Rolling Bearing Fault Diagnosis Method Based on an Adaptive Parameter Gaussian Convolution Kernel Neural Network

Abstract

The present invention relates to a rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network, belonging to the field of rolling bearing fault diagnosis categories, and includes the following steps: obtaining the vibration signal of the rolling bearing in the rotating equipment by using an accelerometer, slicing, obtaining the envelope vector, and making it into a data set; calculating the kurtosis value and the average skewness value of the samples in the data set, and designing three Gaussian convolution kernels with adaptive parameters related to the characteristics of the bearing vibration impact information; using the obtained three Gaussian convolution kernels with adaptive parameters to design an adaptive parameter Gaussian convolution layer for feature extraction, and splicing the obtained three feature vectors to obtain a new feature vector; inputting the obtained new feature vector into a Softmax classifier with a cross-entropy loss function, and establishing a mapping relationship between the fault features and the fault categories to achieve fault diagnosis, improving the training efficiency and classification effect of the fault diagnosis model, and enhancing the reliability and interpretability of the model.

Description

Technical Field

[0001] The present invention relates to the field of rolling bearing fault diagnosis categories, and particularly to a rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network. Background Art

[0002] Rolling bearings are important components in rotating mechanical equipment, playing a key role in transmitting kinetic energy and connecting rotating components. However, due to changes in the operating state of mechanical equipment and the operating environment, various fault types are likely to occur during the operation of rolling bearings, and in severe cases, it may even lead to the shutdown of mechanical equipment. According to relevant statistics, 90% of the faults in rotating mechanical equipment using rolling bearings come from rolling bearings. Therefore, timely and accurate fault diagnosis of rolling bearings has important practical significance.

[0003] In recent years, with the continuous improvement of sensor technology and the continuous development of computational science and computing equipment, deep learning technology has made rapid progress. Considerable research results have been achieved in the fault diagnosis of rolling bearings based on deep learning methods. However, the current research only relies on the black-box principle of deep learning methods, relying on random fitting to extract fault features and establish the association between fault features and fault types. It cannot involve the physical mechanism knowledge of bearings and the signal statistical mechanism knowledge, and cannot reliably and interpretably extract fault features, often ignoring some important mechanism information hidden in the signals. Therefore, it is crucial to design a method that combines signal processing and deep learning, utilizes the statistical mechanism information contained in the signals, modifies and designs some network components in deep learning, enables the network to have mechanism significance, and can reliably and interpretably extract important features according to the signal mechanism to achieve reliable and intelligent analysis of fault signals. Summary of the Invention

[0004] In view of the above disadvantages, the present invention provides a rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network, which combines signal mechanism knowledge, enhances the reliability and interpretability of fault diagnosis technology, improves the rolling bearing fault diagnosis ability, and is a fault diagnosis method with engineering practical value.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network includes the following steps:

[0007] Step 1: Collect vibration signals in the form of one-dimensional time series of a rotating bearing in different health states through an accelerometer, segment and slice the collected vibration signals, calculate the envelope vector of each segment of the slice and label it as a sample, and make all the samples into a data set;

[0008] Step 2: Calculate the kurtosis value for the envelope vector and the average skewness value by segmented sliding window for setting the Gaussian convolution kernel parameters, obtaining three Gaussian convolution kernels with adaptive parameters. The Gaussian convolution layer with adaptive parameters constructed based on the three Gaussian convolution kernels with adaptive parameters is used to extract three fault feature vectors of the envelope vector and splice them to obtain a new fault feature vector;

[0009] Step 3: Input the new fault feature vector extracted in Step 2 into a Softmax classifier with a cross-entropy loss function, and based on the established mapping relationship between the fault features and the fault categories, diagnose the fault categories of the rolling bearing based on the mapping relationship.

[0010] A further improvement of the technical solution of the present invention lies in that Step 1 includes the following steps:

[0011] Step 11: Divide the vibration signal in the form of a one-dimensional time series into m non-overlapping segments of length N according to the rotational speed and the sampling frequency, constituting m one-dimensional vibration vectors. The calculation expression for the length N is as follows:

[0012]

[0013] where f is the sampling frequency of the vibration signal and r is the rotational speed of the rotating bearing;

[0014] Step 12: Perform a Hilbert transform on the one-dimensional vibration vector of length N in Step 11 to obtain an envelope vector in the form of a one-dimensional time series of length N. The calculation expression for obtaining the envelope vector is as follows:

[0015]

[0016] where x(n) is the vibration signal, is the signal obtained by performing a Hilbert transform on x(n), is the envelope vector to be obtained;

[0017] Step 13: Label the corresponding envelope vector according to the bearing health state and the fault type. The label values are divided into 0 (normal) and 1 to P (P is the number of fault types). Take the envelope vector and its corresponding label as a sample, construct all samples into a data set, and divide the data set into a training set, a validation set, and a test set according to a ratio.

