Heavy truck transmission system component machining tool life prediction management method

Abstract

This application relates to the field of predictive data processing technology, specifically to a method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components. The method includes: collecting cutting force signals during the machine tool's operation; performing wavelet decomposition on the cutting force signals; constructing impact load characteristic indicators based on the peak point variation characteristics of the detail coefficients of each decomposed signal layer; adaptively adjusting the transition adjustment parameters of each signal layer based on the differences in the correlation between different signal layers and the degree of irregular fluctuation of the detail coefficients of each signal layer; calculating the adjustment value of each detail coefficient; performing wavelet threshold denoising on the cutting force signals; and combining the denoised spindle bearing housing vibration signal, high-frequency stress wave signal, and tool tip temperature signal to predict tool life. This method avoids the problem of poor denoising effect of the cutting force signal due to high-frequency impact and external electromagnetic interference, thus improving the accuracy of tool life prediction.

Description

Technical Field

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

[0002] As a core component of new energy heavy-duty trucks, the machining quality of the parts in the heavy-duty truck transmission system directly affects the vehicle's power transmission efficiency, reliability, and service life. The heavy-duty truck transmission system includes various precision components with complex curved surface structures, such as the transmission housing, transmission bearings, and gears. Currently, the machining processes for heavy-duty truck transmission system components are not yet perfect, resulting in insufficient machining accuracy, low production efficiency, and high costs, making it difficult to meet the demands of the industry's rapid development. Therefore, to ensure stable and high-quality machining of heavy-duty truck transmission system components and avoid frequent tool changes or tool breakage during machining, it is usually necessary to reasonably predict the service life of machine tool cutting tools.

[0003] When predicting the lifespan of machine tool cutting tools used in the machining of heavy-duty truck transmission system components, it is typically necessary to monitor mechanical changes, vibration signals, and temperature fluctuations during the machining process using sensors. This involves acquiring key parameters reflecting the tool wear state in real time and building a lifespan prediction model for the machine tool to predict its lifespan in real time. However, when acquiring cutting force signals using force sensors, it's important to consider that in milling gears or turning shaft components, the tool and workpiece are in intermittent contact. During milling, the tool continuously enters and exits the workpiece, resulting in high-frequency pulses in the signal. Furthermore, the starting and stopping of high-power equipment (such as frequency converters and motors) in the machining workshop generates strong electromagnetic interference, introducing high-frequency burrs into the signal. These burrs are similar in characteristics to the signal of minor tool chipping. These external interferences can cause confusion between the acquired normal signal and the signal indicating tool damage, leading to significant deviations in the subsequent lifespan prediction of machine tool cutting tools used in the machining of heavy-duty truck transmission system components. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components, thereby resolving the existing issues.

[0005] The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components in this application adopts the following technical solution: One embodiment of this application provides a method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components. The method includes the following steps: Real-time acquisition of cutting force signals on the worktable, spindle bearing housing vibration signals, high-frequency stress wave signals, and tool head temperature signals during the operation of the machine tool; The cutting force data during no-load conditions is deleted. Wavelet decomposition is performed on the cutting force signals in each preset period of the remaining time to obtain the signals of each layer of the cutting force signals in each period. The amplitude distribution characteristics of each peak point in the detail coefficients of each layer of signals are determined. Based on the complexity and irregularity of the data changes of the amplitude distribution characteristics of all peak points, the impact load characteristic index of each layer of signals is constructed. The correlation difference feature value of each layer signal is constructed based on the degree of difference between the detail coefficients of each layer signal and the detail coefficients of each other layer signal; the local feature value of each layer signal is constructed based on the degree of disorder in the distribution of the number of detail coefficients among all adjacent peak points in each layer signal and the degree of disorder in the distribution of the amplitude of all peak points; each detail coefficient of each layer signal is adjusted based on the impact load feature index, the correlation difference feature value and the local feature value, and then wavelet threshold denoising is performed on the cutting force signal; A prediction model is trained based on historically collected denoised cutting force signals, vibration signals, high-frequency stress wave signals, and temperature signals. Tool life is then predicted based on the trained prediction model.

