Method and device for predicting residual service life of cylindrical grinding wheel based on acoustic emission signal

Abstract

The invention discloses a method for predicting the remaining service life of a cylindrical grinding wheel based on acoustic emission signals, which comprises the following steps of: acquiring acoustic emission signal data in the working process of the cylindrical grinding wheel, labeling the acoustic emission signal data according to the number of effective processing parts, and forming a data set by the acoustic emission signal data and labels; constructing an initial model which comprises a data preprocessing module, a multi-scale feature extraction module, a feature fusion module and a prediction module; training the initial model by using the data set to obtain a prediction model for predicting the residual service life of the cylindrical grinding wheel; and inputting acoustic emission signal data during working of the cylindrical grinding wheel into the prediction model so as to output the residual service life of the corresponding cylindrical grinding wheel. The invention further provides a device for predicting the remaining service life of the cylindrical grinding wheel. The method provided by the invention can predict the residual service life of the cylindrical grinding wheel so as to guide production and maintenance, reduce the rejection rate and improve the resource utilization rate.

Description

Technical Field

[0001] This invention belongs to the field of mechanical manufacturing condition monitoring and predictive maintenance technology, and particularly relates to a method and device for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals. Background Technology

[0002] Cylindrical grinding, a high-precision machining method in the field of mechanical manufacturing, is widely used in the final forming of precision parts such as shafts and discs. The grinding wheel, as the core cutting tool in the grinding process, directly determines the surface quality, geometric accuracy, and production cost of the workpiece. As grinding progresses, the grinding wheel experiences abrasive wear, abrasive breakage, binder detachment, and chip groove blockage, leading to a gradual degradation of its grinding performance. Failure to replace a failing grinding wheel in time will directly result in machining quality defects such as workpiece burning, chatter marks, and dimensional deviations, and may even damage the machine tool.

[0003] Patent document CN120068469A discloses a grinding wheel thickness management method, system, device, and medium, including the following steps: obtaining the initial thickness of the target grinding wheel and an initial alarm threshold; obtaining the cumulative compensation amount and average compensation amount of the target grinding wheel after each dressing compensation based on the initial thickness; inputting the initial alarm threshold and average compensation amount into a preset threshold dynamic adjustment model to obtain a dynamic alarm threshold; determining whether the cumulative compensation amount of the target grinding wheel in the current dressing compensation cycle is greater than or equal to the dynamic alarm threshold; if so, sending an alarm message; otherwise, returning to the step of obtaining the cumulative compensation amount and average compensation amount of the target grinding wheel after each dressing compensation based on the initial thickness.

[0004] Patent document CN120178791A discloses a method for predicting the life of a diamond grinding wheel when grinding helical grooves with a solid carbide tool. The method includes the following steps: A machine learning module is used to estimate the wear amount along the grinding wheel diameter along the axial direction. The machine learning input includes tool parameters, grinding wheel parameters, grinding parameters, etc. The machine learning output is the wear amount along the axial diameter of the grinding wheel. The machine learning input is substituted into the machine learning module to predict the wear state of the grinding wheel. If the wear exceeds the set limit, the grinding wheel is re-dressed and feedback is sent to the machine tool. Otherwise, the wear state of the grinding wheel, grinding wheel information, grinding path, and helical groove information are substituted into a module for calculating the rake angle and core thickness errors to obtain the rake angle and core thickness errors and determine whether they meet the tolerance requirements. If the rake angle and core thickness errors meet the requirements, the helical groove is ground. The machined helical groove shape is repeatedly acquired using an image method and fused into a precise contour. Then, the shape of the grinding wheel is calculated using the envelope principle, and the machine learning-predicted grinding wheel wear information is updated. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for predicting the remaining service life of external cylindrical grinding wheels based on acoustic emission signals. This method can achieve highly real-time and highly accurate prediction of the remaining service life of external cylindrical grinding wheels, so as to guide production and maintenance, reduce scrap rate, and improve resource utilization.

