MSST-CNN-based hydroelectric generating set fault diagnosis method and system, and storage medium

Abstract

The invention discloses a hydroelectric generating set fault diagnosis method and system based on MSST-CNN, and a storage medium, and relates to the field of hydroelectric generating set fault diagnosis. The method comprises the following steps: acquiring different state vibration signals of the hydroelectric generating set; preprocessing the vibration signals in different states by using an improved multi-scale synchronous compression change method to obtain a vibration signal image; training a convolutional neural network by taking the vibration signal images as a test set and a training set to obtain an MSST-CNN model; and inputting the to-be-detected signal features into the convolutional neural network for feature extraction and fault classification. The method improves the diagnosis precision, provides a new thought for the fault diagnosis of the hydroelectric generating set, and has certain practical significance.

Description

Technical Field

[0001] The present invention relates to the field of hydropower unit fault diagnosis, and more specifically to a hydropower unit fault diagnosis method, system and storage medium based on MSST-CNN. Background Art

[0002] Hydropower units are developing towards high head and large capacity, and ensuring the safety and stability of the units has become the first priority for the operation of hydropower stations. With the frequent switching of unit operating conditions, hydraulic, mechanical and electromagnetic factors jointly affect the stable operation of the units, and traditional hydropower maintenance methods cannot effectively identify unit failures. According to statistics, 80% of hydropower unit failures are caused by vibration. Therefore, the vibration signal of the unit contains a large amount of fault information. Using vibration signals to achieve diagnosis can quickly identify unit failures and ensure safe operation of hydropower units.

[0003] In the field of rotating machinery fault diagnosis, with the development of sensor technology, it is very common to obtain multi-channel vibration signals at the same time. For example, in the field detection of large rotors, non-contact eddy current sensors are generally installed orthogonally on a certain support section, and then the vibration signals in the orthogonal direction are collected. Even if a sensor fails, on-site monitoring can be carried out. However, the time-frequency analysis technology of multi-channel vibration signals that matches it is still relatively lacking. Therefore, a new method for time-domain analysis of rotating machinery faults based on multi-scale synchronous squeezing transform (MSST) is proposed, which introduces the concept of multi-scale into synchronous squeezing transform to achieve the effect of measuring signal complexity from multiple scales.

[0004] However, the existing technology still has some defects, and it is necessary to further improve the fault diagnosis efficiency and fault diagnosis accuracy of hydropower units. Summary of the invention

[0005] In view of this, the present invention provides a hydropower unit fault diagnosis method, system and storage medium based on MSST-CNN.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A hydropower unit fault diagnosis method based on MSST-CNN includes the following steps:

[0008] Obtain vibration signals of hydropower units in different states;

[0009] The vibration signals of different states are preprocessed by using the improved multi-scale synchronous compression transform method to obtain the vibration signal image.

[0010] The vibration signal images are used as test sets and training sets to train the convolutional neural network and obtain the MSST-CNN model;

[0011] The features of the signal to be detected are input into the convolutional neural network for feature extraction and fault classification.

[0012] Optionally, the improved multi-scale synchronous compression transform method is based on improvements of short-time Fourier transform and continuous wavelet transform.

[0013] Optionally, the convolutional neural network consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the convolutional layer uses filters to detect different pattern features of image edges, shapes, and textures; the pooling layer optimizes the network structure by reducing the number of parameters and the dimension of the feature map; the fully connected layer maps the output to a linearly separable space and cooperates with the classifier to achieve classification of different images.

[0014] Optionally, the improved multi-scale synchronous compression transformation method is specifically as follows: for a time series {x 1 ,x 2 ,…,x N}, a continuous coarse-grained time series is constructed by gradually increasing the number of data points in non-overlapping windows. Each element in the coarse-grained time series All according to the equation:

[0015]

[0016] Where τ represents the scaling factor, and 1≤j≤N / τ. The length of each coarse-grained time series is the integer part of N / τ. When τ=1, the coarse-grained time series is exactly the original time series. Then the multi-scale theory is introduced into SST to calculate the image after multi-scale synchronous compression transformation.