[0006] In one embodiment, the process of obtaining the impact load characteristic index is as follows: Analyze the amplitude difference between each peak point and its two adjacent data points in each layer of signal, and determine the characteristic value of each peak point by combining the amplitude of each peak point; Calculate the permutation entropy of the eigenvalues ​​of all peak points in each layer of the signal; perform curve fitting on the eigenvalues ​​of all peak points in each layer of the signal, and calculate the residuals of all peak points on the obtained fitted line; The impact load characteristic index of each layer signal is determined based on the arrangement entropy and the residual.

[0007] In one embodiment, the process of obtaining the feature values ​​of each peak point is as follows: Calculate the sum of the absolute values ​​of the amplitude differences between each peak point and its two left and right adjacent data points in each layer of signal, and use the sum of the amplitude of each peak point and the sum of the absolute values ​​as the characteristic value of each peak point.

[0008] In one embodiment, the impact load characteristic index is proportional to both the permutation entropy and the residual.

[0009] In one embodiment, the process of obtaining the correlation difference feature value is as follows: The set of all detail coefficients of each layer signal is denoted as the detail coefficient set; the degree of difference between each layer signal of the cutting force signal in the same period and the detail coefficient set of each other layer signal is calculated, and the coefficient of variation of all the said degree of difference of each layer signal is calculated as the correlation difference feature value of each layer signal.

[0010] In one embodiment, the process of obtaining the local feature values ​​is as follows: Calculate the standard deviation of the number of detail coefficients between all adjacent peak points in each layer of the signal, and calculate the standard deviation of the amplitude of all peak points in each layer of the signal. Use the sum of the normalized values ​​of the two standard deviations as the local feature value of each layer of the signal.

[0011] In one embodiment, the process of adjusting each detail coefficient of the signal at each layer is as follows: The adjustment coefficients of each layer of signal are constructed based on the correlation difference feature value, the local feature value, and the impact load feature index; Based on the adjustment coefficient, the initial transition adjustment parameter is adaptively adjusted to obtain the adaptive transition adjustment parameter of each layer signal, which is used to adjust each detail coefficient of each layer signal.

[0012] In one embodiment, the adjustment coefficient is positively correlated with the correlation difference feature value, the local feature value, and the impact load feature index, respectively.

[0013] In one embodiment, the process of obtaining the adaptive transition adjustment parameter is as follows: The adjustment coefficients of each layer signal are inversely proportionally mapped, and the product of the mapping result and the initial transition adjustment parameter is used as the adaptive transition adjustment parameter of each layer signal.

[0014] In one embodiment, the prediction model training based on historically acquired denoised cutting force signals, vibration signals, high-frequency stress wave signals, and temperature signals specifically involves: Historical signal data of various types are acquired, and data under no-load conditions are removed. Then, the variances of the denoised cutting force signal, the denoised vibration signal, and the denoised high-frequency stress wave signal are calculated within each temperature acquisition time interval. The vector composed of the variances of each temperature data and the three signals in the previous temperature acquisition time interval is used as a feature vector. All feature vectors in the historical data are used as inputs to the prediction model to be trained.

[0015] This application has at least the following beneficial effects: This application collects cutting force signals during machine tool operation, performs wavelet decomposition on the cutting force signals, and constructs impact load characteristic indices based on the peak point variation characteristics of the detail coefficients of each decomposed signal to assess the severity of high-frequency noise interference. By analyzing the differences in the correlation between different signal layers and the irregularity of the fluctuations in the detail coefficients of each signal layer, the severity of external electromagnetic interference to the cutting force signals is assessed. Then, the transition adjustment parameters of each signal layer are adaptively adjusted, and the adjustment value of each detail coefficient is calculated to perform wavelet threshold denoising on the cutting force signals. This improves the filtering effect of wavelet denoising algorithm on noise interference in cutting force signals, avoiding the problem in traditional wavelet threshold denoising where fixed transition adjustment parameters are used for filtering operations, which do not consider the impact of high-frequency impact and external electromagnetic interference on the cutting force signals, and directly denoises, easily confusing normal cutting signals with tool damage cutting signals, leading to poor subsequent tool life prediction results. Attached Figure Description

