Method for diagnosing bearing corrosion based on regional convolutional network

Abstract

The invention provides a method for diagnosing bearing corrosion based on a regional convolutional network, and the method comprises the steps: collecting a stray magnetic flux signal through a coil magnetic flux sensor, and decomposing the collected stray magnetic flux signal into an axial component and a radial component through processing; performing power spectrum extraction on the axial and radial components, and generating a continuous spectrum image; processing the continuous spectrum image, taking the cut and adjusted continuous spectrum image as input, and identifying a potential area for diagnosing the corrosion degree of the bearing by using a selective search algorithm; and inputting the identified potential region into a CNN neural network, extracting shallow layer features for diagnosing the corrosion degree of the bearing by using the CNN neural network, and classifying the corrosion degree of the bearing by using the shallow layer features to obtain a final diagnosis result.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault diagnosis, and particularly relates to a method for diagnosing bearing corrosion based on a region convolutional network. Background Art

[0002] With the progress of science and technology, industrial equipment has become more complex and precise. Since industrial equipment plays an important role in actual production, once a failure occurs, the industrial production process will be severely affected. Moreover, according to statistics, most motor failures are caused by bearing electro-corrosion. Therefore, the research on the corrosion degree of motor bearings is of great significance. Previous research on bearing fault diagnosis has mainly focused on the acquisition and feature extraction of vibration signals or current signals. Currently, many aspects have systematically summarized the research on fault diagnosis of rolling bearings using vibration signals or current signals in recent years. However, on the one hand, adding vibration sensors significantly increases the actual production cost. On the other hand, especially under some special working conditions, some harmonics in the current spectrum will be amplified. For the above reasons, it is necessary to select a signal that is less affected by working conditions, such as a method for detecting bearing electro-corrosion using the stray magnetic flux signal of a motor. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides a method for diagnosing bearing corrosion based on a region convolutional network, which can solve the problems of low diagnostic accuracy caused by the installation limitations, high cost of vibration sensors, and harmonic amplification of current signals under special working conditions in the prior art. Technical Solution: The object of the present invention can be achieved by the following technical solution: A method for diagnosing bearing corrosion based on a region convolutional network, comprising the following steps: Step 1: Collect the stray magnetic flux signal through a coil magnetic flux sensor; Step 2: Extract the power spectrum of the obtained stray magnetic flux signal to obtain a continuous spectrum image, that is, a binary histogram accumulated in the power frequency space; Step 3: Crop and adjust the continuous spectrum image in Step 2, and use it as the input of the region convolutional neural network, and use the region convolutional neural network to diagnose the corrosion degree of the motor bearing. Preferably, in Step 2, the power spectrum extraction of the obtained stray magnetic flux signal, the power spectrum extraction expression is: X(f) = [X1(f) X2(f) … X n (f)] In the formula: X m(f) is the discrete Fourier transform (DTF) of the window data centered at time mR; x(n) is the input signal at time n, g(n) is the window function, and R is the number of samples between subsequent DFTs. Preferably, in step 2, the collected stray magnetic flux signal is processed and decomposed into axial and radial components, the power spectra of the axial and radial components are extracted, and a continuous spectrum image is plotted. For the continuous spectrum image, with frequency as the horizontal axis and power spectrum as the vertical axis, a two-dimensional continuous spectrum image is generated, and the continuous spectrum image is used to characterize the correlation characteristics between the bearing corrosion state and the change of the magnetic flux signal. Preferably, in step 3, the process of diagnosing the corrosion degree of the motor bearing using the region convolutional network includes: Step 3-1: Process the continuous spectrum image, use the cropped and adjusted continuous spectrum image as the input, and use the selective search algorithm to identify the potential regions for diagnosing the bearing corrosion degree. Step 3-2: Input the potential regions identified in step 3-1 into the CNN neural network, use the CNN neural network to extract the shallow features for diagnosing the bearing corrosion degree, and classify the bearing corrosion degree using the shallow features to obtain the diagnosis result. Preferably, the continuous spectrum image in step 3-1: It represents the relationship between frequency and power, and for each frequency band of the power frequency pair, it represents the percentage of time it exists in the signal evolution. The redder the color in the continuous spectrum image, the longer the specific power frequency band persists in the signal. Preferably, in step 3-1, the selective search algorithm includes: Investigating all edge regions attributed to different corrosion degrees and having peak amplitudes, and classifying them as anchor points or bounding boxes. Perform a regression task (using non-maximum suppression) on the input continuous spectrum image, and use the bounding boxes to generate a set of potential regions. Preferably, in step 3-2, using the CNN neural network to extract the shallow features for judging the bearing corrosion degree and classifying the bearing corrosion degree using the shallow features