A fault diagnosis method, system and device for a main circulating pump in a converter valve and a medium

Abstract

A kind of fault diagnosis method, system, device and medium of main circulating pump in converter valve, characterized in that, the method includes the following steps: step 1, the three-way vibration signal of main circulating pump in converter valve is collected, and the three-way vibration signal is denoised using VMD-SVD algorithm;Step 2, respectively the denoised one-way vibration signal is converted into two-dimensional gray image, and the two-dimensional gray image is sequentially input into different color channels and then superimposed to obtain feature fusion image;Step 3, the feature fusion image is input into convolutional neural network to realize fault feature extraction.The method is effective and reliable, and the algorithm efficiency is high, and different fault types can be reasonably traced from fault features.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing, and more specifically, to a fault diagnosis method, system, device, and medium for the main circulating pump in a converter valve. Background Technology

[0002] With the increase in DC transmission capacity and voltage levels, converter valves, as core equipment in DC transmission projects, directly affect the safety and stability of the DC transmission system. During operation, converter valves generate a large amount of heat. The main circulating pump of the valve cooling system drives the flow of cooling water to control the junction temperature within a reasonable range. Due to changes in operating conditions and operating status, the main circulating pump is prone to failure, which significantly reduces the heat exchange efficiency of the valve cooling system. This can lead to excessively high junction temperatures in the converter valves, causing DC blockage and severely affecting the stability requirements of the DC transmission system. Therefore, real-time fault diagnosis of the main circulating pump in the valve cooling system is of great significance. Currently, fault diagnosis for the main circulating pump mainly focuses on the analysis of its vibration signals.

[0003] Domestic experts and scholars have conducted active research on fault diagnosis of electrical equipment based on vibration signals. In the extraction of fault features from vibration signals, different research directions have been developed, including time-domain analysis, frequency-domain analysis, and time-frequency analysis. Among these, time-frequency analysis best reflects the time and frequency domain characteristics of non-stationary vibration signals. For example, Zhao Shutao et al. used the Complete EEMD with Adaptive Noise (CEEMDAN) method to decompose non-stationary signals and extract the shape entropy features of the component power spectrum, thus realizing the identification of the noisy state of high-voltage circuit breakers. Yu Chunyu et al. combined Empirical Mode Decomposition (EMD) and an Autoregressive Model to obtain the vibration signal feature matrix by solving the autoregressive model parameters and residual terms, thus realizing the fault diagnosis and location of rolling bearings. Li Bing et al. proposed a motor bearing fault diagnosis method based on an improved sine-cosine algorithm (ISCA) to optimize a stacked denoising auto encoder (SDAE), which achieved good fault diagnosis results.

[0004] Furthermore, deep learning algorithms have demonstrated excellent performance in electrical equipment fault diagnosis and vibration signal processing. For example, Lei Chunli et al. proposed a deep learning model based on transfer learning and an improved residual neural network (ResNet), which effectively diagnosed the faults of rolling bearings in wind turbines. Zhang Xunjie et al. converted vibration signals into two-dimensional images and combined convolutional neural networks (CNNs) and bidirectional gated recurrent units (BiGRUs) to achieve fault mode recognition for rolling bearings. Zhang Long et al. used recursive graph encoding to convert vibration signals into two-dimensional images and used a residual network (ResNet) to adaptively extract fault features from the rolling bearing vibration signals, thus achieving fault diagnosis.

[0005] However, to date, there is still a lack of accurate and effective intelligent algorithm-based fault diagnosis in the field of main circulation pump vibration signal diagnosis. Furthermore, when the main circulation pump experiences various failures, existing methods struggle to accurately pinpoint the source of the fault from similar vibration signals, given the diverse causes of these failures.

[0006] Furthermore, fault signals themselves contain relatively rich three-dimensional spatial information, and depending on the different causes of the fault, there are distinguishable differences in the three-dimensional spatial information of the vibration signals. However, current technologies do not yet use these differences to further determine the cause of the fault. This makes it difficult to accurately trace similar vibration signals under different fault causes. For example, when a coupling experiences two different faults—angular misalignment and parallel misalignment—or two faults—bearing looseness and rotor imbalance—the causes of the faults may be completely different, but the results, that is, the situation reflected in the vibration signals, may be very similar. If the spatial information of the fault signals cannot be deeply explored and analyzed, it is impossible to accurately classify the above two types of faults.