[0018] A further improvement of the technical solution of the present invention lies in that Step 2 includes the following steps:

[0019] Step 21: Calculate the kurtosis value k for the envelope vector in the form of a one-dimensional time series obtained in Step 1 and calculate the average skewness value by segmented sliding window

[0020] The calculation expression for the kurtosis value k is as follows:

[0021]

[0022] Where N is the length of the envelope vector, and X i is the envelope value of the i-th point of the envelope vector, μ is the mean value of the envelope vector, and σ is the standard deviation of the envelope vector;

[0023] The average skewness value The calculation expression is as follows:

[0024]

[0025] Where H is the number of segmented sliding window, h is the length of the segmented sliding window, a is the starting point of the current segmented sliding window, and X t is the envelope value of the t-th point of the envelope vector within the current segmented sliding window, μ′ is the mean value of the envelope vector within the current segmented sliding window, and σ′ is the standard deviation of the envelope vector within the current segmented sliding window;

[0026] Step 22: Use the kurtosis value k and the average skewness value obtained in Step 21 to generate parameter sequences of three Gaussian convolution kernels for the adaptive parameter setting of the Gaussian convolution kernel. The expression for generating the parameter sequences of the Gaussian convolution kernel in Step 2 is as follows:

[0027]

[0028]

[0029]

[0030] Where k is the kurtosis value, is the average skewness value, the parameter q is the length of the convolution kernel, w is an integer sequence of length q from 1 to q, and exp{} is the power operation of e;

[0031] Using the above-generated three Gaussian convolution kernel parameter sequences f1(w), f2(w), and f3(w), three Gaussian convolution kernels with adaptive parameters are obtained;

[0032] Step 23: Use the three Gaussian convolution kernels with adaptive parameters designed in Step 22 to construct three Gaussian convolution layers with adaptive parameters, extract features from the samples in the dataset obtained in Step 1, extract three fault feature vectors, and splice them to obtain a new fault feature vector.

[0033] A further improvement of the technical solution of the present invention is that the calculation expression of the cross-entropy loss function Loss in Step 3 is as follows:

[0034]

[0035] where A is the number of samples, P + 1 is the number of categories (P fault categories and 1 normal category), y jc is the sign function, which is 1 if the type of type j and sample c is the same, otherwise it is 0, and p jc is the predicted probability that the observed sample j belongs to category c.

[0036] Compared with the prior art, the beneficial effects of the rolling bearing fault diagnosis method based on the adaptive parameter Gaussian convolution kernel neural network provided by the present invention are as follows:

[0037] The rolling bearing fault diagnosis method based on the adaptive parameter Gaussian convolution kernel neural network provided by the present invention utilizes the reliability, interpretability of signal statistical mechanism knowledge and the strong fitting ability of the deep learning network, designs three Gaussian convolution kernels with adaptive parameters by using the signal kurtosis value and the average skewness value, and constructs a convolution layer by using the three adaptive Gaussian convolution kernels to extract three different features sensitive to signal impact, and splices the three features to obtain a new feature. Therefore, the present invention can extract richer and more reliable fault diagnosis information, can enhance the classification ability of fault types, and provides a new direction for the field of rolling bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0039] Figure 1 is the flow chart of the rolling bearing fault diagnosis method based on the adaptive parameter Gaussian convolution kernel neural network of the present invention;

[0040] Figure 2 is Figure 1 the flow chart of the adaptive parameter Gaussian convolution kernel generation module in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following will clearly and completely describe the technical solutions of the present invention through specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.