[0006] To achieve the first objective of this invention, the following technical solution is provided: a method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals, comprising the following steps: Acoustic emission signal data are collected during the working process of the external cylindrical grinding wheel, and the acoustic emission signal data is labeled according to the number of effectively processed parts. The acoustic emission signal data and labels are combined into a dataset. The initial model is constructed, which includes a data preprocessing module, a multi-scale feature extraction module, a feature fusion module, and a prediction module; The data preprocessing module is used to perform noise reduction and normalization processing on the input acoustic emission signal data. The multi-scale feature extraction module is used to perform multi-resolution analysis on the acoustic emission signal data processed by the data preprocessing module, so as to output feature components at different frequencies. The feature fusion module is used to fuse feature components at different frequencies to output a corresponding comprehensive feature vector. The prediction module performs a patch segmentation operation on the input comprehensive feature vector to construct multiple local segments, and mixes the multiple local segments in the time dimension channel and the spatial dimension channel respectively to output the time features and spatial features of the corresponding channel. Based on the time features and spatial features, a prediction is made to output the prediction result. The initial model was trained using the dataset to obtain a predictive model for predicting the remaining service life of the external cylindrical grinding wheel; The acoustic emission signal data of the external cylindrical grinding wheel during operation is input into the prediction model to output the remaining service life of the corresponding external cylindrical grinding wheel.

[0007] This invention suppresses strong background noise in the acoustic emission signal during grinding, extracts signal features that are inherently related to the wear of the grinding wheel, and captures its multi-scale time dependence. This enables highly real-time and accurate prediction of the remaining service life of external cylindrical grinding wheels, guiding production and maintenance, reducing scrap rates, and improving resource utilization.

[0008] Specifically, the number of effective processed parts is determined by whether the surface roughness of the current processed part is greater than or equal to a threshold. When the surface roughness is greater than or equal to the threshold, the number of processed parts is recorded. When the surface roughness is less than the threshold, the processing of the next processed part continues.

[0009] Specifically, the noise reduction process is based on a weighted reconstruction noise reduction algorithm using mutual information and variational mode decomposition to process the original signal.

[0010] Specifically, the normalization process maps the denoised acoustic emission signal data to a preset numerical range.

[0011] Specifically, the mapping process employs either the max-min normalization method or the Z-score normalization method.

[0012] Specifically, the feature fusion module uses discrete wavelet transform to fuse feature components at different frequencies.

[0013] Specifically, the discrete wavelet transform method adopts the Daubechies wavelet system.

[0014] To achieve the second objective of this invention, the following technical solution is provided: a device for predicting the remaining service life of an external cylindrical grinding wheel, used to perform the steps of the above-described method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: By employing a weighted reconstruction denoising strategy that combines VMD and mutual information, the core signal components related to grinding wheel wear can be adaptively and accurately separated, effectively suppressing environmental noise and interference, and providing a high-quality data foundation for subsequent accurate prediction. The proposed multi-resolution branch Mixer model innovatively combines the multi-resolution analysis capability of discrete wavelet transform with the powerful feature mixing (extraction) capability of the MLP-Mixer architecture. Wavelet transform decouples the signal across different frequency bands, enabling the model to analyze wear characteristics in parallel from multiple scales and frequency bands; the Mixer block efficiently captures long-range dependencies within the sequence and interactions between features, avoiding some inherent defects of traditional CNN / RNN models, thereby achieving deeper and more comprehensive automatic feature extraction. This enables real-time and precise health management of the grinding wheel's condition, guiding production personnel to replace the grinding wheel at the optimal time. This avoids batch scrapping of workpieces and quality accidents caused by excessive wear of the grinding wheel, while also preventing premature replacement of the grinding wheel, saving production costs, and improving equipment utilization and production efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method for predicting the remaining service life of an external cylindrical grinding wheel provided in this embodiment; Figure 2 This is a schematic diagram of the time-domain decomposition of the original acoustic emission signal provided in this embodiment; Figure 3This is a schematic diagram of the frequency domain decomposition of the original acoustic emission signal provided in this embodiment; Figure 4 This is the spectrum of the signal before reconstruction provided in this embodiment; Figure 5 This is the spectrum of the reconstructed signal provided in this embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown in this embodiment, a method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals is provided. The specific steps are as follows: Acoustic emission signal data are collected during the working process of the external cylindrical grinding wheel, and the acoustic emission signal data is labeled according to the number of effectively processed parts. The acoustic emission signal data and labels are combined into a dataset. The initial model is constructed, which includes a data preprocessing module, a multi-scale feature extraction module, a feature fusion module, and a prediction module; The data preprocessing module is used to perform noise reduction and normalization processing on the input acoustic emission signal data. The multi-scale feature extraction module is used to perform multi-resolution analysis on the acoustic emission signal data processed by the data preprocessing module, so as to output feature components at different frequencies. The feature fusion module is used to fuse feature components at different frequencies to output a corresponding comprehensive feature vector. The prediction module performs a patch segmentation operation on the input comprehensive feature vector to construct multiple local segments, and mixes the multiple local segments in the time dimension channel and the spatial dimension channel respectively to output the time features and spatial features of the corresponding channel. Based on the time features and spatial features, a prediction is made to output the prediction result. The initial model was trained using the dataset to obtain a predictive model for predicting the remaining service life of the external cylindrical grinding wheel; The acoustic emission signal data of the external cylindrical grinding wheel during operation is input into the prediction model to output the remaining service life of the corresponding external cylindrical grinding wheel.