[0007] In addition, the efficiency of existing intelligent algorithms is relatively low, and most of the vibration signals collected by the main circulation pump are in a normal, non-faulty state, which further reduces the efficiency of the algorithm in troubleshooting.

[0008] To address the aforementioned issues, there is an urgent need for a new method, system, device, and medium for fault diagnosis of the main circulating pump in a converter valve. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a fault diagnosis method, system, device, and medium for the main circulating pump in a converter valve. By collecting the three-dimensional vibration signal of the main circulating pump, fault extraction is achieved through noise reduction, image conversion, and convolutional neural networks.

[0010] The present invention adopts the following technical solution.

[0011] The first aspect of this invention relates to a fault diagnosis method for the main circulating pump in a converter valve. The method includes the following steps: Step 1, acquiring the three-dimensional vibration signal of the main circulating pump of the converter valve, and denoising the three-dimensional vibration signal using the VMD-SVD algorithm; Step 2, converting the denoised unidirectional vibration signal into a two-dimensional grayscale image, and then sequentially inputting the two-dimensional grayscale image into different color channels and superimposing them to obtain a feature fusion image; Step 3, inputting the feature fusion image into a convolutional neural network to achieve fault feature extraction.

[0012] Preferably, the three-dimensional vibration signal is the axial, vertical and longitudinal vibration components of the main circulation pump collected by an accelerometer installed on the pump end bearing housing of the main circulation pump; the vibration frequency measurement range of the accelerometer is between 0.2Hz and 10kHz, and the sampling frequency is 12.8kHz.

[0013] Preferably, the VMD-SVD algorithm denoises the unidirectional vibration signal in each direction of the triaxial vibration signal.

[0014] Preferably, the unidirectional vibration signal is converted into a two-dimensional grayscale image using the GAF algorithm, and each two-dimensional grayscale image is labeled with a direction label and a time label.

[0015] Preferably, the historical operating status of the main circulation pump is used as a two-dimensional grayscale image to mark the status labels; the status labels are: main circulation pump normal, main circulation pump rotor unbalanced, coupling angle misalignment, coupling parallel misalignment, and bearing loose.

[0016] Preferably, two-dimensional grayscale images with corresponding time labels are superimposed and then input into the AlexNet network to extract fault features.

[0017] Preferably, the fault features extracted by the AlexNet network are associated with the state labels to achieve fault classification.

[0018] Preferably, in the training set, 2 samples are drawn each time. N The unextracted feature fusion images are used as batch samples and input into the AlexNet network for training until the training of all samples in the training set is completed, thus confirming the completion of one iteration process; N is a natural number.

[0019] Preferably, the order of samples in the training set is randomly rearranged before each iteration.

[0020] A second aspect of the present invention relates to a fault diagnosis system for the main circulating pump in a converter valve. The system implements the steps of the fault diagnosis method for the main circulating pump in a converter valve according to the first aspect of the present invention. The system includes an acquisition module, a fusion module, and an extraction module. The acquisition module includes an accelerometer and a denoising unit. The accelerometer is used to acquire the triaxial vibration signal of the main circulating pump in the converter valve. The denoising unit uses the VMD-SVD algorithm to denoise the triaxial vibration signal. The fusion module converts the denoised uniaxial vibration signal into a two-dimensional grayscale image, and then sequentially inputs the two-dimensional grayscale image into different color channels and superimposes them to obtain a feature fusion image. The extraction module inputs the feature fusion image into a convolutional neural network to extract fault features.

[0021] A third aspect of the present invention relates to a fault diagnosis device for the main circulating pump in a converter valve, the device comprising a processor and a storage medium; wherein the storage medium is used to store instructions; and the processor is used to perform operations according to the instructions to execute the steps of the method in the first aspect of the present invention.

[0022] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method of the first aspect of the present invention.

[0023] The beneficial effects of this invention are that, compared with the prior art, the fault diagnosis method, system, device, and medium for the main circulating pump in a converter valve of this invention can generate two-dimensional grayscale images by collecting the three-dimensional vibration signals of the converter valve, and comprehensively consider the weak vibration features in three different spatial dimensions, using a convolutional neural network to achieve effective and accurate fault feature extraction. The method of this invention is effective and reliable, with high algorithm efficiency, and can achieve reasonable source tracing of different fault types from fault features.