includes inputting the region proposals of size 32×32 into the primary two-dimensional convolutional block to extract fixed-length feature vectors, and these high-level features are input into two fully connected layers, and a softmax layer used as a classifier at the back. Preferably, the softmax layer used as a classifier is the core component of the CNN classification task, and through probabilistic output and cross-entropy loss optimization, efficient multi-classification decision-making is achieved. Preferably, in the softmax layer, the process of probabilistic output includes: Convert the original scores output by the last layer (fully connected layer) of the neural network into a probability distribution, such that the probability value of each category is between 0 and 1, and the sum of the probabilities of all categories is 1: Wherein, For each score z i Perform exponentiation to amplify the influence of high scores, where K is the number of categories; Preferably, in the softmax layer, the cross-entropy loss is optimized: measure the difference between the probability distribution p = [p1, p2,..., p K predicted by the model and the true label distribution y = [y1, y2,..., y K : The loss for a single sample: Wherein, K is the number of categories, p i is the probability distribution predicted by the model, and y i is the true label distribution; The loss for a batch of samples: Wherein, K is the number of categories, N is the number of samples, p i is the probability distribution predicted by the model, and y i is the true label distribution; Beneficial effects: Compared with the prior art, the present invention has at least the following technical effects: 1. The present invention diagnoses the corrosion degree of the motor bearing through the stray magnetic flux signal of the motor. The stray magnetic flux signal is not easily affected by special working conditions, avoiding problems such as the installation limitation of vibration sensors, high cost, and low diagnostic accuracy caused by the harmonic amplification of current signals under special working conditions. 2. Aiming at the problem of identifying the timing characteristics of the stray magnetic flux signal, the applied region convolutional network can first be filtered through the selective search algorithm, and then a simple convolutional neural network is used to classify the degree of electrical corrosion of the bearing, greatly reducing the network scale. Description of the Drawings Figure 1 is the overall framework diagram of the present invention. Figure 2 is the diagnostic flowchart of the corrosion degree of the present invention. Figure 3 is the continuous spectrum image for fault judgment. Detailed Embodiments To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. As Figure 1 shown, it is the overall framework diagram of a method for diagnosing bearing corrosion based on a region convolutional network provided by the present invention. First, a stray magnetic flux signal is collected by a coil magnetic flux sensor, and the collected stray magnetic flux signal is decomposed into axial and radial components through processing; power spectrum extraction is performed on the axial and radial components, and a continuous spectrum image is generated; the continuous spectrum image is processed, and the cropped and adjusted continuous spectrum (PS) image is used as the input, and the selective search algorithm is used to identify the potential regions for judging the bearing corrosion degree; the identified potential regions are input into the CNN neural network, and the CNN neural network is used to extract the shallow features for judging the bearing corrosion degree, and the shallow features are used to classify the bearing corrosion degree to obtain the final diagnosis result. The specific implementation steps are as follows Step 1: Collect the stray magnetic flux signal through a coil magnetic flux sensor; According to the stray magnetic flux signal collected by the coil magnetic flux sensor during the operation of the motor, the coil magnetic flux sensor has a total of 1000 turns, an inner diameter of 65 mm, and a wire diameter of 0.2 mm; The signal initially collected by this method is collected through a coil magnetic flux sensor, and the coil magnetic flux sensor needs to be placed on the frame of the motor under test; Step 2: Perform power spectrum extraction on the obtained stray magnetic flux signal to obtain a continuous spectrum image, that is, a binary histogram accumulated in the power frequency space; Due to the particularity of the collected signal, the collected stray magnetic flux signal needs to be decomposed into axial and radial components through processing, and then power spectrum extraction is performed on the axial and radial components, and a continuous spectrum image is drawn as the input signal of the corrosion degree diagnosis model. Perform power spectrum extraction on the obtained stray magnetic flux signal, and the power spectrum extraction expression is: X(f) = [X1(f) X2(f) … X n (f)] In the formula: X m (f) is the discrete Fourier transform (DTF) of the window data centered on time mR; x(n) is the input signal at time n, g(n) is the window function, and R is the number of samples between subsequent DFTs. Step 3: Crop and adjust the continuous spectrum image in Step 2, and use it as the input of the region convolutional neural network, and use the region convolutional neural network to diagnose the corrosion degree of the motor bearing. Corrosion degree diagnosis process: As Figure 2The specific process of corrosion degree diagnosis is shown as follows: The power spectrum of the obtained stray magnetic flux signal is extracted to obtain a continuous spectrum image. Then, the image is preprocessed, and the processed image is fed into a region convolutional neural network, which first uses the selective search algorithm to identify potential regions. Subsequently, the image data is input into a convolutional neural network model based on the potential regions, and the model further analyzes the image to extract key features. The model calculates the classification probability