Cutter wear prediction method based on multi-source information weighted fusion convolutional neural network

Abstract

The invention provides a tool wear prediction method based on a multi-source information weighted fusion convolutional neural network, and the method comprises the following steps: obtaining four RGB three-channel wavelet time-frequency images in a data preprocessing module, extracting features from each time-frequency image through a regression convolutional neural network, and carrying out the regression to obtain a tool wear amount, and finally, in a weighted average module, carrying out weighted average on the obtained four tool wear losses, so as to obtain the average wear loss of the tool. According to the method, the wavelet time-frequency image is utilized, the one-dimensional signal is converted into the two-dimensional image, the two-dimensional convolutional neural network can be directly utilized for feature extraction and recognition, compared with a traditional machine learning method, the process of manual feature extraction is not needed, feature extraction and abrasion loss prediction are carried out on multi-source data through the convolutional neural network, and the accuracy of feature extraction and abrasion loss prediction is improved. And then weighted averaging is carried out on the predicted abrasion loss, so that the prediction precision is improved.

Description

Technical Field

[0001] This invention belongs to the technical field of tool wear prediction methods, specifically relating to a tool wear prediction method based on a multi-source information weighted fusion convolutional neural network. Background Technology

[0002] The condition of machine tool cutting tools affects machining quality. To ensure machining quality, a conservative tool-changing strategy is usually adopted during machining, i.e., replacing tools early. Studies show that tools that are replaced generally still have 20%-50% of their service life remaining. Monitoring tool wear during production and automatically determining whether tool replacement is necessary can fully utilize the tool's service life. Machining production with tool prediction systems can improve machining efficiency by 10%-50% and reduce costs by 10%-40%. Therefore, predicting tool wear is of great significance. Due to severe instrument measurement obstruction, direct detection of tool wear is ineffective, hence indirect detection methods have been widely used. Indirect detection mainly utilizes sensors to collect force signals, vibration signals, and acoustic emission signals during machining. Through preprocessing and sensitive feature extraction, machine learning models are used to monitor tool wear. This method is limited by the quality of feature extraction; traditional machine learning methods yield unstable results, different models show significant differences in results, and different feature extraction methods have a significant impact on the final result.

[0003] Existing methods for detecting tool wear mainly rely on machine learning. To extract features closely related to tool wear, researchers not only need to be good at observation and discovery, but also require certain skills and experience. Moreover, the extracted features have weak interpretability and universality, are prone to missing sensitive features, and are easily affected by subjective human factors, resulting in recognition performance that cannot meet practical needs.

[0004] Based on this, a tool wear prediction method based on a multi-source information weighted fusion convolutional neural network is proposed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a tool wear prediction method based on multi-source information weighted fusion convolutional neural network to address the shortcomings of the prior art, thereby solving the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a tool wear prediction method based on multi-source information weighted fusion convolutional neural network, comprising the following steps: The data preprocessing module contains four types of RGB three-channel wavelet time-frequency images. For each type of time-frequency image, a regression convolutional neural network is used to extract features and regress to obtain a tool wear amount. Finally, in the weighted averaging module, the four tool wear amounts are weighted and averaged to obtain the average tool wear amount.

[0007] As a further explanation of the present invention, the four types of RGB three-channel wavelet time-frequency images are vibration time-frequency image, force time-frequency image, acoustic emission time-frequency image, and current time-frequency image.

[0008] As a further explanation of the present invention, a three-channel accelerometer is installed on the machine tool to detect the vibration signals of the tool in the x, y, and z directions; a three-channel force sensor is installed to detect the force signals of the tool in the x, y, and z directions; an acoustic emission sensor is installed to detect the acoustic emission signals of the tool; and a current sensor is installed to detect the spindle current. The data preprocessing module performs wavelet time-frequency transformation on the collected vibration signals in the x, y, and z directions to obtain wavelet time-frequency images, and performs grayscale processing. The grayscale images in the x, y, and z directions are fused to form an RGB three-channel wavelet time-frequency image. Similarly, the force signals and vibration signals in the x, y, and z directions are processed to obtain RGB three-channel wavelet time-frequency images. For the one-dimensional acoustic emission RMS signal, wavelet analysis is directly performed to obtain an RGB acoustic emission wavelet time-frequency image. The one-dimensional spindle current signal is also directly analyzed using wavelet analysis to obtain an RGB current wavelet time-frequency image.

[0009] As a further explanation of the present invention, the regressive convolutional neural network consists of 8 layers. The first layer consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The second layer also consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The third and fourth layers both consist of a convolutional layer and a ReLU activation function. The fifth layer also consists of a convolutional layer, a ReLU activation function, and a pooling layer. The sixth and seventh layers both consist of a fully connected layer, a ReLU activation function, and a Dropout layer. The eighth layer consists of a fully connected layer. The last regression layer outputs a predicted value.

[0010] As a further explanation of the present invention, the wavelet time-frequency image is scaled to 227×227×3 and input into the regression convolutional neural network. After passing through each layer, the scale becomes: 55×55×96, 27×27×256, 13×13×184, 13×13×184, 13×13×184, 1×4096, 1×4096, 1×1. The final regression layer outputs a predicted value of tool wear.

[0011] Compared with the prior art, the present invention has the following advantages: 1. This invention utilizes wavelet time-frequency images to convert one-dimensional signals into two-dimensional images, which can be directly used for feature extraction and recognition using two-dimensional convolutional neural networks; 2. Compared with traditional machine learning methods, this invention eliminates the need for manual feature extraction.

