A method and system for predicting the quality of precision angular contact ball bearings

Abstract

This invention discloses a method and system for predicting the quality of precision angular contact ball bearings. By acquiring machine tool processing parameters and vibration signals throughout the entire processing process in real time, and relying on online data processing and prediction models, quality prediction can be completed synchronously during the processing steps. This effectively avoids the problem of workpiece scrapping and rework discovered after the process, reduces the ineffective waste of raw materials, labor time, and equipment energy consumption, and improves the production efficiency and cost control capabilities of precision bearing mass production. Addressing the complex characteristics of bearing processing vibration signals—nonlinearity, non-stationarity, and multi-scale—a dual-path extraction architecture of deep learning automatic features and explicit features is adopted. On the one hand, multi-scale convolutional neural networks autonomously mine the deep correlation features hidden in the signal, avoiding the limitations of manual feature design. On the other hand, multi-dimensional explicit features in the time domain, frequency domain, and time-frequency domain are integrated, taking into account both the physical representation law of the signal and deep abstract information, significantly improving the completeness and accuracy of signal feature mining.

Description

Technical Field

[0001] The present invention relates to the technical fields of intelligent manufacturing of bearings, signal processing and quality prediction, and particularly relates to a quality prediction method and system for precision angular contact ball bearings. Background Art

[0002] The processing process of precision angular contact ball bearings involves multiple processes, various processing parameters and complex machine tool dynamics. There are many factors affecting the processing quality, including but not limited to spindle speed, feed speed, cutting or grinding parameters, cooling and lubrication state, machine tool vibration state, force change at the processing site, and environmental noise disturbance, etc. Especially under the conditions of high-speed and high-precision machining, the machine tool and workpiece system will generate complex vibration responses, and such vibration signals usually have the characteristics of non-stationarity, non-linearity, strong coupling and multi-scale changes. At the same time, the quality parameters of bearing processing often are not independent of each other, but there are obvious coupling relationships and co-variation characteristics, which makes it difficult for traditional quality judgment methods based on single-factor analysis or empirical rules to accurately reflect the actual processing state, restricting the further improvement of the manufacturing level of precision bearings.

[0003] At present, in the actual production process of enterprises, the quality control of bearing processing still depends more on the post-processing detection method, that is, after the workpiece completes the key process or the final processing, surface roughness, roundness error, dimensional deviation, contact angle deviation, radial runout and other quality indicators are obtained through off-line detection means, and then it is judged whether it is qualified according to the detection results, or the previous process parameters are traced and adjusted. This kind of method belongs to post-detection, with obvious hysteresis, unable to timely detect the quality deviation trend during the processing, and even more difficult to achieve early warning and active intervention of quality problems. Once the quality is found to be out of tolerance after processing, it will not only cause workpiece repair or even scrapping, but also cause waste of equipment, working hours and material costs, thus affecting production efficiency and product consistency.

[0004] The existing precision angular contact ball bearing processing quality control technologies still have the following defects: First, it mainly depends on post-processing detection, lacking the ability of early prediction and real-time warning for the processing process; second, the traditional machine learning methods have insufficient ability to mine the characteristics of complex vibration signals and multi-source processing data, and it is difficult to meet the high-precision quality prediction requirements; third, the existing methods do not fully describe the complex coupling relationships among processing parameters, vibration signals and various types of feature information, and cannot effectively support the joint prediction of multiple quality indicators; fourth, there is a lack of a unified modeling method and an extensible prediction framework suitable for massive heterogeneous processing data in actual engineering. Therefore, it is necessary to propose an intelligent quality prediction method for the processing process of precision angular contact ball bearings to solve the problems of quality prediction lag, insufficient accuracy, poor adaptability and difficulty in supporting subsequent process optimization in the existing technology. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting the quality of precision angular contact ball bearings, overcoming the problems of existing technologies, such as reliance on post-processing inspection, delayed detection of quality problems, difficulty in fully exploring the deep nonlinear relationship between complex vibration signals and processing parameters using traditional machine learning methods, and difficulty in meeting the requirements for high-precision joint prediction of multiple quality indicators. This invention can address the massive processing data generated in actual production processes by uniformly analyzing, preprocessing, extracting features, and fusing and modeling multi-source heterogeneous information such as processing parameters, vibration signals, and quality indicators. This enables high-precision joint prediction of multiple processing quality indicators such as bearing roughness, roundness, and shape, and provides a basis for subsequent processing quality assessment, anomaly warning, and processing parameter optimization.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for predicting the quality of precision angular contact ball bearings includes the following steps: Obtain the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted; The acquired vibration signal is subjected to dual-path feature extraction. One path uses a multi-scale convolutional neural network to automatically extract deep features from the vibration signal to obtain automatic deep features; the other path extracts explicit features from the vibration signal to obtain explicit features. The extracted automatic deep features and explicit features are fused from multiple sources to obtain the comprehensive features of the vibration signal. The comprehensive features of the vibration signal are then fused with the machine tool processing parameters from multiple sources. A predictive model is used to predict the features after multi-source fusion to obtain the corresponding processing quality parameters, which include roughness, roundness and shape.