[0017] A hydropower unit fault diagnosis method based on MSST-CNN includes the following steps:

[0018] Hydropower unit vibration signal acquisition module: used to obtain vibration signals of hydropower units in different states;

[0019] Vibration signal preprocessing module: used to preprocess vibration signals of different states using an improved multi-scale synchronous compression transformation method to obtain vibration signal images;

[0020] Convolutional neural network training module: used to train the convolutional neural network using vibration signal images as test sets and training sets to obtain the MSST-CNN model;

[0021] Fault diagnosis module: used to input the features of the signal to be detected into the convolutional neural network for feature extraction and fault classification.

[0022] A computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of any one of the methods for diagnosing hydropower unit faults based on MSST-CNN are implemented.

[0023] It can be seen from the above technical solution that compared with the prior art, the present invention provides a hydropower unit fault diagnosis method, system and storage medium based on MSST-CNN, which improves the diagnosis accuracy, provides a new idea for hydropower unit fault diagnosis, and has certain practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0025] Figure 1 It is a schematic diagram of the fault diagnosis method of a hydropower unit of the present invention;

[0026] Figure 2 Schematic diagram of the MSST-CNN training process;

[0027] Figure 3 Schematic diagram of the CWT-CNN training process;

[0028] Figure 4 Schematic diagram of the MSST-CNN confusion matrix;

[0029] Figure 5 This is a schematic diagram of the CWT-CNN confusion matrix;

[0030] Figure 6 It is a schematic diagram of the convolutional neural network structure of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Example 1

[0033] The embodiment of the present invention discloses a hydropower unit fault diagnosis method based on MSST-CNN. Figure 1 As shown, the following steps are included:

[0034] Step 1: Obtain vibration signals of hydropower units in different states;

[0035] Step 2: preprocess the vibration signals of different states using the improved multi-scale synchronous compression variation method to obtain the vibration signal image;

[0036] Step 3: Use the vibration signal image as a test set and training set to train the convolutional neural network to obtain the MSST-CNN model;

[0037] Step 4: Input the features of the signal to be detected into the convolutional neural network for feature extraction and fault classification.

[0038] Furthermore, in step 2, since the traditional synchronous squeezing transform cannot measure the time complexity of time series signals from multiple time scales, influenced by the multi-scale entropy theory of Costa et al., the multi-scale theory is combined with the synchronous squeezing transform to propose a new image transformation method, Multi-synchro squeezing Transform (MSST).

[0039] Synchro squeezing transform (SST) is an improvement based on short-time Fourier transform (STFT) and continuous wavelet transform (CWT). Compared with traditional STFT and CWT, SST has better adaptability in the analysis process and can be flexibly adjusted according to the different characteristics of the signal. SST has higher resolution in the time-frequency domain, can more accurately distinguish signal components, improve signal quality and reliability, obtain higher time-frequency resolution, and has obvious advantages in suppressing cross terms.

[0040] Short-time Fourier transform is a joint time-frequency analysis method for time-varying and non-stationary signals. It can transform a one-dimensional fault vibration signal into a two-dimensional matrix suitable for CNN processing, a characteristic spectrum containing time domain and frequency domain information. The basic idea of ​​STFT is to use a fixed-length window function to intercept the time domain signal, and perform Fourier transform on the intercepted signal to obtain the local spectrum in a very small time period near time t. By translating the window function on the entire time axis, the final transformation is to obtain a set of local spectra in each time period. Therefore, STFT is a two-dimensional function of time and frequency. The basic calculation formula is as follows:

[0041]

[0042] Where: f(t) is the time domain signal, g(t-τ) is the time window centered at time τ. From this, it can be seen that STFT is the Fourier transform of the signal f(t) multiplied by a window function g(t-τ) centered at τ.

[0043] Continuous Wavelet Transform (CWT) is a tool for analyzing signals in the time-frequency domain, which allows us to capture the local frequency characteristics of non-stationary signals. Unlike Fourier transform, continuous wavelet transform can provide frequency resolution that varies with time, so it is particularly suitable for analyzing the situation where the frequency in time series varies with time.

