A brake disc machining quality control method, device and medium based on hyperparameters

Abstract

This invention discloses a method, equipment, and medium for quality control of brake disc machining based on hyperparameters, relating to the field of machining process quality control technology. The method includes: establishing a data acquisition network to collect multi-source signals and establish relationships between process parameters; preprocessing the signals to extract multi-dimensional features related to roughness and constructing a machining feature vector; establishing a roughness prediction model based on the feature vector and measured samples and setting adjustable hyperparameters; dynamically adjusting the model during machining using hyperparameter optimization to obtain real-time roughness prediction values; comparing the predicted values ​​with target values ​​and correcting the machining parameters according to the optimized hyperparameter constraints. This invention achieves adaptive model updates by setting adjustable hyperparameters and employing Bayesian optimization, improving prediction accuracy and stability; and by introducing dynamic hyperparameter optimization and online correction, it achieves real-time deviation control and automatic parameter adjustment, ensuring that the brake disc roughness remains stable within the target range.

Description

Technical Field

[0001] This invention relates to the field of quality control technology in machining processes, and in particular to a method, equipment, and medium for quality control of brake disc machining based on hyperparameters. Background Technology

[0002] In the field of brake disc manufacturing, the stability and consistency of processing quality are directly related to the safety performance of automotive braking systems. Existing methods mainly rely on process parameter acquisition, surface roughness detection, and statistical modeling. During the brake disc processing, sensors are deployed and corresponding relationships are established with process parameters. Processing features are extracted using time-domain, frequency-domain, and time-frequency analysis methods. These features are then combined with measured roughness samples to establish a roughness prediction model, enabling the prediction and evaluation of brake disc surface quality. This type of method is widely used in industrial applications and provides important basic support for quality control in brake disc processing.

[0003] However, existing processing quality control still has many shortcomings. On the one hand, the hyperparameters of roughness prediction models are usually fixed during the training phase and lack dynamic optimization in the actual processing process, making it difficult for the prediction results to adapt to changes in working conditions in real time. On the other hand, the correction of processing parameters usually relies on experience or linear regression weights, which cannot make full use of the complex correlation between features. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for controlling the processing quality of brake discs based on hyperparameters, which solves the problems of static fixation of model hyperparameters and insufficient accuracy of processing parameter correction in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for controlling the quality of brake disc machining based on hyperparameters, which includes: building a brake disc machining data acquisition network, synchronously acquiring multi-source signals of brake disc machining, and establishing a correspondence between process parameters and multi-source signals of brake disc machining; The multi-source signals of brake disc machining are preprocessed to extract time-domain, frequency-domain, and time-frequency features related to surface roughness. Based on the correspondence between process parameters and brake disc machining signals, a machining process feature vector is constructed. Based on the feature vector of the processing process and the first measured roughness sample, a roughness prediction model is established. The hyperparameters are adjusted through performance indicators to obtain the optimized roughness prediction model and hyperparameters. The roughness prediction model is dynamically adjusted during the machining process using a hyperparameter optimization method to obtain the real-time predicted value of the brake disc surface roughness by the feature vector of the machining process. The real-time predicted value of the brake disc surface roughness is compared with the target roughness, and the machining process is adjusted according to the optimized hyperparameter constraints to achieve real-time machining parameter correction.

[0007] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the steps of establishing a brake disc machining data acquisition network, acquiring and synchronizing multi-source signals of brake disc machining, and establishing the correspondence between process parameters and multi-source signals of brake disc machining are as follows: Sensors are deployed in the brake disc machining area to form a brake disc machining data acquisition network; Multi-source signals from brake disc processing are simultaneously acquired via a brake disc processing data acquisition network. Baseline calibration is performed on multi-source signals from brake disc machining. The calibration coefficients between the calibrated brake disc machining multi-source signals and process parameters are calculated using correlation analysis. Based on the calibration coefficients, a correspondence between process parameters and multi-source signals for brake disc machining is established.