[0007] Preferably, the machine tool machining parameters and vibration signals during the machining process are obtained from the JSON data file generated during the machining of the precision angular contact ball bearing, and the obtained machine tool machining parameters and vibration signals during the machining process are preprocessed to obtain standardized vibration signals.

[0008] Preferably, the vibration signal is preprocessed, specifically including: Perform signal dimensionality reduction on the original vibration signal; Extract the XYZ three-channel vibration signals; The XYZ three-channel vibration signals are combined with the processing position information; The merged signal undergoes a first DC component removal process. Envelope demodulation is performed on the signal after DC removal; Wavelet denoising is performed on the envelope-demodulated signal; Bandpass filtering is applied to the wavelet-denoised signal. Min-max normalization is applied to the bandpass filtered signal; The standardized signal undergoes a second DC component removal process. The normalized vibration signal was obtained by performing PCA principal component analysis on the signal after the second DC removal.

[0009] Preferably, the explicit features include time-domain features, frequency-domain features, and time-frequency-domain features; Temporal characteristics include: mean, standard deviation, variance, root mean square value, peak value, peak-to-peak value, mean absolute value, signal energy, shape factor, peak factor, impulse factor, and sharpness factor.

[0010] Preferably, the frequency domain features include: power spectral density, dominant frequency, center frequency, frequency bandwidth, total energy, harmonic frequency components, spectral mean, spectral variance, spectral kurtosis, spectral skewness, spectral root mean square value, and spectral peak / mean ratio.

[0011] Preferably, the time-frequency domain features are 24 features used to characterize the local energy distribution, frequency evolution law, and non-stationary change characteristics of the vibration signal.

[0012] Preferably, the prediction model is trained using a multi-source fusion dataset. The acquisition of the multi-source fusion dataset specifically includes: obtaining machine tool machining parameters and vibration signals during the machining process from a JSON data file generated during the machining of precision angular contact ball bearings; preprocessing the obtained machine tool machining parameters and vibration signals during the machining process; automatically extracting deep features and explicit features from the preprocessed vibration signals; fusing the automatic deep features and explicit features to form a comprehensive vibration signal feature; and performing multi-source fusion with the preprocessed machine tool machining parameters to obtain the multi-source fusion dataset for machining quality prediction input. The prediction model uses roughness, roundness, and shape as the output labels of the neural network.

[0013] Preferably, the prediction model includes: The Transformer module is used to extract global correlations between features from multiple sources. The LSTM module is used to model the temporal dynamic changes during the processing. A multidimensional linear output layer is used to achieve joint regression output of multiple processing quality parameters.

[0014] Preferably, the method further includes generating quality assessment results, anomaly warning information, and quality risk identification results based on the prediction results of the processing quality parameters.

[0015] A precision angular contact ball bearing quality prediction system includes a data acquisition module, a feature extraction module, a fusion module, and a prediction module; The data acquisition module acquires the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted. The feature extraction module performs dual-path feature extraction on the acquired vibration signal. One path uses a multi-scale convolutional neural network to automatically extract deep features from the vibration signal to obtain automatic deep features; the other path performs explicit feature extraction on the vibration signal to obtain explicit features. The fusion module performs multi-source feature fusion of the extracted automatic deep features and explicit features to obtain comprehensive vibration signal features, and then performs multi-source fusion of the comprehensive vibration signal features with machine tool processing parameters. The prediction module uses a prediction model to predict the corresponding processing quality parameters based on the multi-source fusion features. The processing quality parameters include roughness, roundness, and shape.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: This invention discloses a quality prediction method for precision angular contact ball bearings. By real-time acquisition of machine tool processing parameters and vibration signals throughout the entire processing process, and relying on online data processing and prediction models, quality prediction can be completed synchronously during the processing steps. This effectively avoids the problem of workpiece scrapping and rework discovered after the process, reduces the ineffective waste of raw materials, labor time, and equipment energy consumption, and improves the production efficiency and cost control capabilities of precision bearing mass production. Addressing the complex characteristics of bearing processing vibration signals—nonlinearity, non-stationarity, and multi-scale—a dual-path extraction architecture of deep learning automatic features and explicit features is adopted. On the one hand, multi-scale convolutional neural networks autonomously mine the deep correlation features hidden in the signal, avoiding the limitations of manual feature design. On the other hand, multi-dimensional explicit features from the time domain, frequency domain, and time-frequency domain are integrated, taking into account both the physical representation law of the signal and deep abstract information, compensating for the information omissions and one-sided representation defects of single feature extraction methods, and significantly improving the completeness and accuracy of signal feature mining.