[0024] The beneficial effects of the present invention also include:

[0025] 1. By separately acquiring vibration signals from different directions in three-dimensional space and then denoising and converting them into images, anomalies in the vibration signals for each direction are accurately acquired. For various faults in the main circulation pump caused by different reasons, even if the fault manifestations are similar, there will be subtle differences in a certain spatial dimension. However, for this invention, these subtle differences can be precisely reflected by separately generated two-dimensional grayscale images. Therefore, the algorithm in this invention, which uses superimposed fused images as input, better ensures the acquisition of weak spatial information in a certain dimension, improves the algorithm's ability to identify local differences, and ensures the accuracy of fault classification.

[0026] 2. This invention eliminates high-frequency noise in the vibration signal of the main circulation pump through signal denoising, thereby improving the accuracy of fault diagnosis for the main circulation pump. Furthermore, it overcomes the shortcomings of traditional fault detection systems for main circulation pumps, such as poor targeting, low efficiency, and low level of intelligence.

[0027] 3. The method of the present invention uses RGB channels to realize the superposition of two-dimensional grayscale images in three directions, which cleverly realizes the joint detection of vibration signals in different directions in the same time by convolutional neural network.

[0028] 4. To improve algorithm efficiency, the iterative process of the AlexNet network was modified in this invention. This modification alters the input method and batch calculation method of the AlexNet network without changing the algorithm model, thereby ensuring that the algorithm can achieve fast convergence even when a large number of samples are normal samples, and ensuring the fault feature extraction rate of the AlexNet network. Attached Figure Description

[0029] Figure 1 This is a schematic diagram illustrating the steps of a fault diagnosis method for the main circulating pump in a converter valve according to the present invention.

[0030] Figure 2 This is a schematic diagram illustrating the conversion of unidirectional vibration signals into two-dimensional grayscale images in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention.

[0031] Figure 3 This is a schematic diagram of the hierarchical structure of the AlexNet network in the fault diagnosis method of the main circulating pump in the converter valve of the present invention.

[0032] Figure 4 This is a schematic diagram of the VMD decomposition of vibration signals in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention.

[0033] Figure 5 This is a schematic diagram of vibration signal denoising in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention.

[0034] Figure 6 This is a schematic diagram illustrating the effect of the AlexNet iterative training process in the fault diagnosis method of the main circulation pump in the converter valve of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments not described in this invention obtained by those skilled in the art based on the embodiments described in this invention without creative effort should fall within the protection scope of this invention.

[0036] Figure 1 This is a schematic diagram illustrating the steps of a fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 1 As shown, in a first aspect, the present invention relates to a fault diagnosis method for the main circulating pump in a converter valve, the method comprising steps 1 to 3.

[0037] Step 1: Collect the three-dimensional vibration signal of the main circulation pump of the converter valve, and use the VMD-SVD algorithm to denoise the three-dimensional vibration signal.

[0038] As an important auxiliary system in the power system, the valve cooling system drives the circulation of cooling water through the main circulating pump, thereby cooling and dissipating heat from the converter valves. This ensures the valve bodies operate within a reasonable temperature range, which is beneficial to the operational stability of the converter valves and extends the service life of valve components. Due to changes in operating conditions and operating status, the main circulating pump is prone to failure, resulting in abnormal vibration signals. Compared with the vibration signal of the main circulating pump under normal operating conditions, the vibration signal under fault conditions is reflected in three directions: vertical, longitudinal, and axial.

[0039] Therefore, in this invention, multidimensional vibration signals can be extracted from the main circulation pump of the valve cooling system, and these vibration signals can be used as indicators to accurately analyze various faults of the main circulation pump.

[0040] Preferably, the three-dimensional vibration signal is the axial, vertical and longitudinal vibration components of the main circulation pump collected by an accelerometer installed on the pump end bearing housing of the main circulation pump; the vibration frequency measurement range of the accelerometer is between 0.2Hz and 10kHz, and the sampling frequency is 12.8kHz.

[0041] Because the vibration frequency of moving equipment is relatively high, the selected vibration sensor must possess characteristics such as a wide frequency response range, good stability, reliable operation, high accuracy, and strong anti-interference capability. In one embodiment of this invention, a dedicated accelerometer M603M107 (measuring range 0.2Hz~10kHz) is used to collect the vibration signal for the moving equipment. The sampling frequency of the aforementioned sensor is 12.8kHz.

[0042] This invention sets three vibration signal measuring points on the pump end bearing housing of the main circulation pump to measure vertical, longitudinal, and axial vibration signals, respectively. The sensors are attached to the outer surface of the pump end bearing housing and do not directly contact the rotating parts; therefore, the sensors do not rotate with the rotation of the main circulation pump shaft.