[0042] This embodiment will describe the present invention in detail with reference to the accompanying drawings:

[0043] Such asFigure 1 As shown in the figure, a rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network is provided, including the following steps:

[0044] Step 1: As Figure 1 shown in the figure, use an accelerometer to collect vibration signals in the form of one-dimensional time series of different health states of a rotating bearing from a rotating mechanical device. Segment and slice the collected vibration signals, calculate the envelope vector of each segment slice and label it as a sample, and make all samples into a data set. The specific operation steps are as follows:

[0045] Step 11: Divide the vibration signal in the form of one-dimensional time series into m non-overlapping segments with a length of N according to the rotational speed and sampling frequency to form m one-dimensional vibration vectors. The calculation expression for the length N is as follows:

[0046]

[0047] where f is the sampling frequency of the vibration signal and r is the rotational speed of the rotating bearing;

[0048] Step 12: Perform a Hilbert transform on the one-dimensional vibration vector with a length of N in Step 11 to obtain an envelope vector in the form of one-dimensional time series with a length of N. The calculation expression for finding the envelope vector is as follows:

[0049]

[0050] where x(n) is the vibration signal, is the signal obtained by performing a Hilbert transform on x(n), is the required envelope vector;

[0051] Step 13: Label the corresponding envelope vector according to the bearing health state and fault type. The label values are divided into 0 (normal) and 1 to P (P is the number of fault types). Take the envelope vector and its corresponding label as a sample, construct all samples into a data set, and divide the data set into a training set, a validation set, and a test set according to a ratio.

[0052] Step 2: As Figure 2 shown in the flowchart of the adaptive parameter Gaussian convolution kernel generation module, calculate the kurtosis value and the segmented sliding window to find the average skewness value for the envelope vector, which are used for setting the Gaussian convolution kernel parameters, obtain three Gaussian convolution kernels with adaptive parameters, and use the three Gaussian convolution kernels to construct an adaptive parameter Gaussian convolution layer to extract three fault feature vectors of the envelope vector and splice them to obtain a new fault feature vector. The specific operation steps are as follows:

[0053] Step 21: Calculate the kurtosis value k and the segmented sliding window to calculate the average skewness value for the envelope vector in the form of one-dimensional time series obtained in Step 1

[0054] The calculation expression of the kurtosis value k is as follows:

[0055]

[0056] where N is the length of the envelope vector, X i is the envelope value of the i-th point of the envelope vector, μ is the mean value of the envelope vector, and σ is the standard deviation of the envelope vector;

[0057] The calculation expression of the skewness value s is as follows:

[0058]

[0059] where H is the number of segmented sliding window windows, h is the length of the segmented sliding window window, a is the starting point of the current segmented sliding window window, X t is the envelope value of the t-th point of the envelope vector within the current segmented sliding window window, μ′ is the mean value of the envelope vector within the current segmented sliding window window, and σ′ is the standard deviation of the envelope vector within the current segmented sliding window window;

[0060] Step 22: Use the kurtosis value k and the average skewness value obtained in Step 21 to generate parameter sequences for three Gaussian convolution kernels for the adaptive parameter setting of the Gaussian convolution kernel. The expression for generating the parameter sequence of the Gaussian convolution kernel in Step 2 is as follows:

[0061]

[0062]

[0063]

[0064] where k is the kurtosis value, is the average skewness value, the parameter q is the convolution kernel length, w is an integer sequence of length q from 1 to q, and exp{} is the power operation of e;

[0065] Design three Gaussian convolution kernels with adaptive parameters by using the above-generated three Gaussian convolution kernel parameter sequences f1(w), f2(w), and f3(w);

[0066] Step 23: Construct three Gaussian convolution layers with adaptive parameters by using the three Gaussian convolution kernels with adaptive parameters designed in Step 22, extract features from the samples in the dataset obtained in Step 1, extract three fault feature vectors, and splice them to obtain a new fault feature vector.

[0067] Step 3: Input the new fault feature vector extracted in Step 2 into a Softmax classifier with a cross-entropy loss function to establish the mapping relationship between the fault features and the fault categories, and perform the diagnosis of the rolling bearing fault categories based on the mapping relationship. The specific operations are as follows:

[0068] Step 31: Input the fault features obtained in Step 23 into a Softmax classifier with a cross-entropy loss function to establish the mapping relationship between the fault features and the fault categories, and obtain the final bearing fault diagnosis result. The calculation expression of the cross-entropy loss function Loss is as follows:

[0069]

[0070] where A is the number of samples, P + 1 is the number of categories (P fault categories and 1 normal category), y jc is the sign function, which is 1 if the type of j and the type of sample c are the same, otherwise it is 0, and p jc is the predicted probability that the observed sample j belongs to category c.