[0017] Preferably, a dual fusion strategy is employed, combining internal vibration feature fusion and cross-source fusion of vibration features and machine tool processing parameters, to integrate two core influencing factors: process parameters and equipment vibration. This approach effectively fits the complex mechanism of strong coupling and synergistic changes among multiple variables during precision bearing machining. Simultaneously, a hybrid prediction architecture combining Transformer global correlation mining and LSTM temporal feature modeling adapts to the dynamic temporal changes in the machining process, effectively addressing the shortcomings of traditional methods that cannot analyze multi-factor coupling relationships and suffer from poor model fit.

[0018] Preferably, roughness, roundness, and shape are used as the three core key quality parameters for simultaneous prediction output, breaking through the limitations of traditional single quality index detection and realizing the integrated evaluation of key form and position accuracy and surface quality of precision angular contact ball bearings. Relying on a complete signal optimization process including standardized preprocessing, PCA dimensionality reduction, wavelet denoising, and bandpass filtering, environmental noise and data bias interference are eliminated. Combined with a prediction model trained on a standardized dataset, the accuracy and stability of simultaneous prediction of multiple quality indicators are significantly improved.

[0019] Preferably, data is read based on the raw JSON format generated during processing, and a multi-step joint preprocessing mechanism is configured to achieve unified and standardized processing of multi-source heterogeneous processing data, adapting to the access needs of massive and multi-type production data in industrial sites. The entire methodology is standardized and modularized, which can be quickly adapted to different processing conditions and production scenarios of precision bearings of different specifications. It has strong scalability and is easy to deploy in industrial settings. Based on the quality parameter prediction results, quality assessment, anomaly warning, and risk identification results are generated simultaneously, forming a closed-loop management system for the entire process of data collection, feature analysis, intelligent prediction, and risk warning. It can identify the deviation trend of processing quality in advance, providing data support for real-time adjustment of on-site process parameters and intervention in equipment status maintenance, reducing the probability of defective products from the source, ensuring the consistency and stability of the processing quality of precision angular contact ball bearings, and contributing to the intelligent manufacturing upgrade of bearings. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of a method for predicting the quality of a precision angular contact ball bearing according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating multi-source processing data parsing and sample construction in an embodiment of the present invention; Figure 3 This is a flowchart of vibration signal preprocessing in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the explicit time-domain, frequency-domain, and time-frequency-domain feature extraction of vibration signals in an embodiment of the present invention. Figure 5 This is a schematic diagram of a multi-scale convolutional neural network structure in an embodiment of the present invention; Figure 6 This is a schematic diagram of the composite neural network structure for predicting the machining quality of precision angular contact ball bearings in an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] This invention extracts multi-source data such as machine tool processing parameters, vibration signals, and quality indicators during the bearing processing. It performs noise reduction, depth feature extraction, and time-frequency domain feature extraction on the vibration signals, and combines them with a composite neural network model to achieve joint prediction of multiple quality parameters. This invention can be applied to quality assessment, anomaly warning, and parameter optimization in the processing of precision angular contact ball bearings.

[0024] like Figure 1 As shown, the present invention provides a method for predicting the quality of precision angular contact ball bearings, specifically including the following steps: Obtain the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted; The acquired vibration signal is subjected to dual-path feature extraction. One path uses a multi-scale convolutional neural network to automatically extract deep features from the vibration signal to obtain automatic deep features; the other path extracts explicit features from the vibration signal to obtain explicit features. The extracted automatic deep features and explicit features are fused from multiple sources to obtain the comprehensive features of the vibration signal. The comprehensive features of the vibration signal are then fused with the machine tool processing parameters from multiple sources. A prediction model is used to predict the features after multi-source fusion to obtain the corresponding roughness, roundness and shape prediction results.