[0016] Figure 1 A flowchart of the method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components provided in this application; Figure 2 This is a schematic diagram illustrating the process of obtaining characteristic indicators of impact loads. Figure 3 This is a schematic diagram illustrating the process of adjusting the detail factor. Detailed Implementation

[0017] The following, in conjunction with the accompanying drawings, details the specific scheme of the method for predicting and managing the tool life of heavy-duty truck transmission system components provided in this application.

[0018] One embodiment of this application provides a method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components.

[0019] Specifically, the following methods for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components are provided. Please refer to [link / reference]. Figure 1 The method includes the following steps: Step S1: Real-time acquisition of cutting force signals, spindle bearing housing vibration signals, high-frequency stress wave signals, and tool head temperature signals on the worktable during the operation of the machine tool.

[0020] When predicting the service life of machine tool cutting tools used in the machining of heavy-duty truck transmission system components, it is typically necessary to use force sensors, vibration sensors, acoustic emission sensors, and temperature sensors for real-time data acquisition. This application first installs a platform-type force gauge under the machine tool table to acquire the cutting force signal data of the machine tool table during machining operations in real time. In this embodiment, the sampling frequency of the cutting force signal is set to 10kHz. A vibration sensor is installed at the spindle bearing housing to acquire vibration signals in real time. In this embodiment, the sampling frequency of the vibration signal is set to 5kHz. An acoustic emission sensor is installed on the tool holder (the implementer may also install the acoustic emission sensor in other locations, such as on the workpiece fixture; this application does not impose specific limitations), thereby acquiring the high-frequency stress wave converted electrical signal in real time, with a sampling rate of 100kHz. A thermocouple is installed near the tool tip on the tool holder to acquire temperature signals in real time, with a sampling rate of 1Hz. Since the temperature signal changes relatively slowly compared to the other three signals, the acquisition frequency of the temperature signal is usually lower than the acquisition frequency of the other three signals.

[0021] It should be noted that the implementer can set the acquisition frequency of various signals according to the actual situation, and this application does not impose specific restrictions.

[0022] Furthermore, by using 5G wireless transmission, various signals acquired in real time are transmitted to the terminal processing device. Simultaneously, during signal transmission, the acquired signals are time-calibrated using Precise Time Protocol (PTP) to prevent time deviations.

[0023] Step S2: Delete the cutting force data when unloaded, perform wavelet decomposition on the cutting force signal in each preset period of the remaining time to obtain the layer signal of the cutting force signal in each period; determine the amplitude distribution characteristics in the neighborhood of each peak point in the detail coefficient of each layer signal, and construct the impact load characteristic index of each layer signal based on the data change complexity and irregularity of the amplitude distribution characteristics of all peak points.

[0024] After signal acquisition and wireless transmission via sensors, the terminal processing device performs preprocessing on the real-time acquired signals. Considering the external noise interference during the machining of heavy-duty truck transmission system components, which can easily lead to deviations in the prediction of machine tool life, filtering of the real-time acquired signals is necessary.

[0025] Conventional filtering struggles to effectively differentiate between high-frequency pulses caused by intermittent cutting impacts, electromagnetic interference noise in the environment, and subtle tool damage signals. This leads to the erroneous filtering of valuable signals and the loss of abrupt changes during denoising. Wavelet thresholding, on the other hand, can distinguish noise from useful signals at multiple scales, more effectively removing external interference. However, traditional wavelet thresholding uses fixed transition parameters. When the parameter is too large, it can lead to excessive noise retention in heavily noise-affected situations; conversely, when the parameter is too small, it can result in over-filtering of details and loss of valuable signals in less affected signals.