[0044] Assume Ψ(t) is a square integrable function, that is, Ψ(t)∈L 2 (R), if its Fourier transform Satisfying the admissibility condition:

[0045]

[0046] Then Ψ(t) is called the mother wavelet. The mother wavelet Ψ(t) is scaled and translated. Let its scaling factor (scale factor) be a, and its translation factor be b. Let the function after scaling and translation be Ψ a,b (t), then:

[0047]

[0048] Name a,b (t) is a wavelet basis function that depends on parameters a and b, also called continuous wavelet basis function. Continuous time signal x(t)∈L 2 The continuous wavelet transform of (R) is defined as:

[0049]

[0050] For a discrete time series x m , let t = mδt, b = nδt, where m, n = 0, 1, 2…, N-1, N is the number of sampling points, δt is the sampling interval, then the discrete time series x m The continuous wavelet transform of is:

[0051]

[0052] By changing the time indexes j and n corresponding to the scale factor a and the translation factor b, a continuous wavelet transform coefficient matrix can be obtained, which reflects the change of the amplitude of the continuous wavelet transform coefficients with time and scale.

[0053] For a time series, we first construct a continuous coarse-grained time series by successively increasing the number of data points in non-overlapping windows. Each element in the coarse-grained time series (i.e. All according to the equation:

[0054]

[0055] Where τ represents the scaling factor, and 1≤j≤N / t. For the case where the scaling factor is 1, the coarse-grained time series is the original time series.

[0056] Then, the multi-scale theory is introduced into SST to calculate the image after multi-scale synchronous compression transformation, and then CNN is used for image recognition.

[0057] As an important subfield of deep learning, convolutional neural network is one of the most popular deep learning algorithms and has a significant effect on processing two-dimensional image data. Unlike traditional fully connected networks, CNN uses shared weights and local connections to make full use of two-dimensional input data structures such as image signals, which greatly reduces the number of hyperparameters, simplifies the training process, and speeds up the network. Compared with previous neural network algorithms, CNN has the advantages of equivalent representation, sparse interaction, and parameter sharing, and has good applications in computer vision, language processing, and face recognition.

[0058] like Figure 6 As shown in the figure, the convolutional neural network consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Its structure can contain multiple convolutional layers and pooling layers. Among them, the convolutional layer can detect different pattern features such as image edges, shapes, and textures by using filters. The pooling layer optimizes the network structure by reducing the number of parameters and the dimension of the feature map. As the last layer of the CNN structure, the fully connected layer can map the output to a linearly separable space and cooperate with the classifier to classify different images.

[0059] This embodiment also discloses a hydropower unit fault diagnosis method based on MSST-CNN, comprising the following steps:

[0060] Hydropower unit vibration signal acquisition module: used to obtain vibration signals of hydropower units in different states;

[0061] Vibration signal preprocessing module: used to preprocess vibration signals of different states using an improved multi-scale synchronous compression transformation method to obtain vibration signal images;

[0062] Convolutional neural network training module: used to train the convolutional neural network using vibration signal images as test sets and training sets to obtain the MSST-CNN model;

[0063] Fault diagnosis module: used to input the features of the signal to be detected into the convolutional neural network for feature extraction and fault classification.

[0064] Finally, a computer storage medium is disclosed, on which a computer program is stored. When the computer program is executed by a processor, any one of the steps of a hydropower unit fault diagnosis method based on MSST-CNN is implemented.

[0065] Example 2

[0066] In order to further verify the high fault detection accuracy of the present invention, the following steps are disclosed:

[0067] Step 1, data collection: With the continuous development of the hydropower industry, the large-scale, high-power and integrated degree of hydropower units are constantly improving, resulting in the high cost and certain danger of on-site experiments for fault diagnosis of hydropower units. Therefore, the present invention simulates four common states of units: normal (NOR), friction (FRI), unbalanced (IMB) and misaligned (MIS) through a rotor test bench. Among them, the unit speed is set to 1200r / min, the sampling frequency is 2048Hz, and 100 sets of data are collected for each of the four states of the unit, and each set of data contains 2048 vibration signals.

[0068] Step 2: Data processing: The vibration signal image features after MSST conversion are input into the CNN convolutional neural network for feature extraction and fault classification, and a new method for hydropower unit fault diagnosis based on MSST-CNN is proposed.

[0069] Specific simulation process: The fault diagnosis of hydropower units based on MSST-CNN uses the MSST image of the vibration signal as the signal feature, builds a fault diagnosis model for hydropower units based on the classic CNN, and trains the model using MSST image samples.

