: Cluster all samples based on the operating parameters of the samples; determine the noise variance of each sample based on the time-frequency fluctuation coefficient; input all operating parameters, time-frequency fluctuation coefficients and noise variances of all samples in each cluster into the regression model for training.

[0035] The time-frequency fluctuation coefficient calculated above is used to perform data fluctuation quantification analysis by combining the impact of tool wear and load changes on the monitoring signal. However, since tool wear and load changes are usually coupled during operation, it is difficult to accurately identify the tool wear situation using only the time-frequency fluctuation coefficient. Therefore, it is necessary to combine different working parameters for further cluster analysis and construct a prediction model to improve the accuracy of quantifying signal fluctuations caused by load changes under different working parameters.

[0036] The processing parameters of all samples from the previous batch and the processed samples in the current batch are obtained, with each sample corresponding to one processing step of a transmission system component. The time-frequency fluctuation coefficient of each sample is calculated according to the above. Then, the data sequence of all processing parameters of all samples is used as input, with a neighborhood radius of 0.15, a minimum number of points of 20, and the metric distance being the mean of the DTW distance between all similar data sequences of two samples. DBSCAN clustering is then performed, and the clustering results are output. It should be noted that implementers can set the neighborhood radius and minimum number of points according to their actual situation; this application does not impose specific restrictions. DBSCAN clustering is a well-known technique, and its specific process will not be elaborated further.

[0037] It is understood that this application provides only one clustering method for clustering samples. There are many existing clustering methods, and implementers may also use other clustering algorithms to cluster samples. This application does not impose any specific restrictions.

[0038] In each cluster, all data sequences and time-frequency fluctuation coefficients of each sample within that cluster are input into a regression algorithm for model training. In this embodiment, a weighted Gaussian regression algorithm is used. During Gaussian process regression model training and prediction, a covariance matrix needs to be constructed. The construction of this covariance matrix is ​​closely related to the noise variance of each sample participating in the regression model training. In the original Gaussian regression, the noise term for each sample is consistent. However, due to differences in disturbances under different working conditions during machine tool operation, even within the same cluster, there may be significant fluctuations between different samples. For model training, it is necessary to add the influence of samples with smaller disturbances to improve the model's reliability. The noise variance of each sample corresponds to the data fluctuation during the processing, and this data fluctuation can be quantified using the aforementioned time-frequency fluctuation coefficient. Therefore, the noise variance of each sample is constructed using the time-frequency fluctuation coefficient, and subsequently, this noise variance is used to construct the logarithmic marginal likelihood function for the regression model.

[0039] In the formula, It is the first The noise variance of each sample; It is the original fixed noise variance, which can be obtained by the statistical variance of the time-frequency fluctuation coefficient of the machine tool when processing historical samples with new tools (in good condition); It is the first The time-frequency fluctuation coefficient of each sample; , These are the minimum and maximum values ​​of the time-frequency fluctuation coefficient among all samples participating in the regression, respectively. This is the correction factor, used to control the magnitude and strength of the noise variance correction. The recommended value range is 0.3 to 0.5, and in this embodiment, it is 0.3.

[0040] By assigning high-noise variance to samples with high volatility (suspected wear), their weight in model training is actively reduced, allowing the Gaussian regression model to purely fit the baseline characteristics of the tool under healthy conditions. The predicted values ​​output by the model represent the expected normal volatility level under current operating conditions.

[0041] Step S4: Input all working condition parameters of the current machining sample into the trained regression model for prediction, so as to obtain the predicted value of the time-frequency fluctuation coefficient of the current machining sample, and compare it with the actual value of the time-frequency fluctuation coefficient of the current machining sample to determine the wear state of the tool.

[0042] The trained regression models for each cluster are obtained through the above method. When monitoring tool wear in subsequent machining operations, the fluctuation measure of machining parameters under normal conditions, i.e., the predicted value of the time-frequency fluctuation coefficient, is obtained based on existing prediction models for various working conditions. However, due to the complexity of machining requirements and the diversity of machining materials, new combinations of working condition parameters may occur during online monitoring. Furthermore, the distances to the centers of multiple clusters are relatively uniform. Therefore, to improve prediction accuracy, a distance-weighted fusion method is used to obtain the predicted value of the time-frequency fluctuation coefficient, expressed as: In the formula, It is the predicted value of the time-frequency fluctuation coefficient of the current processed sample p; It represents the number of clusters. It is the distance between the currently processed sample p and the cluster center of the s-th cluster. It is the sum of the distances between the current processed sample p and the cluster centers of all clusters. It is the time-frequency fluctuation coefficient of the current processed sample p predicted by the regression model corresponding to the s-th cluster. The value is a preset minimum positive number, used to prevent the denominator from being 0. In this embodiment... .

[0043] After obtaining the predicted value of the time-frequency fluctuation coefficient, calculate... The difference between the actual value of the time-frequency fluctuation coefficient of the current processing sample p and the actual value of the current processing sample p is considered. If this difference is greater than the set wear threshold U=0.2, it indicates that the tool is worn, which will affect the processing of the parts. If the difference is less than or equal to the set wear threshold, it indicates that the tool is not worn and will not affect the processing of the parts. It should be noted that in this embodiment, the threshold is calculated based on historical data. In other embodiments of this application, the implementer can set the wear threshold according to the actual situation.

[0044] A schematic diagram illustrating the process of obtaining the frequency domain fluctuation index is shown below. Figure 2 As shown.

[0045] Based on the same inventive concept as the above methods, this application also provides a tool wear monitoring system for heavy truck transmission system component processing machine tools, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for monitoring tool wear of heavy truck transmission system component processing machine tools.