[0043] After acquiring the three-dimensional vibration signal, the method of this invention can also accurately extract the feature components of the vibration signal by eliminating noise. Before further extracting the feature components, this invention uses the VMD-SVD (Variational Mode Decomposition-Singular Value Decomposition) algorithm to jointly remove noise from the vibration signal.

[0044] Analysis reveals that the actual purpose of this invention is to extract fault signals from the normal vibration signals of the main circulation pump. Therefore, the noise mentioned here mainly refers to the relatively stable high-frequency components generated during the normal operation of the main circulation pump. In contrast, fault signals typically exhibit more irregular characteristics, such as non-periodic features or imbalances in three-dimensional space.

[0045] Preferably, the VMD-SVD algorithm denoises the unidirectional vibration signal in each direction of the triaxial vibration signal.

[0046] Understandably, this invention first uses the VMD algorithm to perform preliminary decomposition of the vertical, longitudinal, and axial signals, and selects the optimal modal components based on the correlation coefficient method. Subsequently, a Hankle matrix can be constructed using SVD to decompose and denoise the multiple optimal modal components selected in the VMD algorithm, and the denoised intrinsic modal components are spatially reconstructed to obtain a noise-free vibration signal. Subsequent steps of this invention also support converting the denoised vibration signal into a vibration image to train a fault diagnosis model, which can be used to improve the accuracy of fault diagnosis.

[0047] Specifically, to address the problems of unspecified intrinsic mode components and mode confusion in traditional Empirical Mode Decomposition (EMD), Dragomiretskiy et al. proposed a signal decomposition method using non-recursive variational modes, namely Variational Mode Decomposition (VMD), in 2014. This method uses a variational problem as the overall framework to decompose the original signal into k finite-bandwidth intrinsic mode components, each with a center frequency within its finite bandwidth.

[0048] The variational mode decomposition model can be represented as:

[0049]

[0050] In the formula, {U k}={u1,…,u k} represents the set of k eigenmode components obtained from the decomposition.

[0051] {W k}={ω1,…,ω k} represents the set of center frequencies corresponding to each intrinsic mode component;

[0052] In the above set, u k This represents the k-th intrinsic mode component.

[0053] ω k This represents the center frequency corresponding to the k-th intrinsic mode component.

[0054] δ(t) represents the impulse function.

[0055] This indicates finding the partial derivative.

[0056] f represents the vibration signal being decomposed.

[0057] In the VMD algorithm, the model can be transformed into an unconstrained variational problem using a quadratic penalty factor α and the Lagrange multiplication operator λ(t). The extended Lagrange expression then becomes:

[0058]

[0059] By using the alternating direction method of multipliers to find saddle points, the frequency domain updates of each mode are obtained, along with the modal components in the frequency domain. Center frequency Lagrange multiplication operators After the (n+1)th iteration, the solution is as follows:

[0060]

[0061]

[0062]

[0063] Where τ is the noise tolerance, which can be selected according to the fidelity requirements of signal decomposition.

[0064] The above method is used to perform multiple iterative calculations to guide the achievement of accuracy requirements. Then the iteration stops. Here, ε represents the convergence precision, which takes a value greater than 0.

[0065] After decomposing into k consecutive signal components by the above method, the effective components can be selected based on the cross-correlation coefficient method. First, the cross-correlation coefficient between each IMF (Intrinsic Mode Functions) and the original signal can be calculated. After the calculation, the cross-correlation coefficients are sorted, and the IMF components corresponding to the top-ranked coefficients can form the optimal mode components in the present invention.

[0066] It should be noted that in an embodiment of the present invention, the selection of the optimal mode components can be achieved by selecting a cross-correlation coefficient threshold. The components with cross-correlation coefficients greater than the threshold are used as the optimal mode components, while those with small cross-correlation coefficients are excluded.

[0067] In addition, singular value decomposition, as an important matrix decomposition method, has good numerical stability and has been widely used in signal denoising, data processing, etc. For the discrete vibration signal X i ={x i (1), x i (2), …, x i (N)}, based on the phase space reconstruction theory, Cycle matrix, Toeplitz matrix and Hankel matrix can be constructed. Different matrix constructions have different effects on signal denoising. In the present invention, the Hankel matrix is selected to reconstruct the one-dimensional discrete signal into a two-dimensional matrix, and the singular value decomposition of the matrix is performed.