[0012] 3. This invention uses convolutional neural networks to extract features and predict wear amount from multiple sources of data, and then performs a weighted average of the predicted wear amount, thereby improving the accuracy of the prediction. Attached Figure Description

[0013] Figure 1 This is a schematic block diagram of the prediction method of the present invention; Figure 2 This is the data preprocessing process of the data preprocessing module of the present invention; Figure 3 This is a diagram of the regression convolutional neural network structure of this invention. Detailed Implementation

[0014] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] like Figure 1-3 As shown, the present invention provides a technical solution: a tool wear prediction method based on a multi-source information weighted fusion convolutional neural network, comprising the following steps: The data preprocessing module contains four types of RGB three-channel wavelet time-frequency images, namely vibration time-frequency image, force time-frequency image, acoustic emission time-frequency image and current time-frequency image. Specific examples Figure 2 As shown, a three-channel accelerometer is installed on the machine tool to detect the vibration signals of the tool in the x, y, and z directions. A three-channel force sensor is installed to detect the force signals of the tool in the x, y, and z directions. An acoustic emission sensor is installed to detect the acoustic emission signals of the tool. A current sensor is installed to detect the spindle current. The data preprocessing module performs wavelet time-frequency transformation on the collected vibration signals in the x, y, and z directions to obtain wavelet time-frequency images, and performs grayscale processing. The grayscale images in the x, y, and z directions are fused to form an RGB three-channel wavelet time-frequency image. Similarly, the force signals and vibration signals in the x, y, and z directions are processed to obtain RGB three-channel wavelet time-frequency images. The one-dimensional acoustic emission RMS signal is directly subjected to wavelet analysis to obtain an RGB acoustic emission wavelet time-frequency image. The one-dimensional spindle current signal is also directly subjected to wavelet analysis to obtain an RGB current wavelet time-frequency image.

[0016] Then, for each time-frequency image, a regression convolutional neural network extracts features and regresses to obtain a tool wear amount; Specifically, such as Figure 3As shown, the regressive convolutional neural network consists of 8 layers. The first layer consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The second layer also consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The third and fourth layers both consist of a convolutional layer and a ReLU activation function. The fifth layer also consists of a convolutional layer, a ReLU activation function, and a pooling layer. The sixth and seventh layers both consist of a fully connected layer, a ReLU activation function, and a Dropout layer. The eighth layer consists of a fully connected layer. The last regression layer outputs a predicted value.

[0017] The wavelet time-frequency image is scaled to 227×227×3 and input into a regression convolutional neural network. After passing through each layer, the scales become: 55×55×96, 27×27×256, 13×13×184, 13×13×184, 13×13×184, 1×4096, 1×4096, and 1×1, respectively. The final regression layer outputs a predicted value of tool wear.

[0018] Finally, in the weighted average module, the four tool wear values ​​are weighted and averaged to obtain the average tool wear value.

[0019] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0020] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A tool wear prediction method based on a multi-source information weighted fusion convolutional neural network, characterized in that: Includes the following steps: The data preprocessing module contains four types of RGB three-channel wavelet time-frequency images. For each type of time-frequency image, a regression convolutional neural network is used to extract features and regress to obtain a tool wear amount. Finally, in the weighted averaging module, the four tool wear amounts are weighted and averaged to obtain the average tool wear amount.

2. The tool wear prediction method based on a multi-source information weighted fusion convolutional neural network according to claim 1, characterized in that, The four types of RGB three-channel wavelet time-frequency images are vibration time-frequency image, force time-frequency image, acoustic emission time-frequency image, and current time-frequency image.

3. The tool wear prediction method based on a multi-source information weighted fusion convolutional neural network according to claim 2, characterized in that, A three-channel accelerometer is installed on the machine tool to detect the vibration signals of the tool in the x, y, and z directions. A three-channel force sensor is installed to detect the force signals of the tool in the x, y, and z directions. An acoustic emission sensor is installed to detect the acoustic emission signals of the tool. A current sensor is installed to detect the spindle current. The data preprocessing module performs wavelet time-frequency transformation on the collected vibration signals in the x, y, and z directions to obtain wavelet time-frequency images, and performs grayscale processing. The grayscale images in the x, y, and z directions are fused to form an RGB three-channel wavelet time-frequency image. Similarly, the force signals and vibration signals in the x, y, and z directions are processed to obtain RGB three-channel wavelet time-frequency images. The one-dimensional acoustic emission RMS signal is directly subjected to wavelet analysis to obtain an RGB acoustic emission wavelet time-frequency image. The one-dimensional spindle current signal is also directly subjected to wavelet analysis to obtain an RGB current wavelet time-frequency image.

4. The tool wear prediction method based on a multi-source information weighted fusion convolutional neural network according to claim 3, characterized in that, The regressive convolutional neural network consists of 8 layers. The first layer consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The second layer also consists of a convolutional layer, a ReLU activation function, a normalization layer, and a pooling layer. The third and fourth layers both consist of a convolutional layer and a ReLU activation function. The fifth layer also consists of a convolutional layer, a ReLU activation function, and a pooling layer. The sixth and seventh layers both consist of a fully connected layer, a ReLU activation function, and a Dropout layer. The eighth layer consists of a fully connected layer. The last regression layer outputs a predicted value.

5. The tool wear prediction method based on a multi-source information weighted fusion convolutional neural network according to claim 4, characterized in that, The wavelet time-frequency image is scaled to 227×227×3 and input into a regression convolutional neural network. After passing through each layer, the scales become: 55×55×96, 27×27×256, 13×13×184, 13×13×184, 13×13×184, 1×4096, 1×4096, and 1×1, respectively. The final regression layer outputs a predicted value of tool wear.

Images