[0008] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the steps of extracting time-domain, frequency-domain, and time-frequency features related to surface roughness, and constructing a machining process feature vector based on the correspondence between process parameters and brake disc machining signals, are as follows: Preprocessing of multi-source signals from brake disc machining; The root mean square value, peak-to-peak value, kurtosis, impulse factor, and envelope energy of the pre-processed brake disc are calculated to generate a time-domain feature set. The pre-processed brake disc is used to process multi-source signals and perform fast Fourier transform to calculate the main frequency amplitude, bandwidth-to-energy ratio, and spectral centroid, generating a frequency domain feature set. The pre-processed brake disc is used to process multi-source signals and perform short-time Fourier transform to extract time-frequency energy and time-frequency entropy, generating a time-frequency feature set. The correlation weighting of the time-domain feature set, frequency-domain feature set, and time-frequency feature set with the correspondence between process parameters and multi-source signals of brake disc processing is performed to generate a weighted feature set; The weighted feature set is combined with the process parameters within the same sampling window time range and sorted in a fixed order to generate a process feature vector.

[0009] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the preprocessing includes sliding window segmentation, unified timestamp alignment, bandpass filtering, envelope demodulation, drift correction, and standardization.

[0010] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the steps of establishing a roughness prediction model based on the machining process feature vector and the first measured roughness sample, adjusting the hyperparameters through performance indicators, and obtaining the optimized roughness prediction model and hyperparameters are as follows: Align the feature vector of the machining process with the time index of the first measured roughness sample; The aligned machining process feature vector and the first measured roughness sample are divided and standardized to generate a standardized machining process feature vector and a standardized first measured roughness sample. The standardized processing feature vector is used as the input to the roughness prediction model, and the standardized first measured roughness sample is used as the label of the roughness prediction model. The roughness prediction model is established and the hyperparameters of the roughness prediction model are initialized. The roughness prediction model is used to iteratively train the hyperparameters on the training set to obtain the trained roughness prediction model and hyperparameters. The roughness prediction model is evaluated using a validation set, and the hyperparameters are adjusted based on the performance metrics to obtain the optimized roughness prediction model and hyperparameters.

[0011] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the step of dynamically adjusting the roughness prediction model during machining using a hyperparameter optimization method to obtain the real-time predicted value of the brake disc surface roughness by the feature vector of the machining process includes the following specific steps: By using the optimized hyperparameters, the roughness prediction model is used to predict the feature vector of the machining process and obtain the roughness prediction value. During random inspection, a second measured roughness sample corresponding to the feature vector of the processing process is obtained, and the error is calculated with the roughness prediction value to generate an instantaneous error. The instantaneous error is exponentially smoothed and averaged to generate an exponential moving average error. An error threshold is set based on the standard deviation of the measured surface roughness of the brake disc and the standard deviation of the residuals on the roughness prediction model validation set. The exponential moving average error is then compared with the error threshold. If the exponential moving average error is greater than the error threshold, a hyperparameter update request is issued. Based on the hyperparameter update request, the hyperparameters are optimized online using the restricted random perturbation approximation gradient method to generate updated hyperparameters; The updated hyperparameters are used to recalculate the real-time predicted value of the brake disc surface roughness based on the feature vector of the machining process, and the updated hyperparameters are used as the hyperparameters for the next cycle.

[0012] As a preferred embodiment of the brake disc processing quality control method based on hyperparameters described in this invention, the sampling inspection refers to randomly selecting a portion of brake disc samples from the finished or semi-finished brake discs that have come off the processing line during the brake disc processing, and performing surface roughness detection on the brake disc samples to obtain a second measured roughness sample.

[0013] As a preferred embodiment of the hyperparameter-based brake disc machining quality control method of the present invention, the steps of comparing the real-time predicted value of the brake disc surface roughness with the target roughness and adjusting the machining process according to the optimized hyperparameter constraints to achieve real-time machining parameter correction are as follows: The real-time deviation between the real-time predicted value of the brake disc surface roughness and the calculated target roughness is used to generate the real-time roughness deviation. Using the real-time roughness deviation and the optimized hyperparameters, the machining parameter adjustment vector is calculated based on the sensitivity weights and constraint ranges of the process parameters; The machining parameter adjustment vector is applied to the cutting speed, feed rate, and tool compensation to generate the corrected machining parameters. The machining process feature vector is regenerated using the corrected machining parameters and then compared with the real-time predicted value of the brake disc surface roughness in the next cycle.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the hyperparameter-based brake disc machining quality control method described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the hyperparameter-based brake disc machining quality control method described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by setting adjustable hyperparameters in the roughness prediction model and using Bayesian optimization, the roughness prediction model is adaptively updated under different working conditions, improving the accuracy and stability of roughness prediction; by introducing a hyperparameter dynamic optimization and online correction mechanism during the processing, deviation control of real-time prediction results and automatic adjustment of processing parameters are achieved, ensuring that the surface roughness of the brake disc remains stable within the target range. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for controlling the machining quality of brake discs based on hyperparameters.