through the softmax function and continuously updates the parameters during training until convergence. Finally, the model gives a diagnosis based on the classification results, completes the image analysis and recognition tasks, and obtains the final diagnosis result. Specifically: The regional neural network includes the selective search algorithm and the CNN neural network based on potential regions. The steps of the selective search algorithm are described as follows. Step 1. Perform preliminary spectral analysis and segmentation on the candidate corrosion degree features; Step 2. Investigate all edge regions attributed to different corrosion degrees and having peak amplitudes, and classify them as anchor points or bounding boxes; Step 3. Perform a regression task (using non-maximum suppression) on the input continuous spectrum image, and use the bounding boxes to generate a set of potential regions; Step 4. Check the termination condition, that is, judge whether the number of regions detected up to the current iteration reaches the upper limit. If not, calculate an additional potential region and insert it into the hierarchical queue; Step 5. Input the identified potential regions into the CNN neural network. For the CNN neural network, a region proposal of size 32×32 is input into the primary two-dimensional convolutional block to extract a fixed-length feature vector. A batch normalization layer is introduced before the first convolutional layer to ensure dimensional consistency. The three convolutional layers have 128, 64, and 32 filters respectively. These high-level features are input into two fully connected layers, which contain 128 and 64 neurons respectively, stacked together. The entire adaptive feature learning process is completed in the fully connected layers, and there is a softmax layer used as a classifier at the back. The standard ReLU activation function is used throughout the network. The output of ReLU activation is expressed as follows: f(x) = max(0, x) When the input is positive: the output is equal to the input, and the derivative is 1; when the input is negative: the output is 0, and the derivative is 0. The softmax layer used as a classifier is the core component of the CNN classification task, and through probabilistic output and cross-entropy loss optimization, it realizes efficient multi-classification decision-making. For the input vector z = [z1, z2,..., z K (K is the number of categories), the softmax output is: In the softmax layer, the process of probabilistic output includes: converting the raw scores output by the last layer (fully connected layer) of the neural network into a probability distribution, such that the probability value of each category is between 0 and 1, and the sum of the probabilities of all categories is 1; In the formula, For each score z i exponentiation is performed to amplify the influence of high scores, where K is the number of categories; In the softmax layer, cross-entropy loss optimization: measures the difference between the probability distribution p = [p1, p2,..., p K predicted by the model and the true label distribution y = [y1, y2,..., y K : For the loss of a single sample: In the formula, K is the number of categories, p i is the probability distribution predicted by the model, and y i is the true label distribution; For the loss of a batch of samples: In the formula, K is the number of categories, N is the number of samples, p i is the probability distribution predicted by the model, and y i is the true label distribution; To accurately evaluate the performance of the proposed region convolutional neural network model, in the experiment, the region convolutional neural network (R-CNN) and the CNN neural network were respectively used for model training. Figure 3 The continuous spectrum image for fault judgment has the frequency as the horizontal axis and the power spectrum as the vertical axis. Some regions are highlighted in the figure. When the fault deteriorates, the components contained in these regions stay in the signal for a longer time and have a higher amplitude. Moreover, in each case, the continuous spectrum image also shows certain differences. These features enable the region convolutional neural network to distinguish the corrosion degree of the bearing using the continuous spectrum image. In the experiment, the corrosion degree was divided into three levels, namely the healthy state and two different corrosion levels. The first-level corrosion is the damage degree of a brand-new bearing running in the motor for 30 hours, and the second-level corrosion is the damage degree of a brand-new bearing running in the motor for 60 hours. Each corrosion degree has a different topology, which is its main feature. The stray magnetic flux signal in the coil flux sensor was extracted at a sampling frequency of 20 kHz. According to different voltages and load levels (percentage of the rated load), the following different working condition types were designed. Operating condition type: Operating condition type Voltage (V) Load rating Operating condition 1 100 No load Operating condition 2 100 50％ Operating condition 3 100 75％ Operating condition 4 100 100％ Operating condition 5 200 No load Operating condition 6 200 50％ Operating condition 7 200 75％ Operating condition 8 200 100％ Operating condition 9 300 No load Operating condition 10 300 50％ Operating condition 11 300 75％ Operating condition 12 300 100％ Diagnostic accuracy of each model under 100V condition: Diagnostic accuracy of each model under 200V condition: Operating condition R-CNN CNN 5 95.1％ 63.2％ 6 96.4％ 51.5％ 7 96.9％ 58.1％ 8 97.2％ 53.7％ Diagnostic accuracy of each model under 300V condition: Operating condition R-CNN CNN 9 96.3％ 67.4％ 10 97.1％ 64.9％ 11 97.5％ 60.2％ 12 98.6％ 55.8％ Although the specific embodiments of the present invention have been described above, those skilled in the art should understand that these are only examples, and various changes or modifications can be made to these embodiments without departing from the principles and essence of the present invention. The scope of the present invention is only defined by the appended claims.

