A tool wear prediction method based on fir-stacking

Abstract

The present application relates to a kind of tool wear prediction methods based on FIR-Stacking, comprising the following steps: signal data in tool processing process is collected;Signal data is denoised using optimized band-stop filter;The signal data after denoising is expanded to obtain expanded signal data;The time domain, frequency domain and time-frequency domain of expanded signal data are extracted statistics respectively;Statistical quantity is used as feature, and the single-dimensional data in expanded signal data is extracted feature, and the feature of multidimensional data is spliced, to obtain feature dataset;Feature dataset is input into the integrated learning model based on Stacking, to obtain tool wear result.The present application removes signal noise using adaptive FIR filter, and combines the Stacking integrated method of multiple machine learning models, which can effectively improve the accuracy of tool wear prediction.Through the feature extraction and selection of tool monitoring signal, the degradation characteristics of tool wear are better captured, so that the prediction result is more accurate and reliable.

Description

Technical Field

[0001] This patent application relates to a tool wear prediction method based on FIR-Stacking, which belongs to the field of industrial manufacturing. Background Technology

[0002] Tool wear has a significant impact on production efficiency and product quality in industrial manufacturing. However, in actual use, cutting tools often suffer from wear, breakage, and tool breakage, leading to decreased cutting quality, extended machining time, reduced production efficiency, and increased costs. Therefore, accurately predicting tool wear is crucial for improving production efficiency, reducing costs, and extending the service life of cutting tools.

[0003] Currently, data-driven tool wear prediction methods have become a research hotspot. This approach involves collecting and analyzing cutting tool monitoring signal data, and then constructing a prediction model based on signal feature engineering to predict tool wear. However, existing methods suffer from the following problems:

[0004] 1. Sensor signal noise: During the cutting process, signal data is often affected by the machining environment, resulting in a large amount of noise in the signal. This noise will affect the accuracy and reliability of the prediction model.

[0005] 2. The prediction effect of a single traditional machine learning model is not ideal: Current methods usually use a single traditional machine learning model to predict tool wear, but these models often fail to fully explore the features of the data, resulting in limited prediction results. Summary of the Invention

[0006] To address the aforementioned shortcomings and difficulties, this invention provides a tool wear prediction method based on FIR-Stacking.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a tool wear prediction method based on FIR-Stacking, comprising the following steps:

[0008] 1) Acquire signal data during the tool machining process;

[0009] 2) Construct a band-stop filter and optimize its parameters using a Bayesian optimization algorithm. Then, use the optimized band-stop filter to denoise the signal data.

[0010] 3) The denoised signal data is augmented to obtain augmented signal data;

[0011] 4) Extract statistics from the time domain, frequency domain, and time-frequency domain of the augmented signal data;

[0012] 5) Use statistics as features to extract features from single-dimensional data in the augmented signal data, and concatenate the features of multi-dimensional data to obtain a feature dataset;

[0013] 6) Construct a Stacking-based ensemble learning model, input the feature dataset into the Stacking-based ensemble learning model, and obtain the tool wear results.

[0014] The signal data includes multidimensional data: cutting force signal, vibration signal, and acoustic emission signal.

[0015] The construction of the band-stop filter specifically involves using an FIR filter based on the Kaiser window function to construct the band-stop filter.

[0016] The optimization of the band-stop filter parameters using the Bayesian optimization algorithm includes the following steps:

[0017] By using the order, cutoff frequency, bandwidth, and beta value of the Kaiser window function as parameters, a Bayesian optimization algorithm is employed to search the parameter space by setting the parameter space of the above parameters and defining maximizing the signal-to-noise ratio as the objective function, thereby obtaining the optimal filter parameters.

[0018] The signal data is augmented to obtain augmented signal data, specifically by dividing the denoised signal data into N parts of signal data of length L.

[0019] The statistics include mean, skewness, kurtosis, standard deviation, energy, root mean square, peak-to-peak value, waveform factor, impulse factor, and margin factor.

[0020] Among them, mean, standard deviation, and energy are basic statistics used to describe the central trend, fluctuation degree, and signal strength of signal data; skewness and kurtosis are used to describe the asymmetry and peak state of signal data; root mean square and peak-to-peak value are used to reflect the amplitude and range of variation of signal data; waveform factor, impulse factor, and margin factor are used to characterize the shape characteristics and variation patterns of signal data.