[0019] Furthermore, during the construction of the dataset, based on the external cylindrical grinding wheel grinding experiment dataset, a corresponding remaining service life label is created for each data sample in the dataset. The label value is the remaining number of grinding operations for the current sample until the end of the grinding wheel's life.

[0020] The data processing module collects the raw acoustic emission signals generated during the grinding process of the external cylindrical grinding wheel; the raw acoustic emission signals are preprocessed, and the preprocessing process includes: For noise reduction, a weighted reconstruction noise reduction algorithm based on mutual information and variational mode decomposition is used to process the original signal.

[0021] First, the signal is decomposed into a series of intrinsic mode functions (EMFs), and then the mutual information values ​​between each EMF and the original signal are calculated. This is used to quantify the correlation between the two. The formula for calculating mutual information is: ; Where X represents the original signal, Y represents an IMF component, P(x,y) is the joint probability distribution of X and Y, and P(x) and P(y) are the marginal probability distributions.

[0022] The calculated mutual information values ​​of each IMF component Normalization is performed to obtain its weight coefficients. : ; Weighted reconstruction is performed based on the calculated mutual information weights. While preserving the effective information related to grinding wheel wear in the acoustic emission signal to the greatest extent, noise and redundant information are removed, so that the signal characteristics can be highlighted, which is convenient for subsequent prediction.

[0023] Normalization operation: The denoised acoustic emission signal data is mapped to a preset numerical range [-1,1] to eliminate the influence of dimensions and accelerate the subsequent convergence process of the model.

[0024] The normalization operation employs either max-min normalization or Z-score normalization.

[0025] The expression for the maximum-minimum normalization method is: ; Where X represents the original data. and These are the minimum and maximum values ​​in the original data, respectively. The data is after normalization; The Z-score normalization expression is:

[0026] Where μ is the mean of the original data and σ is the standard deviation of the original data. This is the standardized data. The standardized data usually conforms to a distribution with a mean of 0 and a standard deviation of 1, and its numerical range naturally maps to an interval centered at 0, which meets the interval requirement of [-1, 1].

[0027] The multi-scale feature extraction module performs multi-resolution analysis on the preprocessed one-dimensional acoustic emission signal sequence, decomposing it into components of different frequency sub-bands, approximation coefficients, and detail coefficients.

[0028] A specific sub-band component obtained by wavelet transform is processed through multiple parallel branches, where each branch contains at least two cascaded MLP blocks for extracting deep features from the sub-band component. The MLP block is composed of a fully connected layer (linear layer), a GELU activation function, and a Dropout regularization layer connected sequentially.

[0029] The feature fusion module is used to fuse the multi-scale features extracted from each parallel branch through wavelet inverse transform to form a comprehensive feature vector that fully represents the state of the grinding wheel.

[0030] In the prediction module, a patch segmentation operation is performed on the comprehensive feature vector to divide the sequence into multiple local segments (patches) to increase the model's receptive field and reduce computational cost. Subsequently, the processed data is fed into the Mixer layer in two orthogonal dimensions for further processing: first, mixing is performed in the time dimension (patch_size dimension) to capture long-distance dependencies in the sequence; then, mixing is performed in the spatial dimension (num_patch dimension) to achieve information interaction between different feature channels; finally, after flattening, the high-level features after two-dimensional mixing are mapped to the final predicted value of the remaining service life of the external cylindrical grinding wheel through a global pooling layer and a fully connected layer.