[0025] In a specific embodiment of the present invention, machine tool machining parameters and vibration signals during the machining process are obtained from a JSON data file generated during the machining of a precision angular contact ball bearing, and the obtained machine tool machining parameters and vibration signals during the machining process are preprocessed to obtain standardized vibration signals.

[0026] In a specific embodiment of the present invention, a JSON data file generated during the machining process of a precision angular contact ball bearing is obtained, and the data file is parsed to extract multi-source information corresponding to the bearing machining process. The multi-source information includes at least machine tool machining parameters, vibration signals of different machining parts, workpiece identification information, and corresponding quality inspection labels. Based on the workpiece number, process number, timestamp, and data field mapping relationship, the association and sample-level reorganization of the multi-source data are completed to establish an original machining quality sample library.

[0027] Data preprocessing is performed on the original processing quality sample library. The data preprocessing includes: The original vibration signal is subjected to signal dimensionality reduction processing, and the vibration signals of the XYZ three channels are extracted; The vibration signal is combined with the processing position information to maintain the correspondence between the signal and the position state during the processing. The vibration signal is first processed to remove the DC component, so as to reduce the impact of DC bias on subsequent feature extraction. Envelope demodulation of the vibration signal enhances the modulation information reflecting changes in processing status; Wavelet denoising is applied to the vibration signal to suppress complex industrial noise and random interference; By using a bandpass filter, the target frequency band signal related to the processing vibration characteristics is retained, while irrelevant frequency band interference is filtered out; Min-max normalization is used to normalize the signal amplitude to eliminate dimensional differences and improve the stability of model training. The standardized signal is then subjected to a second DC component removal process to further reduce the impact of baseline drift on signal analysis. Principal component analysis (PCA) was used to extract principal components and compress information from the preprocessed multidimensional signal in order to reduce redundancy and highlight the main vibration characteristics. After signal processing, a standardized vibration signal dataset is obtained for subsequent feature extraction and model training. This preprocessing step effectively improves vibration signal quality, enhances feature effectiveness, and improves the stability and accuracy of multi-source fusion modeling.

[0028] A dual-path feature extraction mechanism is constructed for the preprocessed vibration signal. One path uses a multi-scale convolutional neural network to automatically extract deep features from the processed vibration signal. By setting multiple convolutional kernels of different scales, it captures the local change patterns and hierarchical representation information of the vibration signal at different time scales and dimensions. The other path performs explicit feature extraction on the vibration signal to obtain time-domain features, frequency-domain features, and time-frequency-domain features, thereby realizing multi-angle representation of the processed vibration signal.

[0029] In the explicit feature extraction path, the following time-domain features are extracted: mean, standard deviation, variance, root mean square value, peak value, peak-to-peak value, mean absolute value, signal energy, shape factor, peak factor, impulse factor, and sharpness factor, totaling 12 time-domain features; the following frequency-domain features are extracted: power spectral density, dominant frequency, center frequency, frequency bandwidth, total energy, harmonic frequency components, spectral mean, spectral variance, spectral kurtosis, spectral skewness, spectral root mean square value, and spectral peak / mean ratio, totaling 12 frequency-domain features; at the same time, 24 time-frequency domain features are extracted; thus forming an explicit statistical feature set containing time-domain, frequency-domain, and time-frequency domain information.

[0030] The prediction model is trained using a multi-source fusion dataset. The acquisition of this dataset specifically includes: obtaining machine tool machining parameters and vibration signals from a JSON data file generated during the machining of precision angular contact ball bearings; preprocessing the obtained machining parameters and vibration signals; automatically extracting deep and explicit features from the preprocessed vibration signals; fusing these features to form a comprehensive vibration signal feature; and then fusing this comprehensive feature with the preprocessed machine tool machining parameters to obtain the multi-source fusion dataset for predicting machining quality. The prediction model uses roughness, roundness, and shape as the neural network output labels and divides the dataset into training, validation, and test sets according to a preset ratio for subsequent model training, parameter tuning, and performance evaluation.

[0031] The prediction model employs a composite neural network model for predicting the machining quality of precision angular contact ball bearings. This model includes a Transformer network module, an LSTM network module, and a multidimensional linear output module. A self-attention mechanism is integrated into the Transformer module to extract global correlations between multi-source input features and enhance the representation of key features. The LSTM module models the temporal dynamic changes during the machining process, and the multidimensional linear layer BPNN is used to achieve joint regression output of multiple quality indicators. This composite neural network model can simultaneously consider the global dependencies of multi-source information, temporal evolution patterns, and coupling relationships between output dimensions, thereby improving the accuracy of quality prediction and the model's generalization ability.