[0068] The two-dimensional matrix is:

[0069]

[0070] where N is the number of sampling points of the discrete signal, w is the signal truncation length, and the above parameters satisfy 1 < w < N, m = N - w + 1. The real sampling values of multiple truncated discrete signals are repeatedly filled into the matrix to obtain the above two-dimensional matrix. For the above matrix, singular value decomposition can be performed to obtain

[0071] X = USV T

[0072] where U is the left singular matrix, and U ∈ R (N-w+1)×(N-w+1) ,

[0073] V T is the right singular matrix, and V T ∈ R w×w ,

[0074] S is a diagonal matrix, and s ∈ R (N-w+1)×w .

[0075] The main diagonal elements of S are the singular values of matrix X, which are λ i(i = 1, 2, ..., k), where k = min((N - w + 1), n).

[0076] Therefore, the diagonal matrix S = diag(λ1, λ2, ..., λ) k ).

[0077] The singular values ​​of the matrix satisfy λ1≥λ2≥…≥λ k ≥0.

[0078] In this invention, the effective order of singular value decomposition (SVD) can be determined using the one-sided maxima principle, and the optimal modal components can be denoised separately through SVD. Finally, the modal components after SVD can be reconstructed to obtain the denoised vibration signal.

[0079] Noise removal achieved in this way has smaller errors and can better avoid under-decomposition and over-decomposition of the signal during the mode decomposition process. In addition, after noise removal, each mode component still retains good temporal continuity characteristics, which also ensures the effectiveness of subsequent image processing of the vibration signal.

[0080] Step 2: Convert the denoised unidirectional vibration signal into a two-dimensional grayscale image, and then input the two-dimensional grayscale image into different color channels and superimpose them to obtain a feature fusion image.

[0081] Figure 2 This is a schematic diagram illustrating the conversion of unidirectional vibration signals into a two-dimensional grayscale image in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 2 As shown, after signal denoising, the noise-removed vibration signal can be converted into a vibration image, thereby enabling convolution operations and fault feature analysis on the vibration image.

[0082] Preferably, the unidirectional vibration signal is converted into a two-dimensional grayscale image using the GAF (Gramian Angular Field) algorithm, and each two-dimensional grayscale image is labeled with a direction label and a time label.

[0083] Specifically, the main circulation pump vibration signal dataset consists of acceleration signals collected by vibration sensors. The vibration signals measured in three directions—axial, vertical, and longitudinal—are typical time-series signals, which can comprehensively reflect the vibration status of the main circulation pump under fault conditions. However, since the vibration signals maintain a strong dependence on time, the data input into the subsequent convolutional fault diagnosis model must retain as much temporal feature information as possible in order to preserve more fault information in both the time and frequency domains.

[0084] Therefore, the axial, vertical, and longitudinal vibration signals can be converted into two-dimensional images using a Gram matrix, and the two-dimensional images transformed by the Gram matrix retain the time information of the original signals.

[0085] During image transformation, the average value of multiple points can be selected to aggregate the time series. In one embodiment of the present invention, a piecewise aggregation approximation method can be used. Simultaneously, a normalization method can be used to constrain the range of all values ​​in the matrix. Then, polar coordinates are generated by using the inverse cosine of the timestamp radius and the scaling value, thus providing the angle values. Finally, an image of the Gram angle field is generated.

[0086] It should be noted that, in order to achieve the subsequent sample image fusion process, orientation labels and time labels can be constructed for each generated image sample to enable image sample matching in the subsequent process.

[0087] Preferably, the historical operating status of the main circulation pump is used as a two-dimensional grayscale image to mark the status labels; the status labels are: main circulation pump normal, main circulation pump rotor unbalanced, coupling angle misalignment, coupling parallel misalignment, and bearing loose.

[0088] In addition, the present invention can also label image samples according to the operating status of the main circulation pump. It is understood that each image sample can be converted from a signal within a short period of time extracted from a continuous vibration signal. Therefore, not only can time tags be added to image samples, but if a main circulation pump failure occurs within this short period, the failure should also be able to characterize the state of the failure at a specific point in time.

[0089] Therefore, the present invention can assign a fault state to each image. Table 1 shows the vibration signal labels, i.e., the state labels, in one embodiment of the present invention. As shown in the table, different fault types correspond to different label values.

[0090]

[0091] Table 1 Fault Type Table

[0092] It is understandable that, depending on the actual model of the main circulation pump used and the different causes of the faults that need to be collected, other fault state types can also be reasonably designed in this invention based on the accuracy of the algorithm, and corresponding to the corresponding label values.

[0093] Preferably, two-dimensional grayscale images with corresponding time labels are superimposed and then input into the AlexNet network to extract fault features.

