A MEMS gyroscope fault diagnosis method based on improved VMD algorithm and multi-scale CNN

Abstract

This invention relates to the field of gyroscope technology, specifically a MEMS gyroscope fault diagnosis method based on an improved VMD algorithm and a multi-scale CNN. The method comprises the following steps: Step 1: Acquiring MEMS gyroscope signals; Step 2: Cropping the MEMS gyroscope signals; Step 3: Decomposing the MEMS gyroscope signals into sequential signals; Step 4: Dividing the sequential signals into useful signals and noise signals; Step 5: Reconstructing the useful signals into feature signals; Step 6: Reconstructing the feature signals into a 64×64 matrix and converting the matrix into image samples; Step 7: Dividing the image samples into training and testing sets; Step 8: Constructing a multi-scale CNN model; Step 9: Training and testing the multi-scale CNN model; Step 10: Outputting the fault diagnosis results. This invention solves the problems of slow diagnosis speed and low accuracy in existing MEMS gyroscope fault diagnosis methods and is applicable to the fault diagnosis of MEMS gyroscopes.

Description

Technical Field

[0001] This invention relates to the field of gyroscope technology, specifically a MEMS gyroscope fault diagnosis method based on an improved VMD algorithm and multi-scale CNN. Background Technology

[0002] MEMS gyroscopes are widely used in high-precision navigation and control systems due to their advantages such as small size, low power consumption, and high precision. However, in practical applications, environmental factors and long-term operation can cause various malfunctions in MEMS gyroscopes, leading to the inability of high-precision navigation and control systems to operate normally. Therefore, there is a strong need to develop and improve MEMS gyroscope fault diagnosis methods. However, practice shows that existing MEMS gyroscope fault diagnosis methods, which rely on a single algorithm for diagnosis, suffer from slow diagnosis speed and low accuracy. Therefore, it is necessary to invent a MEMS gyroscope fault diagnosis method based on an improved VMD algorithm and multi-scale CNN to address the problems of slow diagnosis speed and low accuracy in existing methods. Summary of the Invention

[0003] To address the problems of slow diagnosis speed and low accuracy in existing MEMS gyroscope fault diagnosis methods, this invention provides a MEMS gyroscope fault diagnosis method based on an improved VMD algorithm and multi-scale CNN.

[0004] This invention is achieved using the following technical solution: A fault diagnosis method for MEMS gyroscopes based on an improved VMD algorithm and multi-scale CNN is proposed, which is implemented through the following steps: Step 1: Acquire MEMS gyroscope signals using sensors; the MEMS gyroscope signals include normal signals and five types of fault signals; the five types of fault signals are: offset fault signal, scaling factor fault signal, device fault signal, transient fault signal, and periodic interference fault signal; Step 2: Use sliding sampling to prune the MEMS gyroscope signal; Step 3: Optimize the number of decomposition modes and penalty factor of the VMD algorithm using the chicken flock optimization algorithm, and decompose the MEMS gyroscope signal into a sequence signal using the VMD algorithm; Step 4: Calculate the sample entropy of the sequence signal, and divide the sequence signal into useful signal and noise signal based on the sample entropy; Step 5: Filter out noise signals and reconstruct the useful signals into characteristic signals; Step 6: Reconstruct the feature signal into a 64×64 matrix, normalize the matrix, and then use Python's matplotlib library to convert the matrix into image samples; the image samples include normal images and five types of fault images; the five types of fault images are: offset fault image, scaling factor fault image, device fault image, transient fault image, and periodic interference fault image. Step 7: Divide the image samples into training and test sets in a 4:1 ratio; Step 8: Construct dilated convolutional blocks and use the dilated convolutional blocks and SE modules to build a multi-scale CNN model; Step 9: Train and test the multi-scale CNN model using the training and test sets; Step 10: Use sensors to acquire MEMS gyroscope signals in real time, and execute steps 2 to 6. Then, input the obtained image samples into the multi-scale CNN model to output the fault diagnosis results.