[0019] Figure 2 A flowchart for establishing the data acquisition network and corresponding relationships for brake disc machining.

[0020] Figure 3 This is a flowchart for constructing feature vectors during the feature extraction and processing process.

[0021] Figure 4 A flowchart for hyperparameter optimization and online updates. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for controlling the machining quality of a brake disc based on hyperparameters, comprising the following steps: S1. Establish a data acquisition network for brake disc processing, synchronously acquire multi-source signals for brake disc processing, and establish the correspondence between process parameters and multi-source signals for brake disc processing.

[0026] Sensors are deployed in the brake disc machining area to form a brake disc machining data acquisition network.

[0027] Furthermore, on the brake disc machining tool, various sensor nodes are deployed at the spindle seat, tool clamping area, and workpiece clamping end to form a brake disc machining data acquisition network.

[0028] Among them, various sensor nodes include vibration sensor nodes, acoustic emission sensor nodes, cutting force sensor nodes, and temperature sensor nodes.

[0029] The brake disc machining data acquisition network synchronously collects multi-source signals from the brake disc machining process.

[0030] Furthermore, a unified master clock signal is set and hardware synchronization is performed with the machine tool spindle encoder pulse, so that the signals collected by all sensors have a unified timestamp. A signal buffer is set up in the brake disc machining data acquisition network to ensure that multi-source signals are acquired in parallel and stored synchronously in each sampling cycle.

[0031] Baseline calibration is performed on the multi-source signals from the brake disc machining.

[0032] Furthermore, the vibration sensor node measures the noise baseline under idling conditions through the vibration channel; the acoustic emission sensor node measures the background energy under no-load cutting conditions through the acoustic emission channel; the cutting force sensor node obtains the sensitivity matrix under cutting conditions on a standard test block through the cutting force channel; and the temperature sensor node performs baseline calibration through a thermocouple standard source.

[0033] Furthermore, under different cutting conditions, the brake disc machining data acquisition network, based on the baseline calibration, acquires multi-source signals.

[0034] The calibration coefficients between the calibrated brake disc machining multi-source signals and process parameters are calculated using correlation analysis.

[0035] Based on the calibration coefficients, a correspondence between process parameters and multi-source signals for brake disc machining is established.

[0036] Furthermore, correlation analysis methods are used to establish the correspondence between process parameters and multi-source signals.

[0037] Furthermore, to quantify the correspondence between process parameters and multi-source signals from brake disc machining, process parameters are defined within the sampling window. Multi-source signals for brake disc processing Corresponding correlation coefficient , is represented as: ; in, Indicates time The values ​​of the process parameters, This represents the average value of the process parameters within the window. Indicates time The brake disc is processed with multi-source signal values. This represents the average value of the multi-source signal values ​​during brake disc machining within the sampling window. Indicates the length of the sampling window.

[0038] It should be noted that the correlation coefficient The value range of is [-1, 1]. The closer the correlation coefficient is to 1, the stronger the positive correlation between the process parameters and the multi-source signals of brake disc processing. The closer the correlation coefficient is to -1, the stronger the negative correlation between the process parameters and the multi-source signals of brake disc processing. The closer the correlation coefficient is to 0, the less obvious the correlation between the two.

[0039] By mapping the process parameters, the multi-source signals for brake disc machining, and the corresponding correlation coefficients one by one, a correspondence between the process parameters and the multi-source signals for brake disc machining is established.

[0040] It should be noted that process parameters refer to adjustable process control quantities that play a decisive role in the machining quality during the brake disc machining process, including but not limited to cutting speed, feed rate, and tool compensation.

[0041] S2. Preprocess the multi-source signals of brake disc machining, extract the time domain, frequency domain and time-frequency features related to surface roughness, and construct the machining process feature vector based on the correspondence between process parameters and brake disc machining signals.

[0042] Preprocess the multi-source signals from the brake disc machining.

[0043] Further preprocessing includes sliding window segmentation, unified timestamp alignment, bandpass filtering, envelope demodulation, drift correction, and normalization.

[0044] The root mean square value, peak-to-peak value, kurtosis, impulse factor, and envelope energy of the pre-processed brake disc are calculated to generate a time-domain feature set.

[0045] Furthermore, for each sampling window, statistical quantities related to surface roughness are calculated by processing multi-source signals and accumulating window energy on each brake disc. These statistical quantities include root mean square value, peak-to-peak value, kurtosis, impulse factor, and envelope energy, generating a time-domain feature set.