[0094] It is understood that in this invention, axial, vertical and longitudinal vibration signals of the main circulation pump of the valve cooling system are measured, and the two-dimensional vibration grayscale image is used as the input layer of the network with three RGB (Red, Green and Blue) channels in parallel. Convolutional layers and pooling layers are used for feature fusion, which improves information utilization while avoiding information bias caused by manual feature extraction.

[0095] This invention collects new data in real time on-site, imports the preprocessed new data into the training set, and retrains the model to continuously improve the model's prediction accuracy.

[0096] Step 3: Input the feature fusion image into the convolutional neural network to achieve fault feature extraction.

[0097] Figure 3 This is a schematic diagram of the hierarchical structure of the AlexNet network in the fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 3 As shown, the convolutional neural network used in this invention can be AlexNet.

[0098] Preferably, the fault features extracted by the AlexNet network are associated with the state labels to achieve fault classification.

[0099] Figure 5 This is a schematic diagram illustrating vibration signal denoising in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 5 As shown, this invention not only divides the vibration image dataset into training and testing sets, inputting the training set into the AlexNet neural network and using the testing set for testing and verification, but also evaluates the model's output results using the accuracy. Furthermore, this invention also associates feature classification methods with different types of fault states in the output layer of AlexNet, thereby achieving the extraction and output of fault types.

[0100] It is understood that the fault diagnosis network framework in this invention can be built based on PyTorch. Table 2 shows the parameters of each layer in a typical AlexNet network architecture in this invention.

[0101] Convolution kernel and fully connected layer parameter settings Convolutional layer 1 kernel_size=11, stride=4, padding=1 Pooling layer 1 kernel_size=3, stride=2 Convolutional layer 2 kernel_size=5, stride=1, padding=2 Pooling layer 2 kernel_size=3, stride=2 Convolutional layer 3 kernel_size=3, stride=1, padding=1 Convolutional layer 4 kernel_size=3, stride=1, padding=1 Convolutional layer 5 kernel_size=3, stride=1, padding=1 Pooling layer 3 kernel_size=3, stride=2 Fully connected layer 1 input = 9216, output = 4096 Fully connected layer 2 input = 4096, output = 4096 Output layer input = 4096, output = 5

[0102] Table 2 AlexNet Network Architecture Parameters

[0103] In this design, the weights of the convolutional layers are randomly initialized, and the ReLU activation function is chosen to accelerate model training and convergence. Max pooling is used in the pooling layers to enhance the robustness of image recognition. Due to the large number of neurons in the fully connected layers, dropout layers with a dropout probability of 0.5 are used to mitigate overfitting. The output layer uses softmax to output the class probability vector.

[0104] Preferably, in the training set, 2 samples are drawn each time. N The unextracted feature fusion images are used as batch samples and input into the AlexNet network for training until the training of all samples in the training set is completed, thus confirming the completion of one iteration process; N is a natural number.

[0105] By extracting samples in small batches, this invention achieves a faster sample reading speed. This not only significantly reduces the computer's processing speed but also ensures that the equipment cost and related parameters supporting the algorithm are reasonably controlled. In one embodiment of this invention, image samples can be read in multiples of 16, such as 32 or 64, and the calculation of all samples in the training library can be achieved through multiple training iterations.

[0106] Preferably, the order of samples in the training set is randomly rearranged before each iteration.

[0107] Understandably, the AlexNet network in this invention can achieve model convergence through multiple iterations. If the order is not shuffled in each iteration, the sampling process needs to ensure sampling speed, so the sampling itself is usually based only on the original order of the samples, such as the order of sample images generated by vibration time. In this case, the mini-batch samples drawn each time will have a high degree of overlap, which is not conducive to model convergence. Therefore, during training, the order of the samples in the training set must be randomly shuffled before each iteration to ensure that the mini-batch samples drawn each time are different, thus achieving the best model convergence effect.

[0108] The fault diagnosis method of this invention will be described below with a specific embodiment. The data of this invention was obtained from on-site measurements at a converter station in Jiangsu Province. The main circulation pump used in this converter station is model MKG200-150-400 / 410H1F2KE-SBQQE. The sampling frequency of the vibration sensor is 12.8K, and a total of 2500 sets of historical data were measured, with each set containing 8000 sampling points. Each fault state of the main circulation pump contains 300 samples, totaling 1500 vibration images forming the training set. 1000 samples were selected as the test set for subsequent model training and testing.