[0005] Furthermore, in step two, the specific parameters of the sliding sampling operation are: the length of the intercepted signal segment is 4096, and the sliding distance is 200.

[0006] Furthermore, in step three, the specific parameters of the chicken flock optimization algorithm are as follows: the number of chickens is 100, the proportions of roosters, hens, and chicks are 20%, 60%, and 20%, respectively, the influence factor of the mother-child relationship is 0.4~0.9, the identity is updated every 5 iterations, and the total number of iterations is 300.

[0007] Furthermore, in step four, the division steps are as follows: first, the sample entropy is differentiated, then the index number + 1 of the last extreme point in the derivative result is taken as the dividing point, then the components after the dividing point are determined as useful signals, and the components before the dividing point are determined as noise signals.

[0008] Furthermore, in step eight, the specific steps for constructing the dilated convolutional block are as follows: First, the input feature map is fed into a convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU. Then, a dilated convolutional layer is added after the convolutional layer; the kernel size of the dilated convolutional layer is 3×3, the dilation rate is K, and the activation function is ReLU; Then, a convolutional layer is added after the dilated convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU; Finally, the feature map is output.

[0009] Furthermore, in step eight, the construction steps of the multi-scale CNN model are as follows: First, the input feature map is subjected to a convolution operation with a kernel size of 5×5 and a channel number of 128; Then, the feature map is input into three branches with dilation rates of 1, 2, and 3, respectively; Then, three dilated convolutional blocks are added to each branch, and an SE module is added after each dilated convolutional block; the number of channels of the three dilated convolutional blocks in each branch are 256, 512, and 1024 respectively. Then, a convolution operation with a kernel size of 1×1 is added to each branch, thereby reducing the number of channels to 256; Then, add a Pooling layer to each branch; Then, the feature maps obtained from the three branches are combined by overlay to form a cross-scale feature fusion layer; Then, a fully connected layer is added after the feature fusion layer; Finally, a Softmax classifier is added after the fully connected layer.

[0010] Furthermore, in step eight, the specific parameters of the multi-scale CNN model are: batch_size=64, epoch=90, init_learning_rate=0.001.

[0011] Compared with existing MEMS gyroscope fault diagnosis methods, the MEMS gyroscope fault diagnosis method based on an improved VMD algorithm and multi-scale CNN described in this invention no longer uses a single algorithm recognition principle for diagnosis. Instead, it utilizes a flock optimization algorithm, dilated convolutional blocks, and SE modules to improve the VMD algorithm and multi-scale CNN model. By fusing the improved VMD algorithm and multi-scale CNN model, fault diagnosis of MEMS gyroscopes is achieved, thereby significantly accelerating the diagnosis speed and greatly improving the diagnosis accuracy. To verify the above beneficial effects, a MEMS gyroscope fault diagnosis experiment was conducted using this invention. The experimental results are as follows: Figure 12 As shown in the figure, the experimental results show that the average precision rate reached 99.28%, the average recall rate reached 99.25%, and the average accuracy rate reached 99.25%.

[0012] This invention effectively solves the problems of slow diagnosis speed and low accuracy of existing MEMS gyroscope fault diagnosis methods, and is applicable to the fault diagnosis of MEMS gyroscopes. Attached Figure Description

[0013] Figure 1 This is a flowchart of the present invention.

[0014] Figure 2 This is a schematic diagram of the sequence signal in step three of the present invention.

[0015] Figure 3 This is a schematic diagram of the feature signals in step five of this invention.

[0016] Figure 4 This is a schematic diagram of step six of the present invention.

[0017] Figure 5 This is a schematic diagram of the offset fault image in step six of the present invention.

[0018] Figure 6 This is a schematic diagram of the scale factor fault image in step six of the present invention.

[0019] Figure 7 This is a schematic diagram of the device fault image in step six of the present invention.

[0020] Figure 8 This is a schematic diagram of the transient fault image in step six of the present invention.

[0021] Figure 9 This is a schematic diagram of the periodic interference fault image in step six of the present invention.