[0032] The composite neural network model is trained using a training set, and the model structure parameters, training parameters, and fusion method are optimized and adjusted using a validation set to obtain an intelligent prediction model for processing quality. The model's prediction performance is then validated using a test set, and the mean absolute percentage error, mean absolute error, and other evaluation indicators are output to assess the model's prediction effect on the three quality indicators of roughness, roundness, and shape.

[0033] The machining parameters, vibration signals, and related extracted features corresponding to the bearing to be predicted are input into the trained intelligent prediction model for machining quality, which outputs the corresponding prediction results for roughness, roundness, and shape. Based on the prediction results, the machining quality status can be further evaluated, abnormal trend warnings can be issued, and quality risk identification can be achieved. The differentiability of the composite neural network model with respect to the input parameters can be used to analyze the sensitivity of changes in machining parameters to the output results of roughness, roundness, and shape. Based on the gradient backpropagation mechanism, some machining parameters can be optimized in reverse, providing a reference for subsequent process parameter adjustments and quality improvement.

[0034] In a specific embodiment of the present invention, such as Figure 1 As shown, the first step is to acquire a massive amount of JSON data files generated during the machining process of the precision angular contact ball bearing. These JSON data files originate from the machining site data acquisition system and are used to record machine tool operation information, vibration data acquisition information, workpiece identification information, and quality inspection results during the bearing machining process.

[0035] In this implementation, the fields in the JSON data file are parsed to extract the following multi-source information: a) Machine tool processing parameters, including but not limited to spindle speed, feed parameters, processing cycle time, process segment markings, and other parameters reflecting the machine tool processing status; b) Vibration signals collected from different parts of the workpiece during processing, wherein the vibration signals are time-series signal data corresponding to the workpiece processing process; c) Workpiece identification information, including workpiece number, process number, batch information, and timestamp information; d) Quality inspection labels, including three processing quality indicators: roughness, roundness, and shape.

[0036] After field extraction, machine tool processing parameters, vibration signals, and quality inspection labels are uniformly associated based on workpiece number, process number, timestamp, and preset mapping relationships to form sample-level data units for individual workpieces or single processing steps, and an original sample library is established. Through the above steps, the structured organization of dispersed and heterogeneous processing data can be achieved, laying a data foundation for subsequent signal processing, feature extraction, and model training.

[0037] Preprocessing of vibration signals: Because vibration signals in real industrial environments have problems such as strong noise interference, obvious non-stationarity, multiple channel dimensions, baseline drift and a lot of redundant information, it is necessary to preprocess the original vibration signals to improve the effectiveness of subsequent feature extraction and the stability of model training.

[0038] In this embodiment, the vibration signal is processed sequentially using the following steps: a) First, perform signal dimensionality reduction and channel organization to extract the XYZ three-channel vibration signal; b) The extracted three-channel vibration signals are merged with the processing position information to establish a correspondence between the vibration signals and the spatial position status during the processing. c) Perform a first DC component removal process on the merged signal to eliminate DC bias in the signal; d) Perform envelope demodulation on the signal after DC removal to enhance the modulation components that reflect changes in processing status; e) Perform wavelet denoising on the envelope demodulated signal to reduce random noise and high-frequency interference in the industrial environment; f) Use a bandpass filter to retain the effective vibration components within the target frequency band and filter out frequency band information that is weakly correlated with processing quality; g) The processed signal is normalized using the Min-max normalization method to ensure uniform numerical scale; h) After normalization, the DC component is removed again to further reduce the impact of baseline drift on subsequent analysis; i) Finally, PCA principal component analysis is used to extract principal components and compress information from the processed multidimensional signal, highlighting the main vibration characteristics and reducing redundant information.

[0039] After the above preprocessing, standardized vibration signal data suitable for subsequent feature extraction and neural network modeling are obtained. This preprocessing process can reduce the impact of noise, redundancy, and irrelevant disturbances on the prediction results while preserving key features of the processing state.

[0040] Explicit Time-Frequency Feature Extraction of Vibration Signals As attached Figure 4 As shown, after completing the vibration signal preprocessing, explicit statistical feature extraction is performed on the vibration signal to characterize the vibration state during the processing from multiple perspectives, including the time domain, frequency domain, and time-frequency domain.