[0026] To achieve better noise reduction, cutting force signals under no-load conditions are first filtered out. Specifically, the real-time acquired cutting force signals are filtered, and the lower limit of the cutting force is set to 1.5 times the average static cutting force when the machine tool spindle is started but not in contact with the workpiece. Signals with cutting forces lower than the lower limit are discarded. In other embodiments of this application, the implementer can set the lower limit of the cutting force according to the actual situation. Each cycle is set to m seconds. In this embodiment, the value of m is set to 1. In other embodiments of this application, the implementer can set the value of m according to the actual situation. The cutting force signal acquired in each cycle during the remaining operation is taken as a segment of the cutting force signal. Each segment of the cutting force signal is used as input, and wavelet decomposition is performed based on db4 wavelet. The decomposition level is set to 7 levels, and the output is a cutting force signal decomposed into 7 levels. db4 wavelet decomposition is a well-known technique and will not be elaborated further. The time of each cycle does not include the time under no-load conditions.

[0027] Furthermore, considering that when machining parts using machine tools for heavy-duty truck transmission systems, the machine tools primarily perform milling of gears or turning of shafts through intermittent cutting operations, significant impact forces are generated at the moment of entry and exit. When acquiring cutting force signals using a platform-type force gauge, the high-frequency impact load caused by the impact manifests as high-frequency noise interference in the signal. The decomposed detail coefficients exhibit rapid, random, and large-amplitude changes, generally displaying non-periodic spike-and-thick-tail pulse characteristics. The waveform is affected by interference, altering its bilateral symmetry. The more severe the noise interference, the more pronounced the characteristic behavior. In contrast, the machining of conventional heavy-duty truck transmission parts typically includes harmonic information of the gear meshing frequency or spindle rotation frequency, with the detail coefficients exhibiting more regular periodic fluctuations and relatively stable amplitude changes.

[0028] Based on the above analysis, taking the k-th layer signal of the cutting force signal in the i-th cycle as an example, a peak-finding algorithm is used to obtain the peak points among all detail coefficients of the signal in that layer. It should be noted that, for the peak-finding algorithm, this embodiment adopts the AMPD peak-finding algorithm. There are many existing peak-finding algorithms, and implementers may also use other peak-finding algorithms, such as the derivative method, the Gaussian product function method, and the covariance method, etc. This application does not impose specific limitations.

[0029] Furthermore, the first-order forward difference value and the first-order backward difference value of each peak point are calculated, and the sum of the absolute values ​​of the two difference values ​​is calculated. The sum of the absolute values ​​and the sum of the amplitude of the peak point are used as the feature value of the peak point. The feature values ​​of all peak points in the signal detail coefficients of this layer are used as input, and the entropy value is calculated by using permutation entropy. The embedding dimension is set to 3 and the time delay is 1. These parameters are selected to cover a complete high-frequency cutting pulse cycle, thereby obtaining the permutation entropy of the feature values ​​of all peak points. Implementers can set the embedding dimension and time delay according to their actual situation; this application does not impose specific restrictions. The larger the obtained permutation entropy value, the more severely the periodicity and symmetry of the detail coefficients in the signal are disrupted, and the more significant the noise interference. Permutation entropy is a well-known technique and will not be elaborated further.

[0030] Furthermore, all feature values ​​obtained from the k-th layer of the cutting force signal in the i-th cycle are curve-fitted using the least squares method. Considering that the more severe the noise interference, the faster and more random the feature values ​​change, leading to ineffective fitting of the feature values ​​to the corresponding curve values, and the more severe the interference, the more random and large-amplitude the fluctuations, the more significant the oscillations. Based on this, the absolute value of the difference between each feature value and its corresponding point on the fitted curve is calculated, and the sum of all such absolute differences is taken as the residual feature value. When the residual eigenvalue is large, it indicates that the signal in that layer undergoes rapid, random, and large-amplitude changes. During curve fitting, the eigenvalue cannot be effectively fitted to the curve, and there is a significant deviation between the eigenvalue and the corresponding value on the curve, consistent with the characteristics of high-frequency noise interference caused by high-frequency impact loads. Calculate the permutation entropy. With residual eigenvalues The product of these two values ​​serves as a characteristic index of the impact load on the signal of that layer. .