[0022] Figure 10 This is a schematic diagram of the hollow convolution block in step eight of this invention.

[0023] Figure 11 This is a schematic diagram of the multi-scale CNN model in step eight of this invention.

[0024] Figure 12 This is a schematic diagram of the experimental results of the present invention. Detailed Implementation

[0025] A fault diagnosis method for MEMS gyroscopes based on an improved VMD algorithm and multi-scale CNN is proposed, which is implemented through the following steps: Step 1: Acquire MEMS gyroscope signals using sensors; the MEMS gyroscope signals include normal signals and five types of fault signals; the five types of fault signals are: offset fault signal, scaling factor fault signal, device fault signal, transient fault signal, and periodic interference fault signal; Step 2: Use sliding sampling to prune the MEMS gyroscope signal; Step 3: Optimize the number of decomposition modes and penalty factor of the VMD algorithm using the chicken flock optimization algorithm, and decompose the MEMS gyroscope signal into a sequence signal using the VMD algorithm; Step 4: Calculate the sample entropy of the sequence signal, and divide the sequence signal into useful signal and noise signal based on the sample entropy; Step 5: Filter out noise signals and reconstruct the useful signals into characteristic signals; Step 6: Reconstruct the feature signal into a 64×64 matrix, normalize the matrix, and then use Python's matplotlib library to convert the matrix into image samples; the image samples include normal images and five types of fault images; the five types of fault images are: offset fault image, scaling factor fault image, device fault image, transient fault image, and periodic interference fault image. Step 7: Divide the image samples into training and test sets in a 4:1 ratio; Step 8: Construct dilated convolutional blocks and use the dilated convolutional blocks and SE modules to build a multi-scale CNN model; Step 9: Train and test the multi-scale CNN model using the training and test sets; Step 10: Use sensors to acquire MEMS gyroscope signals in real time, and execute steps 2 to 6. Then, input the obtained image samples into the multi-scale CNN model to output the fault diagnosis results.

[0026] In step two, the specific parameters for the sliding sampling operation are: the length of the intercepted signal segment is 4096, and the sliding distance is 200.

[0027] In step three, the specific parameters of the chicken flock optimization algorithm are as follows: the number of chickens is 100, the proportions of roosters, hens, and chicks are 20%, 60%, and 20%, respectively, the influence factor of the mother-child relationship is 0.4~0.9, the identity is updated every 5 iterations, and the total number of iterations is 300.

[0028] In step four, the division steps are as follows: First, the sample entropy is differentiated, and then the index number + 1 of the last extreme point in the derivative result is taken as the dividing point. Then, the components after the dividing point are determined as useful signals, and the components before the dividing point are determined as noise signals.

[0029] In step eight, the specific steps for constructing the dilated convolutional block are as follows: First, the input feature map is fed into a convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU. Then, a dilated convolutional layer is added after the convolutional layer; the kernel size of the dilated convolutional layer is 3×3, the dilation rate is K, and the activation function is ReLU; Then, a convolutional layer is added after the dilated convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU; Finally, the feature map is output.

[0030] In step eight, the construction steps of the multi-scale CNN model are as follows: First, the input feature map is subjected to a convolution operation with a kernel size of 5×5 and a channel number of 128; Then, the feature map is input into three branches with dilation rates of 1, 2, and 3, respectively; Then, three dilated convolutional blocks are added to each branch, and an SE module is added after each dilated convolutional block; the number of channels of the three dilated convolutional blocks in each branch are 256, 512, and 1024 respectively. Then, a convolution operation with a kernel size of 1×1 is added to each branch, thereby reducing the number of channels to 256; Then, add a Pooling layer to each branch; Then, the feature maps obtained from the three branches are combined by overlay to form a cross-scale feature fusion layer; Then, a fully connected layer is added after the feature fusion layer; Finally, a Softmax classifier is added after the fully connected layer.

[0031] In step eight, the specific parameters of the multi-scale CNN model are: batch_size=64, epoch=90, init_learning_rate=0.001.