[0071] The above-described embodiments are merely descriptions of the preferred embodiments of the present invention and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention device.

Claims

1. A rolling bearing fault diagnosis method based on an adaptive parameter Gaussian convolution kernel neural network, characterized in that It includes the following steps: Step 1: Collect vibration signals in the form of one-dimensional time series of different health states of the rotating bearing through an accelerometer. Segment and slice the collected vibration signals, calculate the envelope vector of each segment slice and label it as a sample, and make all samples into a data set; Step 2: Calculate the kurtosis value and the average skewness value by segmented sliding window for the envelope vector, which are used for setting the Gaussian convolution kernel parameters, obtain three Gaussian convolution kernels with adaptive parameters. The Gaussian convolution layer with adaptive parameters based on the three Gaussian convolution kernels with adaptive parameters is used to extract three fault feature vectors of the envelope vector and splice them to obtain a new fault feature vector; Step 2 includes the following steps: Step 21: Calculate the kurtosis value k of the envelope vector in the form of one-dimensional time series obtained in Step 1 and calculate the average skewness value by segmented sliding window The calculation expression of the kurtosis value k is as follows: where N is the length of the envelope vector, and X i is the envelope value of the i-th point of the envelope vector, μ is the mean of the envelope vector, and σ is the standard deviation of the envelope vector; The average skewness value The calculation expression is as follows: where H is the number of segmented sliding window, h is the length of the segmented sliding window, a is the starting point of the current segmented sliding window, and X t is the envelope value of the t-th point of the envelope vector within the current segmented sliding window, μ′ is the mean value of the envelope vector within the current segmented sliding window, and σ′ is the standard deviation of the envelope vector within the current segmented sliding window; Step 22: Using the kurtosis value k and the average skewness value obtained in Step 21 Generate parameter sequences for three Gaussian convolution kernels for the adaptive parameter setting of the Gaussian convolution kernel. The parameter sequence generation expression of the Gaussian convolution kernel described in Step 2 is as follows: where k is the kurtosis value, is the average skewness value, the parameter q is the convolution kernel length, w is an integer sequence of length q from 1 to q, and exp{} is the power operation of e; Using the above three Gaussian convolution kernel parameter sequences f1(w), f2(w), f3(w) generated, obtain three Gaussian convolution kernels with adaptive parameters; Step 23: Use the three Gaussian convolution kernels with adaptive parameters designed in Step 22 to construct three Gaussian convolution layers with adaptive parameters, extract features from the samples in the data set obtained in Step 1, extract three fault feature vectors, and splice them to obtain a new fault feature vector; Step 3: Input the new fault feature vector extracted in Step 2 into a Softmax classifier with a cross-entropy loss function, and based on the established mapping relationship between the fault features and the fault categories, diagnose the fault categories of the rolling bearing based on the mapping relationship.

2. The rolling bearing fault diagnosis method of the adaptive parameter Gaussian convolution kernel neural network according to claim 1, characterized in that, Step 1 includes the following steps: Step 11: Divide the vibration signal in the form of one-dimensional time series into m non-overlapping segments with a length of N according to the rotational speed and the sampling frequency, constituting m one-dimensional vibration vectors. The calculation expression of the length N is as follows: where f is the sampling frequency of the vibration signal and r is the rotational speed of the rotating bearing; Step 12: Perform Hilbert transform on the one-dimensional vibration vector with a length of N in Step 11 to obtain an envelope vector in the form of one-dimensional time series with a length of N. The calculation expression for obtaining the envelope vector is as follows: where \(x(n)\) is the vibration signal, is the signal obtained by performing the Hilbert transform on \(x(n)\), is the desired envelope vector; Step 13: Label the corresponding envelope vector according to the bearing health state and the fault type. The label values are divided into 0 (normal) and 1 to P (P is the number of fault types). Take the envelope vector and its corresponding label as a sample, construct all samples into a data set, and divide the data set into a training set, a validation set, and a test set according to a ratio.

3. The rolling bearing fault diagnosis method of the adaptive parameter Gaussian convolution kernel neural network according to claim 1, characterized in that The calculation expression of the cross-entropy loss function Loss in Step 3 is as follows: Where A is the number of samples, P + 1 is the number of classes (P failure classes and 1 normal class), and y jc is the sign function, which is 1 if the type of type j and sample c is the same, otherwise 0, and p jc is the predicted probability that the observed sample j belongs to class c.

Images