[0031] The impact load characteristic indices of all layers of the cutting force signals from the most recent M periods in history are obtained by using Min-Max normalization. normalized value In this embodiment, the value of M is set to 100. When using Min-Max normalization, a minimal constant is added to the denominator. This avoids the denominator being zero, which would prevent calculation. In this embodiment... Implementers can set it themselves according to the actual situation. This application does not impose any specific restrictions.

[0032] Step S3: Construct correlation difference feature values ​​for each layer of signals based on the degree of difference between the detail coefficients of each layer of signals and the detail coefficients of each other layer of signals; construct local feature values ​​for each layer of signals based on the degree of disorder in the distribution of the number of detail coefficients among all adjacent peak points in each layer of signals and the degree of disorder in the distribution of the amplitude of all peak points; adjust each detail coefficient of each layer of signals based on the impact load feature index, the correlation difference feature values ​​and the local feature values, and then perform wavelet threshold denoising on the cutting force signal.

[0033] Furthermore, considering the presence of numerous high-power electrical devices such as frequency converters, servo drives, and high-power motors in the machining environment during the processing of heavy-duty truck transmission system components, these devices generate strong electromagnetic radiation during start-up, shutdown, or operation. This radiation may enter the sensor and signal lines of the force gauge through power lines or spatial coupling. In the presence of such high-frequency electromagnetic noise interference, the detail coefficients in the wavelet layer typically exhibit frequent and irregular random intensity spikes. The more severe the interference, the more pronounced these characteristics become. Simultaneously, since the cutting force generated during the cutting operation responds at different scales, under normal cutting force conditions, the detail coefficients of different layers show a high correlation. However, under external electromagnetic interference, because the noise is random and lacks correlation characteristics, the noise leads to a decrease in the correlation between different layers. The more severe the noise interference, the more severely the correlation between different layers is affected.

[0034] Based on the above analysis, firstly, all detail coefficients of each layer of signal are taken as the detail coefficient set of that layer. Taking the k-th layer signal of the i-th period as an example, the discrete Fréchet distances between the data in the detail coefficient set of that layer and the data in the detail coefficient sets of each of the other layers are calculated. Then, the coefficient of variation of all the discrete Fréchet distances of that layer is calculated as the correlation difference characteristic value of that layer signal. ,when A larger value indicates a greater difference in the correlation of detail coefficients between different layers, and a significant fluctuation in the correlation between the target layer and the other layers, consistent with a situation of significant external electromagnetic noise interference; when The smaller the value, the less the correlation between the target layer and the detail coefficients of the other layers fluctuates, which is consistent with the situation where it is less affected by external interference. Among them, the discrete Friesian distance is a well-known technique, and the specific process will not be described in detail.

[0035] It should be noted that this application provides only one distance metric algorithm for the difference distance between sets of detail coefficients. There are many existing distance metric algorithms, and implementers may also use other distance metric algorithms to calculate the difference distance between sets of detail coefficients. This application does not impose any specific restrictions.

[0036] Furthermore, considering that under severe external electromagnetic interference, the detail coefficients in wavelet layers often exhibit frequent and irregular random intensity spikes, based on this, taking the k-th layer signal of the i-th period as an example, we calculate the total number of detail coefficients between any two adjacent peak points in the detail coefficients of this layer signal, and calculate the standard deviation of the total number of detail coefficients between all peak points; simultaneously, we calculate the standard deviation of the amplitude of all peak points; after performing maximum and minimum value normalization on the two obtained standard deviations respectively, we sum the two normalized standard deviations to obtain the local eigenvalues ​​of the signal of this layer. . Acquired A larger value indicates that the random peak intervals and amplitude fluctuations occur frequently within this layer, consistent with the distribution characteristics of detail coefficients under severe external electromagnetic noise interference. Calculate local eigenvalues. Correlation differential eigenvalues The product is multiplied by a factor of 1, and a very small constant of 0.01 is added to the denominator to ensure that the denominator is not zero. This product is then used as an electromagnetic interference characteristic index of the signal at this layer. .