[0046] The pre-processed brake disc is used to process multi-source signals and perform fast Fourier transform to calculate the main frequency amplitude, bandwidth-to-energy ratio, and spectral centroid, generating a frequency domain feature set.

[0047] Furthermore, a Fast Fourier Transform (FFT) is performed on the multi-source signal of the brake disc processing within the sampling window. Through FFT, power spectrum estimation, and energy ratio calculation, the main frequency amplitude, bandwidth-energy ratio, and spectral centroid are calculated to generate a frequency domain feature set.

[0048] The pre-processed brake disc is used to process multi-source signals and perform short-time Fourier transform to extract time-frequency energy and time-frequency entropy, generating a time-frequency feature set.

[0049] Furthermore, a short-time Fourier transform is performed on the multi-source signals of brake disc processing within the sampling window. Time-frequency energy and time-frequency entropy are extracted through CWT sub-band integration, energy normalization, and entropy measurement to generate a time-frequency feature set.

[0050] The correlation weighting of the time-domain feature set, frequency-domain feature set, and time-frequency feature set with the correspondence between process parameters and multi-source signals of brake disc processing is performed to generate a weighted feature set.

[0051] Furthermore, correlation coefficients are used to perform correlation weighting on time-domain features, frequency-domain features, and time-frequency features to generate a weighted feature set. Correlation weighting refers to the normalized weight calculation and weighted fusion of the absolute values ​​of the correlation coefficients between process parameters and multi-source signals of brake disc processing.

[0052] The normalized weight calculation and weighted fusion are expressed as follows: ; ; in, Indicates the first The normalized weights of each feature among all features, with values ​​ranging from [0,1]. Indicates the first The correlation coefficient between multiple source signals and process parameters in the machining of a brake disc Represents the total number of features. Indicates the first The correlation coefficient between multiple source signals and process parameters in the machining of a brake disc This represents the first signal extracted from the multi-source signal of brake disc machining. The original values ​​of each feature Indicates the first Weighted features.

[0053] The weighted feature set includes weighted time-domain features, weighted frequency-domain features, and weighted time-frequency features.

[0054] It should be noted that after normalization based on the sum of the absolute values ​​of the correlation coefficients, the weight of each feature is standardized into a proportional form, and the normalized weight ranges from [0,1].

[0055] The weighted feature set is combined with the process parameters within the same sampling window time range and sorted in a fixed order to generate a process feature vector.

[0056] Furthermore, within the same sampling time window, the weighted features in the weighted feature set are combined with the process parameters in a one-to-one correspondence in numerical structure, and then sorted in a fixed order.

[0057] The fixed order involves arranging different types of weighted features and their corresponding process parameters in the order of weighted time-domain features, weighted frequency-domain features, and weighted time-frequency features to generate a processing feature vector, ensuring that the dimension and position of the weighted feature vector for each processing window remain consistent.

[0058] S3. Based on the feature vector of the processing process and the first measured roughness sample, establish a roughness prediction model, adjust the hyperparameters through performance indicators, and obtain the optimized roughness prediction model and hyperparameters.

[0059] Align the feature vector of the machining process with the time index of the first measured roughness sample.

[0060] Furthermore, after the brake disc is machined, according to batch requirements, roughness values ​​are collected on the outer cylindrical surface and friction ring area using a surface roughness meter, and each roughness value is matched with the time window index of the brake disc machining data acquisition network to obtain the first measured roughness sample.

[0061] Furthermore, based on the fact that all multi-source signals collected by all sensors during the brake disc processing have a unified timestamp and are divided into sampling windows of fixed length, the processing feature vector and the first measured roughness sample in the corresponding sampling time window are matched one-to-one to form training sample pairs. The first measured roughness samples that do not overlap in the sampling time window are removed and the index is aligned.

[0062] The aligned machining process feature vector and the first measured roughness sample are divided and standardized to generate the standardized machining process feature vector and the standardized first measured roughness sample.

[0063] Furthermore, the aligned processing feature vector set and the aligned first measured roughness sample set are divided into training set, validation set and test set. Based on the training set, the standardized parameters are estimated to standardize the divided processing feature vector and the first measured roughness sample, generating standardized processing feature vector and standardized first measured roughness sample.

[0064] The standardized machining process feature vector is used as the input to the roughness prediction model, and the standardized first measured roughness sample is used as the label of the roughness prediction model. The roughness prediction model is established and its hyperparameters are initialized.