[0041] In this embodiment, the time-domain features include the following 12 items: mean; standard deviation; variance; root mean square value; peak value; peak-to-peak value; mean absolute value; signal energy; shape factor; peak factor; impulse factor; and sharpness factor.

[0042] The frequency domain characteristics include the following 12 items: power spectral density; dominant frequency; center frequency; frequency bandwidth; total energy; harmonic frequency components; spectral mean; spectral variance; spectral kurtosis; spectral skewness; spectral root mean square value; and spectral peak / mean ratio.

[0043] Simultaneously, 24 time-frequency domain features were extracted to further characterize the local energy distribution, frequency evolution law, and non-stationary change characteristics of the signal in the time-frequency plane.

[0044] This results in an explicit statistical feature set consisting of 12 time-domain features, 12 frequency-domain features, and 24 time-frequency-domain features. This set can comprehensively describe the vibration response characteristics of the machine tool and workpiece system during machining, thus providing highly interpretable input information for machining quality prediction.

[0045] Automatic feature extraction using multi-scale convolutional neural networks In addition to explicit statistical features, this invention also uses a multi-scale convolutional neural network to automatically mine deep features in vibration signals, so as to further characterize the local patterns, hierarchical relationships and multi-scale variation laws of vibration signals under complex processing conditions.

[0046] like Figure 5 As shown, the multi-scale convolutional neural network takes the preprocessed vibration signal as input and sets multiple convolutional kernels of different scales. It extracts features from the vibration signal through parallel or hierarchical convolutional structures. Different convolutional kernels correspond to receptive fields of different time scales, which can capture short-term impact features, periodic change features, and cross-scale local pattern information respectively.

[0047] In this embodiment, the processing of the multi-scale convolutional neural network includes convolution operations, nonlinear activation, feature map generation, and feature convergence. This structure can automatically learn a deep representation from the original processed signal that differs from manually calculated statistical features. Compared to single-scale convolutional networks, multi-scale convolutional structures can more effectively adapt to the variations in vibration signals across different frequency bands and time resolutions, thereby improving the completeness and effectiveness of feature extraction.

[0048] Finally, the multi-scale convolutional neural network outputs a deep automatic feature representation of the vibration signal.

[0049] Multi-source feature fusion and dataset partitioning: After obtaining explicit statistical features and deep automatic features, the two are integrated and further fused with machine tool processing parameters to construct a multi-source input sample for the processing quality prediction model.

[0050] Specifically, a) by Figure 4 The process shown extracts explicit time-domain, frequency-domain, and time-frequency-domain features. b) by Figure 5The deep automatic features extracted by the multi-scale convolutional neural network are shown; c) machine tool processing parameters obtained by parsing JSON data; are uniformly spliced, organized and encoded to form a comprehensive input feature vector for a single sample.

[0051] Meanwhile, roughness, roundness, and shape are used as output labels for the corresponding samples to form supervised learning sample pairs. To complete model training and performance validation, all samples are divided into training, validation, and test sets according to a preset ratio. The training set is used for learning model parameters, the validation set is used for adjusting model structure and training parameters, and the test set is used for the final performance evaluation of the model.

[0052] Predictive models such as Figure 6 As shown, a composite neural network model for predicting the machining quality of precision angular contact ball bearings is adopted. The model takes multi-source fusion features as input and consists of a Transformer module, an LSTM module, and a multi-dimensional linear output layer, wherein the self-attention mechanism is set inside the Transformer module.

[0053] Specifically: a) The Transformer module is used to extract global correlations between multi-source input features. Through the self-attention mechanism, it can model the dependencies between different time steps, different channels, and different feature dimensions, and enhance the expressive power of key features; b) The LSTM module is used to model the temporal dynamic characteristics of the machining process. Since bearing machining vibration signals and machining states have significant temporal evolution characteristics, LSTM can capture long-term and short-term temporal dependencies, enhancing the model's ability to learn dynamic trends. c) Multidimensional linear layer BPNN is used to map the aforementioned deep features to multiple quality output dimensions to achieve joint regression prediction of three indicators: roughness, roundness, and shape.

[0054] With the above structure, the model can simultaneously take into account global dependency modeling, temporal relationship modeling, and multi-output mapping modeling, thereby better expressing the complex nonlinear coupling relationship between processing parameters, vibration signal characteristics, and multiple quality indicators.

[0055] Model training and prediction output: After constructing the composite neural network model, the model's parameters are trained using training set samples, and the hyperparameters, training epochs, and network structure are adjusted and optimized using the validation set. Once the model training is complete, its performance is validated using the test set.