[0037] The electromagnetic interference characteristic indices of all layers of the cutting force signals over the most recent M periods are normalized using Min-Max normalization to obtain... normalized value When using Min-Max normalization, add a minimal constant to the denominator. This is to avoid the denominator being 0, which would prevent calculation from being possible.

[0038] Based on the above analysis, the adjustment coefficients for each layer of signals are constructed, and their expressions are as follows: ,in, is the adjustment coefficient of the k-th layer signal of the cutting force signal within the i-th period. As a preset weighting coefficient, in this embodiment, Set it to 0.5.

[0039] Considering that conventional filtering of wavelet layers after signal decomposition using wavelet coefficients results in a steeper transition when the transition adjustment parameter is large, approaching a hard threshold and preserving details effectively, it can easily lead to excessive noise retention in noisy environments. Conversely, a smaller transition adjustment parameter results in a smoother transition, approaching a soft threshold and providing better smoothing, but it can also lead to excessive filtering of details and loss of effective signal in signals less affected by interference. Therefore, the initial transition adjustment parameter in the algorithm is adaptively adjusted based on the adjustment coefficient to obtain the adaptive transition adjustment parameters for each signal layer. Preferably, in this embodiment, the expression for the adaptive transition adjustment parameter is: ,in, The adaptive transition adjustment parameter is the k-th layer signal of the cutting force signal within the i-th cycle. As the initial transition adjustment parameter, the value of b is set to the default value of 1 in this embodiment; The adjustment coefficient for the k-th layer signal of the cutting force signal within the i-th period is obtained by using... This ensures that the denominator ranges from [0.5, 1.5], thus avoiding invalid value ranges when adjusting the initial transition parameters.

[0040] Furthermore, the detail coefficients of each layer of signal are adaptively adjusted according to the adaptive transition adjustment parameters, as expressed in the following expression: In the formula, , These are the adjusted and unadjusted values ​​of the t-th detail coefficient in the k-th layer of the cutting force signal within the i-th period, respectively. It is a symbolic function; The wavelet threshold corresponding to the cutting force signal in the i-th period is obtained in this embodiment by using the fixed threshold rule (sqtwolog); In this embodiment, 'a' is a preset shape adjustment parameter, and 'a' takes the default value of 1. The adaptive transition adjustment parameter is the k-th layer signal of the cutting force signal within the i-th cycle; This is an exponential function with the natural constant e as the base. Among them, the fixed threshold method and... The calculation formulas are all well-known techniques, and the specific process will not be elaborated here.

[0041] Existing technologies typically set wavelet coefficients in wavelet threshold denoising algorithms based on expert experience. However, they fail to consider the challenges of predicting the lifespan of machine tool cutting tools used in heavy-duty truck transmission system components. During milling gears or turning shafts, the tool and workpiece are in intermittent contact, leading to frequent high-frequency impact loads and high-frequency noise interference. Furthermore, electromagnetic interference in the machining environment can severely affect the acquired signal, potentially causing significant deviations in subsequent machine tool lifespan predictions. Therefore, based on this prior knowledge, this application fully considers the distribution characteristics of the detail coefficients of each wavelet layer under different external electromagnetic interference intensities and high-frequency impact load interference caused by intermittent cutting. It adaptively adjusts the wavelet coefficients corresponding to different wavelet layers, effectively filtering the acquired cutting force signal to avoid serious deviations in subsequent predictions of the lifespan of machine tool cutting tools used in heavy-duty truck transmission system components.

[0042] The cutting force signal of the current period is used as the input to the wavelet threshold denoising algorithm. The detail coefficients in each layer of the current period's cutting force signal are calculated, and these detail coefficients are adjusted using the method described above. The adjusted detail coefficients are then used for subsequent denoising and reconstruction, outputting the filtered and denoised cutting force signal. Wavelet threshold denoising is a well-known technique, and its specific steps will not be elaborated further.