Claims

1. A method for diagnosing bearing corrosion based on a region convolutional network, characterized in that, It includes the following steps: Step 1: Collect stray magnetic flux signals through a coil magnetic flux sensor; Step 2: Extract the power spectrum of the acquired stray magnetic flux signals to obtain a continuous spectrum image, that is, a binary histogram accumulated in the power frequency space; Step 3: Crop and adjust the continuous spectrum image in Step 2 and use it as the input of a region convolutional neural network. The region convolutional neural network includes a selective search algorithm and a CNN neural network, and use the region convolutional neural network to diagnose the corrosion degree of the motor bearing.

2. The method for diagnosing bearing corrosion based on a regional convolutional network according to claim 1, wherein In Step 2, the collected stray magnetic flux signals are processed and decomposed into axial and radial components, the power spectra of the axial and radial components are extracted, and a continuous spectrum image is plotted; For the continuous spectrum image, with frequency as the horizontal axis and power spectrum as the vertical axis, a two-dimensional continuous spectrum image is generated. The continuous spectrum image is used to characterize the correlation characteristics between the bearing corrosion state and the change of the magnetic flux signal.

3. A method for diagnosing bearing corrosion based on a region convolutional network according to claim 1, characterized in that, In Step 2, the expression for extracting the power spectrum of the acquired stray magnetic flux signals is: X(f) = [X1(f) X2(f) … X n (f)] Where: X m (f) is the discrete Fourier transform (DTF) of the window data centered on the time mR; x(n) is the input signal at time n, g(n) is the window function, and R is the number of samples between successive DFTs.

4. A method for diagnosing bearing corrosion based on a region convolutional network according to claim 1, characterized in that, In Step 3, the process of using the region convolutional network to diagnose the corrosion degree of the motor bearing includes: Step 3-1: Process the continuous spectrum image, use the cropped and adjusted continuous spectrum image as the input, and use the selective search algorithm to identify potential regions for diagnosing the corrosion degree of the bearing; Step 3-2: Input the potential regions identified in Step 3-1 into the CNN neural network, use the CNN neural network to extract shallow features for judging the corrosion degree of the bearing, and classify the corrosion degree of the bearing using the shallow features to obtain a diagnosis result.

5. The method for diagnosing bearing corrosion based on a region convolutional network according to claim 3, wherein The continuous spectrum image in Step 3-1 represents the relationship between frequency and power, and for each frequency band of the power frequency pair, it represents the percentage of time present in the signal evolution. The redder the color in the continuous spectrum image, the longer the specific power frequency band persists in the signal.

6. The method for diagnosing bearing corrosion based on a region convolutional network according to claim 3, wherein, In Step 3-1, the selective search algorithm includes: investigating all edge regions attributed to different corrosion degrees and having peak amplitudes, and classifying them as anchor points or bounding boxes, performing a regression task on the input continuous spectrum image using non-maximum suppression, and generating a set of potential regions using the bounding boxes. For the non-maximum suppression, loU >= 0.5, and its process is:

1. Arrange the scores of all regions from large to small; 2. Eliminate redundancy, all regions with loU >= 0.5 with the region having the maximum score; 3. Retain the region with the maximum score, and the remaining regions are used as a new candidate set.

7. A method for diagnosing bearing corrosion based on a region convolutional network according to claim 3, characterized in that In Step 3-2, using the CNN neural network to extract shallow features for judging the corrosion degree of the bearing and classifying the corrosion degree of the bearing using the shallow features includes inputting a region proposal with a size of 32×32 into a primary two-dimensional convolutional block to extract a fixed-length feature vector. These high-level features are input into two fully connected layers, and a softmax layer used as a classifier later.

8. A method for diagnosing bearing corrosion based on a regional convolutional network according to claim 6, characterized in that, The softmax layer used as a classifier is the core component of the CNN classification task, and through probabilistic output and cross-entropy loss optimization, efficient multi-classification decisions are achieved.

9. A method for diagnosing bearing corrosion based on a regional convolutional network according to claim 7, characterized in that, In the softmax layer, the process of probabilistic output includes: Convert the raw scores output by the last layer of the neural network, i.e., the fully connected layer, into a probability distribution such that the probability value for each category is between 0 and 1, and the sum of the probabilities for all categories is 1; wherein, for each score z i is exponentiated to amplify the influence of high scores, and K is the number of categories.

10. A method for diagnosing bearing corrosion based on a region convolutional network according to claim 7, characterized in that In the softmax layer, the cross-entropy loss is optimized, which measures the difference between the probability distribution p = [p1, p2,..., p K predicted by the model and the true label distribution y = [y1, y2,..., y K : The loss for a single sample: where K is the number of categories, p i is the probability distribution predicted by the model, and y i is the true label distribution; The loss for a batch of samples: where K is the number of classes, N is the number of samples, and p i is the probability distribution predicted by the model, and y i is the true label distribution.

Images