[0065] Furthermore, kernel ridge regression is used as the roughness prediction model, with the input being the processing feature vector of a single sampling time window and the output being the roughness prediction value of the corresponding sampling time window.

[0066] The label of the roughness prediction model refers to the label used to guide the roughness prediction model in learning the correspondence between the feature vectors of the processing process and the surface roughness.

[0067] Furthermore, the hyperparameters of the roughness prediction model include hyperparameter boundaries, window length, step length, kernel width, and regularization coefficient.

[0068] Specifically, the upper limit of the window length and the lower limit of the step length are determined based on the sampling rate and hardware computing power to obtain the hyperparameter boundaries of the roughness prediction model; the initial values ​​of the window length and step length are selected within the hyperparameter boundaries based on the main frequency band coverage and time-frequency resolution requirements; the median of the pairwise Euclidean distance is calculated for the feature vectors of the training set, and the initial value of the kernel width is obtained based on the median method; a log-interval regularization candidate set is constructed, and rapid evaluation is performed on a small validation set under the conditions of the initial values ​​of the window length, step length, and kernel width, and the one with the highest regularization score is selected as the initial value of the regularization coefficient.

[0069] The roughness prediction model is used to iteratively train the hyperparameters on the training set, recording the hyperparameter data of the current and previous rounds until the hyperparameters are stable or the number of iterations is reached, thus completing the training and obtaining the trained roughness prediction model and hyperparameters.

[0070] Furthermore, the standardized training set is used as the input to the roughness prediction model. Based on the kernel ridge regression optimization objective, the kernel ridge regression coefficients are solved, and the kernel width and regularization coefficient are registered as hyperparameters of the roughness prediction model.

[0071] The kernel ridge regression optimization objective is expressed as: ; in, Describe the objective function. This indicates the number of the first measured roughness sample. Indicates the first The value of the first measured roughness sample. Indicates the first The roughness prediction value of the machining process feature vector corresponding to the first measured roughness sample. Represents the regularization coefficient. This represents the fitting function in the kernel function space.

[0072] It should be noted that the fitting function is expressed as: ; in, Indicates the first The kernel ridge regression coefficients of the machining process feature vectors corresponding to the first measured roughness sample, with values ​​ranging from [0,1]. Indicates the first training set The feature vector of the machining process corresponding to the first measured roughness sample. This represents the feature vector of the processing procedure to be predicted.

[0073] It should be noted that, based on the optimization objective of kernel ridge regression, the kernel ridge regression coefficients of the feature vectors in the training set are normalized to make the kernel ridge regression coefficients a weighting factor to express the feature contribution ratio, and the range of kernel ridge regression coefficients is set to [0,1].

[0074] The roughness prediction model is evaluated using a validation set, and the hyperparameters are adjusted based on the performance metrics to obtain the optimized roughness prediction model and hyperparameters.

[0075] Furthermore, the mean absolute error of the validation set is used as a performance evaluation metric. Bayesian optimization is employed to search for the trained hyperparameters within the hyperparameter boundaries to obtain the optimal roughness prediction model and hyperparameters.

[0076] Furthermore, Bayesian optimization is employed to search for trained hyperparameters within the hyperparameter boundary. Specifically, the hyperparameter boundary is used as the search space for Bayesian optimization; the mean absolute error of the validation set is used as a performance evaluation metric to construct the optimization objective function; initial sampling points are selected in the search space, and the optimization objective function values ​​at the initial sampling points are used to perform a Gaussian process proxy to approximate the optimization objective function values ​​throughout the entire search space, thus obtaining the expression function; new hyperparameters are selected at the candidate points with the largest expression function values, and the optimization objective function values ​​are calculated using the validation set; then, the process of obtaining the expression function, selecting candidate points, and calculating the optimization objective function values ​​is repeated for the new hyperparameters and their corresponding optimization objective function values ​​until the error of the optimization objective function values ​​calculated between the validation sets no longer decreases, thus obtaining the optimized hyperparameters and the roughness prediction model corresponding to the hyperparameters.

[0077] The objective function of Bayesian optimization is expressed as: ; in, This represents the objective function of Bayesian optimization. Indicates the number of validation sets. Indicates the first Roughness prediction value of each processing process feature vector. Indicates the first The value of the first measured roughness sample corresponding to the feature vector of each processing process.

[0078] S4. The roughness prediction model is dynamically adjusted during the processing by using hyperparameter optimization methods to obtain the real-time predicted value of the surface roughness of the brake disc by the feature vector of the processing process.