[0032] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of MEMS gyroscopes based on an improved VMD algorithm and multi-scale CNN, characterized in that: This method is implemented using the following steps: Step 1: Acquire MEMS gyroscope signals using sensors; the MEMS gyroscope signals include normal signals and five types of fault signals; the five types of fault signals are: offset fault signal, scaling factor fault signal, device fault signal, transient fault signal, and periodic interference fault signal; Step 2: Use sliding sampling to prune the MEMS gyroscope signal; Step 3: Optimize the number of decomposition modes and penalty factor of the VMD algorithm using the chicken flock optimization algorithm, and decompose the MEMS gyroscope signal into a sequence signal using the VMD algorithm; Step 4: Calculate the sample entropy of the sequence signal, and divide the sequence signal into useful signal and noise signal based on the sample entropy; Step 5: Filter out noise signals and reconstruct the useful signals into characteristic signals; Step 6: Reconstruct the feature signal into a 64×64 matrix, normalize the matrix, and then use Python's matplotlib library to convert the matrix into image samples; the image samples include normal images and five types of fault images; the five types of fault images are: offset fault image, scaling factor fault image, device fault image, transient fault image, and periodic interference fault image. Step 7: Divide the image samples into training and test sets in a 4:1 ratio; Step 8: Construct dilated convolutional blocks and use the dilated convolutional blocks and SE modules to build a multi-scale CNN model; Step 9: Train and test the multi-scale CNN model using the training and test sets; Step 10: Use sensors to acquire MEMS gyroscope signals in real time, and execute steps 2 to 6. Then, input the obtained image samples into the multi-scale CNN model to output the fault diagnosis results.

2. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN as described in claim 1, characterized in that: In step two, the specific parameters for the sliding sampling operation are: the length of the intercepted signal segment is 4096, and the sliding distance is 200.

3. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN according to claim 1, characterized in that: In step three, the specific parameters of the chicken flock optimization algorithm are as follows: the number of chickens is 100, the proportions of roosters, hens, and chicks are 20%, 60%, and 20%, respectively, the influence factor of the mother-child relationship is 0.4~0.9, the identity is updated every 5 iterations, and the total number of iterations is 300.

4. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN according to claim 1, characterized in that: In step four, the division steps are as follows: First, the sample entropy is differentiated, and then the index number + 1 of the last extreme point in the derivative result is taken as the dividing point. Then, the components after the dividing point are determined as useful signals, and the components before the dividing point are determined as noise signals.

5. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN according to claim 1, characterized in that: In step eight, the specific steps for constructing the dilated convolutional block are as follows: First, the input feature map is fed into a convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU. Then, a dilated convolutional layer is added after the convolutional layer; the kernel size of the dilated convolutional layer is 3×3, the dilation rate is K, and the activation function is ReLU; Then, a convolutional layer is added after the dilated convolutional layer; the kernel size of this convolutional layer is 1×1 and the activation function is ReLU; Finally, the feature map is output.

6. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN according to claim 1, characterized in that: In step eight, the construction steps of the multi-scale CNN model are as follows: First, the input feature map is subjected to a convolution operation with a kernel size of 5×5 and a channel number of 128; Then, the feature map is input into three branches with dilation rates of 1, 2, and 3, respectively; Then, three dilated convolutional blocks are added to each branch, and an SE module is added after each dilated convolutional block; the number of channels of the three dilated convolutional blocks in each branch are 256, 512, and 1024 respectively. Then, a convolution operation with a kernel size of 1×1 is added to each branch, thereby reducing the number of channels to 256; Then, add a Pooling layer to each branch; Then, the feature maps obtained from the three branches are combined by superposition to form a cross-scale feature fusion layer; Then, a fully connected layer is added after the feature fusion layer; Finally, a Softmax classifier is added after the fully connected layer.

7. The MEMS gyroscope fault diagnosis method based on the improved VMD algorithm and multi-scale CNN according to claim 1, characterized in that: In step eight, the specific parameters of the multi-scale CNN model are: batch_size=64, epoch=90, init_learning_rate=0.001.

Images