[0031] This embodiment also provides a device for predicting the remaining service life of an external cylindrical grinding wheel, used to perform the steps of the method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals provided in the above embodiment.

[0032] To better illustrate the beneficial effects of the technical solution of the present invention, the grinding wheel of the external cylindrical grinding machine used for machining shaft parts is used as the monitoring object, and the implementation process of the present invention is described in detail.

[0033] Data collection and label creation Experimental setup: Through-feed grinding experiments were conducted on a high-precision cylindrical grinding machine. The workpiece material was 45# steel, and the grinding wheel was a white corundum wheel. An acoustic emission sensor with a resonant frequency of 150 kHz was installed on the housing near the grinding position to collect acoustic emission signals during the grinding process. The sensor signal was amplified by a preamplifier with a gain of 40 dB, and then acquired and stored by a data acquisition card with a sampling rate of 1.25 MHz.

[0034] Failure threshold setting: The surface roughness Ra value of the workpiece was used as the criterion for grinding wheel failure. When the Ra value of the workpiece surface produced by continuous grinding exceeded 1.6 μm, the grinding wheel was deemed to have failed, and the experiment was stopped. A total of 276 workpieces were ground from the start of the new grinding wheel until failure.

[0035] Tagging: The continuous acoustic emission signal data collected throughout the grinding process is sliced ​​into two sets of data samples, each with 276 grinding cycles, using the grinding cycle of a single workpiece as the unit. A Remaining Useful Life (RUL) tag is created for each sample. For example, the RUL tag for the 36th workpiece sample is 240 (276-36), and the RUL tag for the 200th workpiece sample is 76.

[0036] Dataset creation: The collected data from each grinding cycle were cleaned: outliers were removed; signals from the non-grinding state were removed; and the stages immediately following and immediately after grinding were removed. Simultaneously, each data point was mapped to a label indicating remaining service life and saved as a CSV file.

[0037] Signal preprocessing VMD Decomposition: Taking the original acoustic emission signal sequence s(t) within a workpiece grinding cycle, according to Table 1, the VMD decomposition mode number K=3 is set. This fully extracts the effective information from the acoustic emission signal without causing redundancy. The penalty factor α=2000, using the default convergence tolerance. The VMD algorithm decomposes s(t) into three intrinsic mode functions (IMF) components, namely IMF1 ~ IMF3, as follows... Figure 2 and Figure 3 As shown, where Figure 2 For the time domain dimension, Figure 3 For the frequency domain dimension.

[0038] In addition, as shown in Table 1, the center frequencies of different IMFs for a signal at different K values ​​are shown.

[0039] .

[0040] Mutual information calculation and filtering: Calculate the mutual information value MI between each IMF component and the original signal s(t).

[0041] Signal reconstruction: the mutual information values ​​of each calculated IMF component. Normalization is performed to obtain its weight coefficients. :

[0042] The reconstructed signal is obtained by weighting and reconstructing the signal based on the calculated mutual information weights. : .

[0043] Comparison of the spectrum diagrams of the signal before and after reconstruction, for example Figure 4 and Figure 5 As shown, where Figure 4 This is the spectrum of the original data signal. Figure 5 This is the spectrum of the reconstructed data signal.

[0044] Normalization operation: The denoised signal s_denoised(t) is processed using the max-min normalization method, linearly mapping it to the interval [-1, 1]. Taking the max-min normalization method as an example, the normalized data can be obtained as follows: ; Where X represents the original data. and These are the minimum and maximum values ​​in the original data, respectively. This is the normalized data.

[0045] Predictive Model Building and Training The model constructed in this embodiment is based on the multi-resolution branch Mixer architecture of discrete wavelet transform, and its specific construction steps are as follows: Discrete Wavelet Transform Layer: The input signal is subjected to a 3-level Discrete Wavelet Transform (DWT) using the Db4 wavelet basis. After decomposition, four sub-band components are obtained: one three-level approximation coefficient (cA3, low frequency) and three detail coefficients (cD1, cD2, cD3, high frequency to low frequency).

[0046] Multi-resolution branching module: Four parallel branches are set up for corresponding dimensions. The structure consists of two or more layers of sequential connections (Linear + GELU + Dropout), and each MLP block undergoes residual connections. It processes the four sub-band component sequences cA3, cD3, cD2, and cD1 respectively.