[0070] The parameters are set as follows: initial learning rate is 0.0001, number of iterations is 10, maximum number of iterations is 150, MiniBatchSize is 4, and MaxEpochs is 6. The CNN network contains 1 two-dimensional convolution layer, two ReLU activation functions, a Softmax layer, and a classification output layer. In order to verify the superiority of the proposed method, CWT and MSST are introduced for comparison in the feature extraction process. Figure 2 and Figure 3 These are the training process curves of the MSST-CNN and CWT-CNN models respectively.

[0071] The specific classification results of MSST-CNN and CWT-CNN are as follows Figure 4 and Figure 5 As shown. Figure 4It can be seen that the MSST-CNN model achieved a 100% recognition rate and can effectively distinguish the four unit states. Figure 5 It can be seen that the CWT-CNN model achieved a recognition rate of 97.3%, 1 "NOR" fault state was misjudged as "FRI" state, 3 "FRI" fault states were misjudged as "NOR" state, and 4 "FRI" fault states were misjudged as "MIS" state. In summary, the model comparison verifies that MSST-CNN is a hydropower unit fault diagnosis model with good diagnostic accuracy.

[0072] This paper proposes a hydropower unit fault diagnosis method based on the combination of multi-scale synchronous compression and convolutional neural network, and draws the following conclusions through experiments:

[0073] Since the time-frequency analysis technology of multi-channel vibration signals is still relatively lacking, a new method of time-domain analysis of rotating machinery faults based on multi-scale synchronous compression is proposed. The multi-scale concept is introduced into the synchronous compression transform to achieve the effect of measuring signal complexity from multiple scales.

[0074] The recognition of unit status by the MSST-CNN and CWT-CNN models was compared, and it was found that the proposed model achieved the highest diagnostic accuracy, reaching 100.00%, which verified that the MSST-CNN model is a hydropower unit fault diagnosis model with good diagnostic accuracy.

[0075] The experimental results show that the hydropower unit diagnosis method proposed in the present invention is a method with excellent diagnostic accuracy, which provides a new idea for hydropower unit fault diagnosis and has certain practical significance.

[0076] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0077] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault diagnosis method for hydropower units based on MSST-CNN, characterized in that: The following steps are involved: Obtain vibration signals of hydropower units in different states; The vibration signals of different states are preprocessed by using the improved multi-scale synchronous compression transform method to obtain the vibration signal image. The vibration signal images are used as test sets and training sets to train the convolutional neural network and obtain the MSST-CNN model; The features of the signal to be detected are input into the convolutional neural network for feature extraction and fault classification.

2. A hydropower unit fault diagnosis method based on MSST-CNN according to claim 1, characterized in that: The improved multi-scale synchronous compression transform method is based on the improvement of short-time Fourier transform and continuous wavelet transform.

3. A hydropower unit fault diagnosis method based on MSST-CNN according to claim 1, characterized in that: The convolutional neural network consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the convolutional layer uses filters to detect different pattern features of image edges, shapes, and textures; The pooling layer optimizes the network structure by reducing the number of parameters and feature map dimensions; the fully connected layer maps the output to a linearly separable space and cooperates with the classifier to achieve classification of different images.

4. A hydropower unit fault diagnosis method based on MSST-CNN according to claim 1, characterized in that: The improved multi-scale synchronous compression transformation method is specifically as follows: for a time series {x1 ,x2,…,x N }, a continuous coarse-grained time series is constructed by gradually increasing the number of data points in non-overlapping windows. Each element in the coarse-grained time series All according to the equation: Where τ represents the scaling factor, and 1≤j≤N / τ; the length of each coarse-grained time series is the integer part of N / τ; the multi-scale theory is introduced into SST to calculate the image after multi-scale synchronous compression transformation.

5. A hydropower unit fault diagnosis method based on MSST-CNN, characterized in that: The following steps are involved: Hydropower unit vibration signal acquisition module: used to obtain vibration signals of hydropower units in different states; Vibration signal preprocessing module: used to preprocess vibration signals of different states using an improved multi-scale synchronous compression transformation method to obtain vibration signal images; Convolutional neural network training module: used to train the convolutional neural network using vibration signal images as test sets and training sets to obtain the MSST-CNN model; Fault diagnosis module: used to input the features of the signal to be detected into the convolutional neural network for feature extraction and fault classification.

6. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program is executed by the processor, the steps of a hydropower unit fault diagnosis method based on MSST-CNN as described in any one of claims 1 to 5 are implemented.

Images