[0021] The construction of the stacking-based ensemble learning model includes the following steps:

[0022] The signal data from the historical tool processing process are processed through steps 1) to 5) to obtain the feature dataset; the Pearson correlation coefficient is used to select features from the training set in the feature dataset, and features with correlation coefficients greater than the threshold are selected as input to the ensemble learning model.

[0023] The selected training set is divided into three parts and input into LightGBM, SVR and MLP respectively for training and prediction to obtain the tool wear prediction value. These prediction values ​​are added to the training set as new features and finally fed into the Lasso model for regression prediction to obtain the final ensemble learning model composed of LightGBM, SVR and MLP and Lasso model.

[0024] A tool wear prediction system based on FIR-Stacking includes:

[0025] The data acquisition module is used to collect signal data during the tool machining process;

[0026] The data denoising module is used to construct a band-stop filter and optimize the parameters of the band-stop filter using a Bayesian optimization algorithm. The optimized band-stop filter is then used to denoise the signal data.

[0027] The data augmentation module is used to augment the denoised signal data to obtain augmented signal data; and to extract statistics from the time domain, frequency domain, and time-frequency domain of the augmented signal data respectively.

[0028] The feature extraction module is used to extract features from single-dimensional data in the augmented signal data by using statistics as features, and to concatenate the features of multi-dimensional data to obtain a feature dataset.

[0029] The tool wear prediction module is used to build a stacking-based ensemble learning model. The feature dataset is input into the stacking-based ensemble learning model to obtain the tool wear results.

[0030] A tool wear prediction device based on FIR-Stacking includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the tool wear prediction method based on FIR-Stacking when the computer program is executed.

[0031] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned tool wear prediction method based on FIR-Stacking.

[0032] The present invention has the following beneficial effects and advantages:

[0033] 1. High-accuracy prediction: By employing an adaptive FIR filter to remove signal noise and combining it with a stacking ensemble method of multiple machine learning models, the accuracy of tool wear prediction can be effectively improved. Through feature extraction and selection of tool monitoring signals, the degradation characteristics of tool wear are better captured, making the prediction results more accurate and reliable.

[0034] 2. Strong noise resistance: The adaptive FIR filter can automatically adjust the filtering parameters according to the actual signal conditions, effectively reducing the impact of noise in the processing environment on the prediction results. By denoising the signal, the model's ability to interpret the signal is improved, thus enhancing the stability and reliability of the prediction.

[0035] 3. Comprehensive Feature Extraction: Utilizing common statistical measures in the time domain, frequency domain, and time-frequency domain as signal features, it can comprehensively and multidimensionally describe the characteristics of tool wear. By combining different features and selecting effective features, it can better capture the changing patterns of tool wear and improve prediction accuracy.

[0036] 4. Multi-model ensemble: Employing a stacking ensemble method using multiple machine learning models can combine the advantages of each model, improving the robustness and generalization ability of predictions. By integrating the prediction results of the base model with the meta-model, the complementarity between models can be better utilized, improving the accuracy and reliability of prediction results.

[0037] 5. High Practicality: The method of this invention is based on existing machine learning algorithms and signal processing technology, and has high practicality and operability. It provides a simple, efficient, and reliable solution for tool wear monitoring and prediction in the industrial manufacturing field, helping enterprises reduce unnecessary production downtime and maintenance costs, and improve production efficiency and product quality. Attached Figure Description

[0038] Figure 1 This is a flowchart of the method modules of the present invention;

[0039] Figure 2 This is a diagram of the Stacking integration model structure. Detailed Implementation

[0040] The method of the present invention will be further described in detail below with reference to the accompanying drawings.

[0041] A tool wear prediction method based on FIR-Stacking includes the following steps:

[0042] 1. Use an adaptive FIR filter to remove signal noise;

[0043] 2. Commonly used statistics are employed in the time domain, frequency domain, and time-frequency domain as signal features, and the features of multi-source signals are concatenated.

[0044] 3. Use Pearson correlation coefficient to screen important features;

[0045] 4. Using LightGBM, SVR, and MLP as base models and Lasso as a meta-model, a Stacking ensemble model is constructed for final tool wear prediction.