[0047] Feature fusion module: The multi-scale features extracted from each parallel branch are fused through wavelet inverse transform to form a comprehensive feature vector that fully represents the state of the grinding wheel.

[0048] Spatiotemporal dual-dimensional hybrid module: This module receives the comprehensive feature sequence output by the feature fusion module, first performs a patch segmentation operation on it, dividing the sequence into patches with a window size of patch_len and a stride; then the processed data is sent to the Mixer layer for processing on the two newly generated dimensions. The Mixer structure is: LayerNorm + Linear + GELU + Dropout + Linear + Dropout.

[0049] Finally, after flattening, a fully connected layer is used to map the high-level features, which have undergone two-dimensional hybrid processing, into the final predicted value of the remaining service life of the external cylindrical grinding wheel.

[0050] Model training: Two sets of 248 samples each (90%) were used as the training set, and 28 samples each (10%) were used as the test set. The Adam optimizer was used with an initial learning rate of 0.001 and the mean squared error (MSE) loss function. The training epochs were 200, and an early stopping strategy was employed to prevent overfitting.

[0051] Prediction and Results Analysis The 56 samples from the test set are input into the trained model for prediction. The predicted RUL values ​​are then compared with the true RUL labels.

[0052] To demonstrate the superiority of this invention, this embodiment compares the method of this invention with conventional methods on the same test set. The evaluation metrics used are mean mean square error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Performance comparisons of different prediction methods are shown in Table 2.

[0053] The results show that the prediction method provided by this invention has significantly lower evaluation indicators than the comparative methods, demonstrating higher prediction accuracy and reliability, and can be effectively applied to the prediction of the remaining service life of external cylindrical grinding wheels.

[0054] Furthermore, the terms "upper," "lower," "inner," "outer," "front," and "rear" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0055] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of the present invention. All equivalent changes or modifications made to the structure, features and principles described in the claims of the present invention should be included in the scope of the claims of the present invention.

[0056] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals, characterized in that, Includes the following steps: Acoustic emission signal data are collected during the working process of the external cylindrical grinding wheel, and the acoustic emission signal data is labeled according to the number of effectively processed parts. The acoustic emission signal data and labels are combined into a dataset. The initial model is constructed, which includes a data preprocessing module, a multi-scale feature extraction module, a feature fusion module, and a prediction module; The data preprocessing module is used to perform noise reduction and normalization processing on the input acoustic emission signal data. The multi-scale feature extraction module is used to perform multi-resolution analysis on the acoustic emission signal data processed by the data preprocessing module, so as to output feature components at different frequencies. The feature fusion module is used to fuse feature components at different frequencies to output a corresponding comprehensive feature vector. The prediction module performs a patch segmentation operation on the input comprehensive feature vector to construct multiple local segments, and mixes the multiple local segments in the time dimension channel and the spatial dimension channel respectively to output the time features and spatial features of the corresponding channel. Based on the time features and spatial features, a prediction is made to output the prediction result. The initial model was trained using the dataset to obtain a predictive model for predicting the remaining service life of the external cylindrical grinding wheel; The acoustic emission signal data of the external cylindrical grinding wheel during operation is input into the prediction model to output the remaining service life of the corresponding external cylindrical grinding wheel.

2. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 1, characterized in that, The number of effective processed parts is determined by whether the surface roughness of the current processed part is greater than or equal to a threshold. When the surface roughness is greater than or equal to the threshold, the number of processed parts is recorded. When the surface roughness is less than the threshold, the processing of the next processed part continues.

3. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 1, characterized in that, The noise reduction process is based on a weighted reconstruction noise reduction algorithm using mutual information and variational mode decomposition to process the original signal.

4. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 1, characterized in that, The normalization process maps the denoised acoustic emission signal data to a preset numerical range.

5. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 4, characterized in that, The mapping process employs either the max-min normalization method or the Z-score normalization method.

6. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 1, characterized in that, The feature fusion module uses discrete wavelet transform to fuse feature components at different frequencies.

7. The method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals according to claim 6, characterized in that, The discrete wavelet transform method adopts the Daubechies wavelet system.

8. A device for predicting the remaining service life of an external cylindrical grinding wheel, characterized in that, The steps are for performing the method for predicting the remaining service life of an external cylindrical grinding wheel based on acoustic emission signals as described in any one of claims 1 to 7.

Images