[0046] In summary, this application provides a method for monitoring tool wear on machine tools used in heavy-duty truck transmission system component processing. It collects various operating parameters of the machine tool during the processing of each sample at different times. By performing time-frequency domain analysis on these parameters and examining the correlation between their data changes, a time-frequency fluctuation coefficient is constructed to assess the impact of tool wear and load changes on the data. This enhances the differentiation of different machine tool operating states and improves the reliability of subsequent prediction models. Then, based on the operating parameters, historical samples are clustered, and prediction models for the time-frequency fluctuation coefficient are constructed for each cluster. The noise variance of the samples is adaptively determined using the time-frequency fluctuation coefficient, which helps decouple tool wear and load changes, improving the accuracy of subsequent tool wear state identification and reducing the impact of load fluctuations on tool wear identification. All operating parameters of the current processing sample are input into a trained regression model for prediction to obtain the predicted value of the time-frequency fluctuation coefficient for the current processing sample. This predicted value is compared with the actual value of the time-frequency fluctuation coefficient for the current processing sample to determine the tool wear state. By utilizing the prediction model, the monitoring error caused by dynamic load fluctuations is reduced, the accuracy of tool wear state identification is improved, and the reliability of tool monitoring is enhanced.

Claims

1. A method for monitoring tool wear on machine tools used in the processing of heavy-duty truck transmission system components, characterized in that, The method includes the following steps: Collect various operating parameters of the machine tool when processing each sample at different times, and construct a data sequence of various operating parameters for each sample; A time-domain fluctuation index is constructed for each sample based on the trend change characteristics and the degree of disorder in the data distribution of the data sequence; a frequency-domain fluctuation index is constructed for each sample based on the frequency energy distribution in the spectrum of the data sequence and the correlation between the spectrum of each data sequence and the spectrum of historical data of the same type; a coupling degradation index is constructed for each sample based on the degree of correlation between different types of data sequences; and a time-frequency fluctuation coefficient for each sample is constructed by combining the time-domain fluctuation index and the frequency-domain fluctuation index. All samples are clustered based on their operating parameters; the noise variance of each sample is determined based on the time-frequency fluctuation coefficient; all operating parameters, time-frequency fluctuation coefficients, and noise variances of all samples in each cluster are input into the regression model for training. All working parameters of the current machining sample are input into the trained regression model for prediction to obtain the predicted value of the time-frequency fluctuation coefficient of the current machining sample. The predicted value is then compared with the actual value of the time-frequency fluctuation coefficient of the current machining sample to determine the wear state of the tool. The process of obtaining the time-domain fluctuation index is as follows: The trend sequence of each data sequence is obtained by using a time series decomposition algorithm, and the trend sequence is fitted by a linear fitting algorithm to obtain the slope of the fitted line. The temporal volatility index of each sample is constructed based on the slope and the degree of distribution disorder of all data sequences of each sample. The temporal volatility index is positively correlated with the slope and the degree of distribution disorder, respectively. The process of obtaining the frequency domain fluctuation index is as follows: The energy in the frequency range where the frequency value in the spectrum is greater than the preset frequency threshold is taken as high-frequency energy, and the ratio of high-frequency energy to total energy in the spectrum is recorded as the high-frequency energy ratio. The correlation coefficient between the spectrogram of each data sequence of each sample and the spectrogram of the same type of data sequence of each sample processed in the past is calculated by the correlation algorithm. The frequency domain fluctuation index of each sample is calculated based on the high-frequency energy ratio and the correlation coefficient of all data sequence spectrograms of each sample. The frequency domain fluctuation index is positively correlated with the high-frequency energy ratio and negatively correlated with the correlation coefficient. The process of obtaining the coupling degradation index is as follows: Calculate the average correlation coefficient between any two types of data sequences for each sample, and use the inverse proportional mapping value of the average value as the coupling degradation index for each sample; The process for obtaining the noise variance of each sample is as follows: The time-frequency fluctuation coefficient of each sample is mapped to a positive number, and the product of this positive number and the original fixed noise variance is used as the adjustment amount. The sum of the adjustment amount and the original fixed noise variance is used as the final noise variance of each sample. The process for obtaining the predicted value of the time-frequency fluctuation coefficient is as follows: All operating parameters of the current processing sample are input into the trained regression model corresponding to each cluster for prediction, and the prediction results of each regression model are obtained; the distance between the current processing sample and the cluster center of each cluster is determined based on all operating parameters of the current processing sample. The predicted value of the time-frequency fluctuation coefficient of the current processed sample is determined based on the prediction results and the distance.

2. The method for monitoring tool wear in machine tools for processing heavy truck transmission system components as described in claim 1, characterized in that, The time-frequency fluctuation coefficient of each sample is the positive fusion value of the time-domain fluctuation index, the frequency-domain fluctuation index, and the coupling degradation index.

3. The method for monitoring tool wear in machine tools for processing heavy truck transmission system components as described in claim 1, characterized in that, The step of determining the predicted value of the time-frequency fluctuation coefficient of the current processed sample based on the prediction result and the distance is as follows: The normalized value of the inverse proportional mapping of the distance is used as the weight of the prediction result corresponding to each cluster; the weighted sum of the prediction results corresponding to all clusters is used as the predicted value of the time-frequency fluctuation coefficient of the current processed sample.

4. The method for monitoring tool wear in machine tools for processing heavy truck transmission system components as described in claim 1, characterized in that, The determination of the wear condition of the cutting tool specifically involves: If the difference between the predicted value of the time-frequency fluctuation coefficient of the current processing sample and its actual value is greater than the preset threshold, it is determined that the current tool is worn. If the value is less than or equal to the preset threshold, it is determined that the current tool has no wear.

5. A tool wear monitoring system for machine tools used in the processing of heavy-duty truck transmission system components, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-4.

Images