[0046] like Figure 1 As shown, the method of the present invention includes the following steps:

[0047] 1. To predict tool wear, a total of seven channels of signals were collected during the machining process: cutting force signal (3 channels, along the X, Y, and Z directions), vibration signal (3 channels, along the X, Y, and Z directions), and acoustic emission signal (1 channel). Tool wear was measured offline after each machining cycle, and this wear amount was used as a label for training the final model. This invention uses seven-channel signal data to predict tool wear. During machining, the feed and unfeed operations cause fluctuations in the signal data; these fluctuations are considered outliers and need to be removed. Through observation and analysis, it was decided to remove outliers from the first and last 10% of the signal data. The retained signal data has stability and continuity, making it more suitable for subsequent experimental research. Furthermore, considering the different signal lengths generated by different machining cycles, a signal with a length of 100,000 was selected for subsequent experiments. These processing measures ensure the accuracy and reliability of the experiments.

[0048] 2. For the signal data after outlier removal, a band-stop filter based on a Kaiser window function FIR filter is designed to remove noise from the signal data. To remove noise bandwidth as much as possible while preserving the original information, this study uses a Bayesian optimization algorithm to optimize the filter parameters. The Bayesian optimization algorithm uses the filter order, cutoff frequency, bandwidth, and the beta value of the Kaiser window function as parameters. By setting the parameter space of these parameters and defining maximizing the signal-to-noise ratio as the objective function, it searches the parameter space to obtain a set of optimal filter parameters. The filter designed in this way can more effectively remove noise and improve signal quality.

[0049] 3. Data augmentation techniques are employed to expand the dataset, preventing problems such as model overfitting due to limited sample size. Specifically, the denoised signal data is divided into 10 parts, that is, a signal data of length 100,000 is split into 10 signal data of length 10,000. This increases the sample size of a single tool from 315 to 3150, thereby increasing the diversity and quantity of data. This strategy further improves the accuracy and robustness of the model.

[0050] 4. Use Fourier transform to analyze the frequency domain information of the signal, and use short-time Fourier transform to obtain the time-frequency domain information of the signal. Extract ten commonly used statistical quantities from the time domain, frequency domain, and time-frequency domain of the signal, including mean, skewness, kurtosis, standard deviation, energy, root mean square, peak-to-peak value, waveform factor, impulse factor, and margin factor.

[0051] These statistics reflect the characteristics of signal data in different domains and play an important role in the analysis and prediction of tool wear. Mean, standard deviation, and energy are basic statistics used to describe the central tendency, fluctuation degree, and signal strength of the signal; skewness and kurtosis describe the asymmetry and peak state of the signal data; root mean square and peak-to-peak value reflect the amplitude and range of variation of the signal; waveform factor, impulse factor, and margin factor reveal the shape characteristics and variation patterns of the signal data.

[0052] By extracting features from the single-dimensional signal, a total of 30 features were obtained in the time domain, frequency domain, and time-frequency domain. Then, the 7-dimensional signal features were concatenated to achieve feature fusion, resulting in a total of 210 features.

[0053] 5. Randomly divide the dataset obtained in the previous step into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%. Use the Pearson correlation coefficient to select features from the training set, filtering out features with a correlation coefficient greater than 0.2 as input to the final model. Since the Pearson correlation coefficient measures the degree of linear correlation between two variables, this step calculates the Pearson correlation coefficient between each feature and the target variable (wear amount), identifying features with a correlation coefficient greater than 0.2 as important features that need to be retained. This approach reduces the number of data features as much as possible, thereby reducing computational burden and improving model performance.

[0054] 6. A stacking-based ensemble learning model is used. The base models are three strong and differentiated regression models for tool wear prediction: LightGBM, SVR, and MLP. This allows the ensemble model to learn the advantages of these three models. The meta-model is a simple regression model: Lasso regression, thus avoiding overfitting. Figure 2 As shown, the training set obtained in the previous step is divided into three parts, which are then used by LightGBM, SVR, and MLP for training and prediction, respectively, to obtain predicted tool wear values. These predicted values ​​are added to the training set as new features, and finally, the Lasso model is used for regression prediction. During model training, a set of optimal model parameters is obtained by minimizing the model's prediction error. The final model is obtained by setting these parameters. The final model is then used to predict the test set to verify its prediction accuracy.

[0055] The feature inputs obtained from steps 1-5 are used to predict tool wear results using the final model. Specifically, the 7-channel signal data generated by each round of tool processing are used to obtain features from steps 1-5, which are then input into the model to obtain the tool wear amount.

[0056] The tool wear prediction method of the present invention can not only improve the performance and efficiency of the model, but also adaptively remove signal noise generated in different industrial environments. At the same time, since the model integrates three different models: LightGBM, SVR and MLP, the final model generalization ability will be better, thereby further improving the accuracy of tool wear prediction.