[0109] In the process of implementing mode decomposition in VMD, this invention selects the number of mode decompositions as k=4 and the penalty factor as 0.005.

[0110] Figure 4 This is a schematic diagram of the VMD decomposition of vibration signals in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 4 As shown, taking the vertical vibration signal under normal conditions as an example, Figure 4The intrinsic mode components time-domain images and signal decomposition results are shown. It can be seen that the VMD algorithm effectively suppresses mode aliasing and has a good signal decomposition effect.

[0111] Furthermore, this invention employs the Pearson correlation coefficient matrix to adaptively select the optimal modal components. Table 3 shows the cross-correlation matrix of each modal component in the vertical direction under bearing loosening conditions. It can be seen that the intrinsic modal component labeled 4 has the lowest cross-correlation coefficient, and thus the first three intrinsic modal components are selected for spatial reconstruction.

[0112] IMF component 1 IMF component 2 IMF component 3 IMF component 4 IMF component 1 1.0000 0.0411 0.0035 0.0017 IMF component 2 0.0411 1.0000 0.0081 0.0025 IMF component 3 0.0035 0.0081 1.0000 0.0079 IMF component 4 0.0017 0.0025 0.0079 1.0000

[0113] Table 3 Optimal Modal Component Selection Table

[0114] After selecting the first three components, a 200-order Hankel matrix is ​​constructed using the SVD algorithm, and the singular value threshold is set to 0.01 to select the SVD reconstructed components. Figure 5 This is a schematic diagram illustrating vibration signal denoising in a fault diagnosis method for the main circulating pump in a converter valve according to the present invention. Figure 5 As shown, taking the vertical vibration signal under normal conditions as an example, the high-frequency white noise in the signal has been filtered out, and the time-frequency domain characteristics of the vibration signal have been preserved.

[0115] Figure 6 This diagram illustrates the effect of the AlexNet iterative training process in the fault diagnosis method for the main circulation pump in a converter valve according to the present invention. Considering computational cost requirements, the present invention selects the first 500 points of each vibration signal group for fault detection. That is, the constructed training set contains a total of 500 sample images.

[0116] Taking the vibration waveform and vibration spectrum of the main circulation pump under normal conditions as an example, after the segmented aggregation approximation with a window size of window_size=2 and the image transformation by the Gram matrix, each group of vibration signals is converted into a vibration image with a pixel value of 250×250.

[0117] When inputting data, images are read in mini-batches of 32 images, and the image order is randomly shuffled before each iteration to ensure the fitting effect and convergence speed of the neural network. Based on Figure 6 As can be seen, with the increase of the number of iterations, the loss function value of the main circulation pump fault diagnosis model continues to decrease, and tends to stabilize at 0.1 in the 10th iteration (epoch). The accuracy of the model on the test set reaches 91%. Therefore, the model proposed in this invention can effectively complete the high-precision fault diagnosis of the main circulation pump of the valve cooling system.

[0118] A second aspect of this invention relates to a fault diagnosis system for the main circulating pump in a converter valve. The system implements the steps of a fault diagnosis method for the main circulating pump in a converter valve as described in the first imitation of this invention. The system includes an acquisition module, a fusion module, and an extraction module. The acquisition module includes an accelerometer and a denoising unit. The accelerometer acquires the triaxial vibration signals of the main circulating pump in the converter valve, and the denoising unit uses the VMD-SVD algorithm to denoise the triaxial vibration signals. The fusion module converts the denoised uniaxial vibration signals into two-dimensional grayscale images, and sequentially inputs the two-dimensional grayscale images into different color channels for superposition to obtain a feature fusion image. The extraction module inputs the feature fusion image into a convolutional neural network to extract fault features.

[0119] A third aspect of the present invention relates to a fault diagnosis device for the main circulating pump in a converter valve, the device comprising a processor and a storage medium; wherein the storage medium is used to store instructions; and the processor is used to perform operations according to the instructions to execute the steps of the method in the first aspect of the present invention.

[0120] It is understood that, in order to implement the various functions in the methods provided in the embodiments of this application, the fault diagnosis device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, in conjunction with the algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0121] This application embodiment can divide fault diagnosis into functional modules based on the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0122] The device includes at least one processor, a bus system, and at least one communication interface. The processor may be a central processing unit (CPU), or it may be replaced by a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), or other hardware, or the FPGA or other hardware may work together with the CPU as a processor.

[0123] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a bus. The memory can also be integrated with the processor.