[0079] By using the optimized hyperparameters, the roughness prediction model is used to predict the feature vector of the machining process and obtain the roughness prediction value.

[0080] Furthermore, the optimized hyperparameters are set into the roughness prediction model, and the roughness prediction value is obtained by predicting the roughness of the feature vector of the machining process through the roughness prediction model.

[0081] During random inspection, a second measured roughness sample corresponding to the feature vector of the processing process is obtained, and the error is calculated with the roughness prediction value to generate an instantaneous error.

[0082] Furthermore, during the brake disc processing, a sample of finished or semi-finished brake discs is randomly selected from the finished or semi-finished brake discs produced from the processing line, and the surface roughness of the brake disc samples is tested to obtain a second measured roughness sample.

[0083] Based on the second measured roughness sample, the instantaneous error between the roughness prediction value and the second measured roughness sample is calculated.

[0084] Specifically, the instantaneous error is calculated by comparing the predicted roughness value with the value of the second measured roughness sample in the same sampling time window, calculating the difference between the two, and taking the absolute value of the difference as the instantaneous error.

[0085] The instantaneous error is exponentially smoothed and averaged to generate an exponential moving average error.

[0086] Furthermore, after obtaining the instantaneous error at the current moment, the instantaneous error is weighted and averaged with the exponential moving average error at the previous moment to generate the exponential moving average error.

[0087] An error threshold is set based on the standard deviation of the measured surface roughness of the brake disc and the standard deviation of the residuals on the roughness prediction model validation set. The exponential moving average error is then compared with the error threshold.

[0088] Furthermore, in the brake disc surface roughness detection process, the surface roughness of the same brake disc is repeatedly measured, and the standard deviation of the brake disc surface roughness is obtained through statistical analysis.

[0089] The residuals between the roughness prediction values ​​and the measured roughness samples are calculated on the validation set of the roughness prediction model, and the variance of the residuals is calculated to obtain the standard deviation of the roughness prediction model.

[0090] Calculate the composite standard deviation based on the measurement standard deviation and the residual standard deviation, and use the composite standard deviation as the error threshold.

[0091] The error threshold is expressed as: ; in, Indicates the error threshold. Indicates the standard deviation of the measurement. This represents the standard deviation of the residuals.

[0092] If the exponential moving average error is greater than the error threshold, a hyperparameter update request is issued.

[0093] Based on the hyperparameter update request, the hyperparameters are optimized online using the restricted random perturbation approximation gradient method to generate updated hyperparameters.

[0094] Furthermore, based on the received hyperparameter update request, the hyperparameters in the current roughness prediction model are evaluated by two-point sampling and updated stepwise to generate updated hyperparameters.

[0095] Furthermore, the two-point sampling evaluation specifically involves generating a random direction vector based on the current roughness prediction model hyperparameters, and adding a perturbation amplitude to both the positive and negative directions of the randomly generated direction vector to obtain two new hyperparameter candidate points. The roughness prediction model is then run on the two new hyperparameter candidate points to obtain new roughness prediction values, and performance evaluation values ​​are calculated on the validation set. By comparing the differences between the two performance evaluation values, the performance change trend of the hyperparameters in the current random direction is determined, thereby approximately estimating the optimization direction of the hyperparameters.

[0096] Furthermore, the step update specifically optimizes the direction of the approximate estimated hyperparameters obtained from the two-point sampling evaluation, with a learning rate of [missing information]. The current hyperparameters are updated using the step size.

[0097] The current hyperparameter update formula is expressed as:

[0098] in, This indicates the updated hyperparameters. This indicates the hyperparameters before the update. Indicates the learning rate. This represents the approximate gradient vector obtained from the two-point sampling evaluation.

[0099] Meanwhile, to prevent hyperparameters from going out of bounds, the updated hyperparameters are subjected to boundary projection to confine them within the hyperparameter boundaries.

[0100] The updated hyperparameters are used to recalculate the real-time predicted value of the brake disc surface roughness based on the feature vector of the machining process, and the updated hyperparameters are used as the hyperparameters for the next cycle.

[0101] S5. Compare the real-time predicted value of the brake disc surface roughness with the target roughness, and adjust the machining process according to the optimized hyperparameter constraints to achieve real-time machining parameter correction.

[0102] The real-time deviation between the real-time predicted value of the brake disc surface roughness and the calculated target roughness is used to generate the real-time roughness deviation.