[0057] By comprehensively applying the above processing techniques, the tool wear prediction method of the present invention can effectively remove the interference of signal noise generated in the industrial environment on tool wear prediction. Furthermore, the use of an integrated model improves the generalization of the model and enhances the tool wear prediction effect, thus demonstrating better performance and application prospects in industrial manufacturing tasks such as tool wear prediction and tool life prediction.

Claims

1. A tool wear prediction method based on FIR-Stacking, characterized in that, Includes the following steps: 1) Acquire signal data during the tool machining process; 2) Construct a band-stop filter and optimize its parameters using a Bayesian optimization algorithm. Then, use the optimized band-stop filter to denoise the signal data. 3) The denoised signal data is augmented to obtain augmented signal data; 4) Extract statistics from the time domain, frequency domain, and time-frequency domain of the augmented signal data; 5) Use statistics as features to extract features from the single-dimensional data in the augmented signal data, and concatenate the features of the multi-dimensional data to obtain the feature dataset; 6) Construct an ensemble learning model based on Stacking, input the feature dataset into the ensemble learning model based on Stacking, and obtain the tool wear results; The construction of the band-stop filter specifically involves using an FIR filter based on the Kaiser window function to construct a band-stop filter. The optimization of the band-stop filter parameters using the Bayesian optimization algorithm includes the following steps: By using the order, cutoff frequency, bandwidth, and beta value of the Kaiser window function as parameters, the Bayesian optimization algorithm searches the parameter space by setting the parameter space of the above parameters and defining maximizing the signal-to-noise ratio as the objective function, thereby obtaining the optimal filter parameters. The construction of the stacking-based ensemble learning model includes the following steps: The signal data from the historical tool processing process are processed through steps 1) to 5) to obtain the feature dataset; the Pearson correlation coefficient is used to select features from the training set in the feature dataset, and features with correlation coefficients greater than the threshold are selected as input to the ensemble learning model. The selected training set is divided into three parts and input into LightGBM, SVR and MLP respectively for training and prediction to obtain the tool wear prediction value. These prediction values ​​are added to the training set as new features and finally fed into the Lasso model for regression prediction to obtain the final ensemble learning model composed of LightGBM, SVR and MLP and Lasso model.

2. The tool wear prediction method based on FIR-Stacking according to claim 1, characterized in that, The signal data includes multidimensional data: cutting force signal, vibration signal, and acoustic emission signal.

3. The tool wear prediction method based on FIR-Stacking according to claim 1, characterized in that, The step of expanding the denoised signal data to obtain expanded signal data specifically involves dividing the denoised signal data into N parts of signal data with a length of L.

4. The tool wear prediction method based on FIR-Stacking according to claim 1, characterized in that, The statistics include mean, skewness, kurtosis, standard deviation, energy, root mean square, peak-to-peak value, waveform factor, impulse factor, and margin factor. Among them, mean, standard deviation, and energy are basic statistics used to describe the central trend, fluctuation degree, and signal strength of signal data; skewness and kurtosis are used to describe the asymmetry and peak state of signal data; root mean square and peak-to-peak value are used to reflect the amplitude and range of variation of signal data; waveform factor, impulse factor, and margin factor are used to characterize the shape characteristics and variation patterns of signal data.

5. A tool wear prediction system based on FIR-Stacking, the system being used to implement the tool wear prediction method based on FIR-Stacking as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to collect signal data during the tool machining process; The data denoising module is used to construct a band-stop filter and optimize the parameters of the band-stop filter using a Bayesian optimization algorithm. The optimized band-stop filter is then used to denoise the signal data. The data augmentation module is used to augment the denoised signal data to obtain augmented signal data; and to extract statistics from the time domain, frequency domain, and time-frequency domain of the augmented signal data respectively. The feature extraction module is used to extract features from single-dimensional data in the augmented signal data by using statistics as features, and to concatenate the features of multi-dimensional data to obtain a feature dataset. The tool wear prediction module is used to build a stacking-based ensemble learning model. The feature dataset is input into the stacking-based ensemble learning model to obtain the tool wear results.

6. A tool wear prediction device based on FIR-Stacking, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a tool wear prediction method based on FIR-Stacking as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements a tool wear prediction method based on FIR-Stacking as described in any one of claims 1-4.

Images