[0056] In this embodiment, the model outputs three continuous quality prediction values, corresponding to roughness, roundness, and shape, respectively. By comparing the model's prediction results with the actual detected labels, the model's error evaluation metrics can be calculated, including Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and other evaluation metrics.

[0057] In this embodiment, the results obtained after model training show that the method has high accuracy in bearing processing quality prediction tasks, with MAPE reaching about 0.07 and MAE reaching about 0.01 to 0.02, indicating that the present invention can accurately achieve joint prediction of multiple quality indicators.

[0058] When new machining data for a workpiece is input, its corresponding machining parameters, vibration signals, and relevant extracted features are input into the trained composite neural network model, which then outputs predicted results for its roughness, roundness, and shape. Based on the prediction results, the workpiece's quality status can be further determined, providing a basis for quality assessment, anomaly warning, and process adjustment.

[0059] Further reverse optimization implementation of processing parameters. In a further embodiment of the present invention, since the constructed composite neural network model is differentiable as a whole, the processing parameters can be subjected to sensitivity analysis and reverse optimization based on the gradient information of the model output regarding the input processing parameters.

[0060] Specifically, under the premise of setting a target quality range, some process conditions can be fixed, and the machining parameters can be used as variables to be optimized. Gradient backpropagation analysis can be used to determine the direction and degree of influence of different machining parameters on the predicted values ​​of roughness, roundness, and shape, thereby obtaining parameter adjustment directions that help improve quality indicators. This method can provide a reference for subsequent machine tool process parameter optimization and quality closed-loop control.

[0061] It should be noted that the above-mentioned reverse optimization implementation of processing parameters is a further extension of the present invention. Without affecting the realization of the core quality prediction scheme of the present invention, it can be selected whether to implement it according to the actual process conditions and application requirements.

[0062] Compared with existing methods that rely on post-processing detection or traditional single machine learning modeling, this invention can uniformly parse and reconstruct samples from heterogeneous JSON processing data, making it highly applicable to engineering projects. A preprocessing workflow including DC component removal, envelope demodulation, wavelet denoising, bandpass filtering, normalization, and PCA improves vibration signal quality and modeling stability. Dual-path collaborative extraction of explicit statistical features and multi-scale convolutional deep features enables multi-angle representation of processing vibration signals. A composite neural network consisting of Transformer, LSTM, and multi-dimensional output layers achieves joint prediction of three quality indicators: roughness, roundness, and shape, improving model accuracy and generalization ability. This invention can be further extended to quality early warning, process optimization, and other precision mechanical parts processing quality prediction tasks, demonstrating significant potential for widespread application.

[0063] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, the computer program including program instructions, and the processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), or field-programmable gate arrays (FPGAs). Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., are the computing core and control core of the terminal. They are suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function. The processor described in this embodiment of the invention can be used for the operation of precision angular contact ball bearing quality prediction, including the following steps: acquiring the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted; performing dual-path feature extraction on the acquired vibration signals, one path using a multi-scale convolutional neural network to automatically extract deep features from the vibration signals to obtain automatic deep features; the other path performing explicit feature extraction on the vibration signals to obtain explicit features; fusing the extracted automatic deep features and explicit features with multi-source features to obtain comprehensive vibration signal features, and fusing the comprehensive vibration signal features with the machine tool processing parameters with multi-source features; using a prediction model to predict the multi-source fused features to obtain the corresponding processing quality parameters, including roughness, roundness, and shape.

[0064] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0065] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the precision angular contact ball bearing quality prediction method in the above embodiments. One or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: acquiring the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted; performing dual-path feature extraction on the acquired vibration signals, one path using a multi-scale convolutional neural network to automatically extract deep features from the vibration signals to obtain automatic deep features; the other path performing explicit feature extraction on the vibration signals to obtain explicit features; fusing the extracted automatic deep features and explicit features with multi-source features to obtain comprehensive vibration signal features; fusing the comprehensive vibration signal features with the machine tool processing parameters with multi-source features; using a prediction model to predict the multi-source fused features to obtain the corresponding processing quality parameters, including roughness, roundness, and shape.