[0103] Furthermore, by utilizing the real-time roughness deviation and the optimized hyperparameters, and based on the sensitivity weights and constraint ranges of the process parameters, the machining parameter adjustment vector is calculated.

[0104] The machining parameter adjustment vector includes the cutting speed adjustment, feed rate adjustment, and tool compensation adjustment.

[0105] Furthermore, based on the calculated real-time roughness deviation, the cutting speed adjustment, feed rate adjustment, and tool compensation adjustment are calculated respectively through the hyperparameter constraint function according to the sensitivity weights corresponding to different process parameters.

[0106] The hyperparameter constraint function is expressed as: ; in, Indicates the first The processing parameter adjustment vector corresponding to each process parameter. Indicates the first Sensitivity weights of each process parameter Indicates the real-time deviation of surface roughness. Indicates the first The lower limit of the adjustment range of each process parameter Indicates the first The upper limit of the adjustment range of each process parameter This represents the truncation function.

[0107] Among them, the hyperparameter constraint function The truncation function expansion is expressed as: ; It should be noted that the upper and lower limits of the process parameter adjustment range are determined by the physical capabilities of the machine tool and cutting tool.

[0108] The sensitivity weight of process parameters refers to the linear regression between process parameters and measured roughness by collecting a large number of process feature vectors and measured roughness samples during the processing, and using the linear regression coefficient as the sensitivity weight of the process parameters.

[0109] The machining parameter adjustment vector is applied to the cutting speed, feed rate, and tool compensation to generate the corrected machining parameters.

[0110] The machining process feature vector is regenerated using the corrected machining parameters and then compared with the real-time predicted value of the brake disc surface roughness in the next cycle.

[0111] This embodiment also provides a computer device applicable to the case of a brake disc machining quality control method based on hyperparameters, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the brake disc machining quality control method based on hyperparameters as proposed in the above embodiment.

[0112] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0113] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the hyperparameter-based brake disc machining quality control method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0114] In summary, this invention achieves adaptive updating of the roughness prediction model under different working conditions by setting adjustable hyperparameters in the roughness prediction model and adopting Bayesian optimization method, thereby improving the accuracy and stability of roughness prediction. By introducing a dynamic optimization and online correction mechanism for hyperparameters during the processing, deviation control of real-time prediction results and automatic adjustment of processing parameters are achieved, ensuring that the surface roughness of the brake disc remains stable within the target range.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling the machining quality of brake discs based on hyperparameters, characterized in that: include, Establish a data acquisition network for brake disc processing, synchronously acquire multi-source signals for brake disc processing, and establish the correspondence between process parameters and multi-source signals for brake disc processing; The multi-source signals of brake disc machining are preprocessed to extract time-domain, frequency-domain, and time-frequency features related to surface roughness. Based on the correspondence between process parameters and brake disc machining signals, a machining process feature vector is constructed. Based on the feature vector of the processing process and the first measured roughness sample, a roughness prediction model is established. The hyperparameters are adjusted through performance indicators to obtain the optimized roughness prediction model and hyperparameters. The roughness prediction model is dynamically adjusted during the machining process using a hyperparameter optimization method to obtain the real-time predicted value of the brake disc surface roughness by the feature vector of the machining process. The real-time predicted value of the brake disc surface roughness is compared with the target roughness, and the machining process is adjusted according to the optimized hyperparameter constraints to achieve real-time machining parameter correction. The method of dynamically adjusting the roughness prediction model during the processing by using hyperparameter optimization to obtain the real-time predicted value of the surface roughness of the brake disc by the feature vector of the processing process is as follows: using the optimized hyperparameters, the roughness prediction model is used to predict the feature vector of the processing process to obtain the roughness prediction value. During random inspection, a second measured roughness sample corresponding to the feature vector of the processing process is obtained, and the error is calculated with the roughness prediction value to generate an instantaneous error. The instantaneous error is exponentially smoothed and averaged to generate an exponential moving average error. An error threshold is set based on the standard deviation of the measured surface roughness of the brake disc and the standard deviation of the residuals on the roughness prediction model validation set. The exponential moving average error is then compared with the error threshold. If the exponential moving average error is greater than the error threshold, a hyperparameter update request is issued. Based on the hyperparameter update request, the hyperparameters are optimized online using the restricted random perturbation approximation gradient method to generate updated hyperparameters; The updated hyperparameters are used to recalculate the real-time predicted value of the brake disc surface roughness based on the feature vector of the machining process, and the updated hyperparameters are used as the hyperparameters for the next cycle.