[0043] Meanwhile, the vibration signal, high-frequency stress wave converted electrical signal, and temperature signal acquired in the current cycle are filtered and denoised using a conventional wavelet threshold denoising algorithm. The filtered and denoised vibration signal, high-frequency stress wave converted electrical signal, and temperature signal are then output for subsequent life prediction operations of machine tool cutting tools used in the processing of heavy truck transmission system components.

[0044] Step S4: Based on the historically acquired denoised cutting force signal, vibration signal, high-frequency stress wave signal and temperature signal, a prediction model is trained, and the tool life is predicted based on the trained prediction model.

[0045] The process involves acquiring all historical cutting force, vibration, high-frequency stress wave, and temperature signals corresponding to machine tool tools from the start of their use to the occurrence of damage. Data from the no-load state is removed, and denoising is then performed using the methods described above. Further, the variances of the denoised cutting force signal, vibration signal, and high-frequency stress wave signal are calculated within the time interval between each temperature data point and the sampling time of its predecessor. A vector composed of the variances of these three signals within each temperature data point and its corresponding time interval is used as a feature vector. All feature vectors from the historical data are used as input, and a Long Short-Term Memory (LSTM) network is used to train the machine tool prediction model. The percentage of remaining machine tool service life corresponding to each historical feature vector is used as the label data for each feature vector. The input feature dimension is set to 4, corresponding to the aforementioned four-dimensional feature vector. The hidden layer dimension is set to 64, the stacking layer number to 1, the time step to 0.1s, the adamW optimizer is selected, and the training epochs are set to 100. Implementers can set the hidden layer dimension, stacking layer number, time step, optimizer, and training epochs according to actual conditions; this application does not impose specific restrictions. The corresponding machine tool life prediction model for heavy-duty truck transmission system components is obtained through training. Long Short-Term Memory (LSTM) networks are well-known technologies and will not be elaborated further.

[0046] Furthermore, the denoised cutting force, vibration, high-frequency stress wave corresponding electrical signals and temperature signals of the current cycle are obtained, and all feature vectors corresponding to the current cycle are obtained through the above feature vector construction method, and input into the above trained prediction model to perform real-time prediction of tool life.

[0047] A schematic diagram of the process for obtaining the characteristic indicators of impact load is shown below. Figure 2 As shown; a schematic diagram of the detail factor adjustment process is shown below. Figure 3 As shown.

[0048] In summary, this embodiment of the application collects cutting force signals during the operation of a machine tool, performs wavelet decomposition on the cutting force signals, constructs impact load characteristic indicators based on the peak point variation characteristics of the detail coefficients of each layer of signals obtained from the decomposition, and assesses the severity of high-frequency noise interference on the signal. By analyzing the differences in the correlation characteristics between different layers of signals and the degree of irregular fluctuation of the detail coefficients of each layer of signals, the severity of external electromagnetic interference on the cutting force signals is assessed. Furthermore, the transition adjustment parameters of each layer of signals are adaptively adjusted, and the adjustment value of each detail coefficient is calculated to perform wavelet threshold denoising on the cutting force signals. This improves the filtering effect of the wavelet denoising algorithm on noise interference in the cutting force signals, and avoids the problem that traditional wavelet threshold denoising uses fixed transition adjustment parameters for filtering operations, which does not consider the impact of high-frequency impact and external electromagnetic interference on the cutting force signals, and directly denoises, easily confusing the normal cutting signals of the tool with the cutting signals of tool damage, resulting in poor subsequent tool life prediction results.