[0066] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the quality of precision angular contact ball bearings, characterized in that, Includes the following steps: Obtain the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted; The acquired vibration signal is subjected to dual-path feature extraction. One path uses a multi-scale convolutional neural network to automatically extract deep features from the vibration signal to obtain automatic deep features; the other path extracts explicit features from the vibration signal to obtain explicit features. The extracted automatic deep features and explicit features are fused from multiple sources to obtain the comprehensive features of the vibration signal. The comprehensive features of the vibration signal are then fused with the machine tool processing parameters from multiple sources. A predictive model is used to predict the features after multi-source fusion to obtain the corresponding processing quality parameters, which include roughness, roundness and shape.

2. The method for predicting the quality of a precision angular contact ball bearing according to claim 1, characterized in that, The machine tool machining parameters and vibration signals during the machining process are obtained from the JSON data file generated during the machining of precision angular contact ball bearings. The obtained machine tool machining parameters and vibration signals during the machining process are then preprocessed to obtain standardized vibration signals.

3. The method for predicting the quality of a precision angular contact ball bearing according to claim 2, characterized in that, The vibration signal is preprocessed, specifically including: Dimensionality reduction of the original vibration signal; Extract the XYZ three-channel vibration signals; The XYZ three-channel vibration signals are combined with the processing position information; The merged signal undergoes a first DC component removal process. Envelope demodulation is performed on the signal after DC removal; Wavelet denoising is performed on the envelope-demodulated signal; Bandpass filtering is applied to the wavelet-denoised signal. Min-max normalization is applied to the bandpass filtered signal; The standardized signal undergoes a second DC component removal process. The normalized vibration signal was obtained by performing PCA principal component analysis on the signal after the second DC removal.

4. The method for predicting the quality of a precision angular contact ball bearing according to claim 1, characterized in that, The explicit features include time-domain features, frequency-domain features, and time-frequency-domain features; Temporal characteristics include: mean, standard deviation, variance, root mean square value, peak value, peak-to-peak value, mean absolute value, signal energy, shape factor, peak factor, impulse factor, and sharpness factor.

5. The method for predicting the quality of a precision angular contact ball bearing according to claim 4, characterized in that, The frequency domain features include: power spectral density, dominant frequency, center frequency, frequency bandwidth, total energy, harmonic frequency components, spectral mean, spectral variance, spectral kurtosis, spectral skewness, spectral root mean square value, and spectral peak / mean ratio.

6. The method for predicting the quality of a precision angular contact ball bearing according to claim 4, characterized in that, The time-frequency domain features are 24 features used to characterize the local energy distribution, frequency evolution, and non-stationary change characteristics of vibration signals.

7. The method for predicting the quality of a precision angular contact ball bearing according to claim 1, characterized in that, The prediction model is trained using a multi-source fusion dataset. The acquisition of this dataset specifically includes: obtaining machine tool machining parameters and vibration signals from a JSON data file generated during the machining process of a precision angular contact ball bearing; preprocessing the obtained machine tool machining parameters and vibration signals; automatically extracting deep and explicit features from the preprocessed vibration signals; fusing these features to form a comprehensive vibration signal feature; and then performing multi-source fusion with the preprocessed machine tool machining parameters to obtain the multi-source fusion dataset for machining quality prediction input. The prediction model uses roughness, roundness, and shape as the neural network output labels.

8. The method for predicting the quality of a precision angular contact ball bearing according to claim 7, characterized in that, Predictive models include: The Transformer module is used to extract global correlations between features from multiple sources. The LSTM module is used to model the temporal dynamic changes during the processing. A multidimensional linear output layer is used to achieve joint regression output of multiple processing quality parameters.

9. The method for predicting the quality of a precision angular contact ball bearing according to claim 1, characterized in that, It also includes generating quality assessment results, anomaly warning information, and quality risk identification results based on the prediction results of the processing quality parameters.

10. A precision angular contact ball bearing quality prediction system, characterized in that, It includes a data acquisition module, a feature extraction module, a fusion module, and a prediction module; The data acquisition module acquires the machine tool processing parameters and vibration signals during the processing of the workpiece to be predicted. The feature extraction module performs dual-path feature extraction on the acquired vibration signal. One path uses a multi-scale convolutional neural network to automatically extract deep features from the vibration signal to obtain automatic deep features; the other path performs explicit feature extraction on the vibration signal to obtain explicit features. The fusion module performs multi-source feature fusion of the extracted automatic deep features and explicit features to obtain comprehensive vibration signal features, and then performs multi-source fusion of the comprehensive vibration signal features with machine tool processing parameters. The prediction module uses a prediction model to predict the corresponding processing quality parameters based on the multi-source fusion features. The processing quality parameters include roughness, roundness, and shape.

Images