2. The brake disc machining quality control method based on hyperparameters as described in claim 1, characterized in that: The steps involved establishing a data acquisition network for brake disc processing, collecting and synchronizing multi-source signals from brake disc processing, and establishing a correspondence between process parameters and these multi-source signals. Sensors are deployed in the brake disc machining area to form a brake disc machining data acquisition network; Multi-source signals from brake disc processing are simultaneously acquired via a brake disc processing data acquisition network. Baseline calibration is performed on multi-source signals from brake disc machining. The calibration coefficients between the calibrated brake disc machining multi-source signals and process parameters are calculated using correlation analysis. Based on the calibration coefficients, a correspondence between process parameters and multi-source signals for brake disc machining is established.

3. The brake disc machining quality control method based on hyperparameters as described in claim 2, characterized in that: The extraction of time-domain, frequency-domain, and time-frequency features related to surface roughness, and the construction of a machining process feature vector based on the correspondence between process parameters and brake disc machining signals, are detailed in the following steps: Preprocessing of multi-source signals from brake disc machining; The root mean square value, peak-to-peak value, kurtosis, impulse factor, and envelope energy of the pre-processed brake disc are calculated to generate a time-domain feature set. The pre-processed brake disc is used to process multi-source signals and perform fast Fourier transform to calculate the main frequency amplitude, bandwidth-to-energy ratio, and spectral centroid, generating a frequency domain feature set. The pre-processed brake disc is used to process multi-source signals and perform short-time Fourier transform to extract time-frequency energy and time-frequency entropy, generating a time-frequency feature set. The correlation weighting of the time-domain feature set, frequency-domain feature set, and time-frequency feature set with the correspondence between process parameters and multi-source signals of brake disc processing is performed to generate a weighted feature set; The weighted feature set is combined with the process parameters within the same sampling window time range and sorted in a fixed order to generate a process feature vector.

4. The brake disc machining quality control method based on hyperparameters as described in claim 3, characterized in that: The preprocessing includes sliding window segmentation, unified timestamp alignment, bandpass filtering, envelope demodulation, drift correction, and normalization.

5. The brake disc machining quality control method based on hyperparameters as described in claim 4, characterized in that: The roughness prediction model is established based on the feature vector of the processing process and the first measured roughness sample. The hyperparameters are then adjusted using performance indicators to obtain the optimized roughness prediction model and hyperparameters. The specific steps are as follows: Align the feature vector of the machining process with the time index of the first measured roughness sample; The aligned machining process feature vector and the first measured roughness sample are divided and standardized to generate a standardized machining process feature vector and a standardized first measured roughness sample. The standardized processing feature vector is used as the input to the roughness prediction model, and the standardized first measured roughness sample is used as the label of the roughness prediction model. The roughness prediction model is established and the hyperparameters of the roughness prediction model are initialized. The roughness prediction model is used to iteratively train the hyperparameters on the training set to obtain the trained roughness prediction model and hyperparameters. The roughness prediction model is evaluated using a validation set, and the hyperparameters are adjusted based on the performance metrics to obtain the optimized roughness prediction model and hyperparameters.

6. The method for controlling the machining quality of brake discs based on hyperparameters as described in claim 5, characterized in that: The sampling inspection refers to randomly selecting a portion of brake disc samples from the finished or semi-finished brake discs that have come off the production line during the brake disc processing, and then performing surface roughness testing on the brake disc samples to obtain a second measured roughness sample.

7. The brake disc machining quality control method based on hyperparameters as described in claim 6, characterized in that: The process involves comparing the real-time predicted value of the brake disc surface roughness with the target roughness, and adjusting the machining process based on the optimized hyperparameter constraints to achieve real-time machining parameter correction. The specific steps are as follows: The real-time deviation between the real-time predicted value of the brake disc surface roughness and the calculated target roughness is used to generate the real-time roughness deviation. Using the real-time roughness deviation and the optimized hyperparameters, the machining parameter adjustment vector is calculated based on the sensitivity weights and constraint ranges of the process parameters; The machining parameter adjustment vector is applied to the cutting speed, feed rate, and tool compensation to generate the corrected machining parameters. The machining process feature vector is regenerated using the corrected machining parameters and then compared with the real-time predicted value of the brake disc surface roughness in the next cycle.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the brake disc machining quality control method based on hyperparameters as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the brake disc machining quality control method based on hyperparameters as described in any one of claims 1 to 7.

Images