Claims

1. A method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components, characterized in that, The method includes the following steps: Real-time acquisition of cutting force signals on the worktable, spindle bearing housing vibration signals, high-frequency stress wave signals, and tool head temperature signals during the operation of the machine tool; The cutting force data during no-load conditions is deleted. Wavelet decomposition is performed on the cutting force signals in each preset period of the remaining time to obtain the signals of each layer of the cutting force signals in each period. The amplitude distribution characteristics of each peak point in the detail coefficients of each layer of signals are determined. Based on the complexity and irregularity of the data changes of the amplitude distribution characteristics of all peak points, the impact load characteristic index of each layer of signals is constructed. The correlation difference feature value of each layer signal is constructed based on the degree of difference between the detail coefficients of each layer signal and the detail coefficients of each other layer signal; the local feature value of each layer signal is constructed based on the degree of disorder in the distribution of the number of detail coefficients among all adjacent peak points in each layer signal and the degree of disorder in the distribution of the amplitude of all peak points; each detail coefficient of each layer signal is adjusted based on the impact load feature index, the correlation difference feature value and the local feature value, and then wavelet threshold denoising is performed on the cutting force signal; A prediction model is trained based on historically collected denoised cutting force signals, vibration signals, high-frequency stress wave signals, and temperature signals. Tool life is then predicted based on the trained prediction model.

2. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 1, characterized in that, The process for obtaining the impact load characteristic indicators is as follows: Analyze the amplitude difference between each peak point and its two adjacent data points in each layer of signal, and determine the characteristic value of each peak point by combining the amplitude of each peak point; Calculate the permutation entropy of the eigenvalues ​​of all peak points in each layer of the signal; perform curve fitting on the eigenvalues ​​of all peak points in each layer of the signal, and calculate the residuals of all peak points on the obtained fitted line; The impact load characteristic index of each layer signal is determined based on the arrangement entropy and the residual.

3. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 2, characterized in that, The process for obtaining the feature values ​​of each peak point is as follows: Calculate the sum of the absolute values ​​of the amplitude differences between each peak point and its two left and right adjacent data points in each layer of signal, and use the sum of the amplitude of each peak point and the sum of the absolute values ​​as the characteristic value of each peak point.

4. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 2, characterized in that, The impact load characteristic index is proportional to the permutation entropy and the residual, respectively.

5. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 1, characterized in that, The process for obtaining the correlation difference feature values ​​is as follows: The set of all detail coefficients of each layer signal is denoted as the detail coefficient set; the degree of difference between each layer signal of the cutting force signal in the same period and the detail coefficient set of each other layer signal is calculated, and the coefficient of variation of all the said degree of difference of each layer signal is calculated as the correlation difference feature value of each layer signal.

6. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 1, characterized in that, The process for obtaining the local feature values ​​is as follows: Calculate the standard deviation of the number of detail coefficients between all adjacent peak points in each layer of the signal, and calculate the standard deviation of the amplitude of all peak points in each layer of the signal. Use the sum of the normalized values ​​of the two standard deviations as the local feature value of each layer of the signal.

7. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 1, characterized in that, The process of adjusting each detail coefficient of the signal at each layer is as follows: The adjustment coefficients of each layer of signal are constructed based on the correlation difference feature value, the local feature value, and the impact load feature index; Based on the adjustment coefficient, the initial transition adjustment parameter is adaptively adjusted to obtain the adaptive transition adjustment parameter of each layer signal, which is used to adjust each detail coefficient of each layer signal.

8. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 7, characterized in that, The adjustment coefficients are positively correlated with the correlation difference feature value, the local feature value, and the impact load feature index, respectively.

9. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 7, characterized in that, The process of obtaining the adaptive transition adjustment parameters is as follows: The adjustment coefficients of each layer signal are inversely proportionally mapped, and the product of the mapping result and the initial transition adjustment parameter is used as the adaptive transition adjustment parameter of each layer signal.

10. The method for predicting and managing the tool life of machine tools used in the processing of heavy-duty truck transmission system components as described in claim 1, characterized in that, The prediction model is trained based on the denoised cutting force signal, vibration signal, high-frequency stress wave signal, and temperature signal collected historically. Specifically: Historical signal data of various types are acquired, and data under no-load conditions are removed. Then, the variances of the denoised cutting force signal, the denoised vibration signal, and the denoised high-frequency stress wave signal are calculated within each temperature acquisition time interval. The vector composed of the variances of each temperature data and the three signals in the previous temperature acquisition time interval is used as a feature vector. All feature vectors in the historical data are used as inputs to the prediction model to be trained.

Images