[0124] The hard drive can be a mechanical hard drive or a solid-state drive (SSD), etc. The interface card can be a host bus adapter (HBA), a redundant array of independent disks (RID), an expander card, or a network interface controller (NIC), etc., and this embodiment of the invention is not limited to any particular type. The interface card in the hard drive module communicates with the hard drive. The storage node communicates with the interface card of the hard drive module to access the hard drive in the hard drive module.

[0125] The hard drive interface can be Serial Attached Small Computer System Interface (SAS), Serial Advanced Technology Attachment (SATA), or Peripheral Component Interconnect Express (PCIe), etc.

[0126] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks, SSDs).

[0127] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method of the first aspect of the present invention.

[0128] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0129] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0130] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0131] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0133] The beneficial effects of this invention are that, compared with the prior art, the fault diagnosis method, system, device, and medium for the main circulating pump in a converter valve of this invention can generate two-dimensional grayscale images by collecting the three-dimensional vibration signals of the converter valve, and comprehensively consider the weak vibration features in three different spatial dimensions, using a convolutional neural network to achieve effective and accurate fault feature extraction. The method of this invention is effective and reliable, with high algorithm efficiency, and can achieve reasonable source tracing of different fault types from fault features.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A fault diagnosis method for the main circulating pump in a converter valve, characterized in that, The method includes the following steps: Step 1: Collect the three-dimensional vibration signal of the main circulation pump of the converter valve, and use the VMD-SVD algorithm to denoise the three-dimensional vibration signal; Step 2: Convert the denoised unidirectional vibration signal into two-dimensional grayscale images respectively. Mark each two-dimensional grayscale image with a direction label and a time label. Then, input the two-dimensional grayscale images with corresponding time labels into different color channels and superimpose them to obtain a feature fusion image. Step 3: Input the feature fusion image into a convolutional neural network to extract fault features.

2. The fault diagnosis method for the main circulating pump in a converter valve according to claim 1, characterized in that: The triaxial vibration signal is based on the axial, vertical and longitudinal vibration components of the main circulating pump, collected by an accelerometer installed on the pump end bearing housing of the main circulating pump. The vibration frequency measurement range of the accelerometer is between 0.2 Hz and 10 kHz, and the sampling frequency is 12.8 kHz.

3. The fault diagnosis method for the main circulating pump in a converter valve according to claim 2, characterized in that: The VMD-SVD algorithm denoises the unidirectional vibration signal in each direction of the triaxial vibration signal.

4. The fault diagnosis method for the main circulating pump in a converter valve according to claim 3, characterized in that: The unidirectional vibration signal is converted into a two-dimensional grayscale image using the GAF algorithm.

5. A fault diagnosis method for the main circulating pump in a converter valve according to claim 4, characterized in that: The historical operating status of the main circulation pump is used to mark the status label of the two-dimensional grayscale image; The status labels are: main circulation pump normal, main circulation pump rotor unbalanced, coupling angle misalignment, coupling parallel misalignment, and bearing loose.

6. The fault diagnosis method for the main circulating pump in a converter valve according to claim 5, characterized in that: Two-dimensional grayscale images are superimposed and then input into the AlexNet network to extract fault features.

7. A fault diagnosis method for the main circulating pump in a converter valve according to claim 6, characterized in that: The fault features extracted by the AlexNet network are associated with the state labels to achieve fault classification.

8. A fault diagnosis method for the main circulating pump in a converter valve according to claim 7, characterized in that: In the training set, each extraction The unextracted feature fusion images are used as batch samples and input into the AlexNet network for training until the training of all samples in the training set is completed, thus confirming the completion of one iteration process. It is a natural number.

9. A fault diagnosis method for the main circulating pump in a converter valve according to claim 8, characterized in that: Before each iteration, the order of the samples in the training set is randomly rearranged.

10. A fault diagnosis system for the main circulating pump in a converter valve, characterized in that: The system is used to implement the steps of the fault diagnosis method for the main circulating pump in a converter valve as described in any one of claims 1-9, and; The system includes a data acquisition module, a fusion module, and an extraction module; wherein... The acquisition module includes an accelerometer and a noise reduction unit. The accelerometer is used to acquire the triaxial vibration signal of the main circulation pump of the converter valve, and the noise reduction unit uses the VMD-SVD algorithm to denoise the triaxial vibration signal. The fusion module is used to convert the denoised unidirectional vibration signal into a two-dimensional grayscale image, mark each two-dimensional grayscale image with a direction label and a time label, and then input the two-dimensional grayscale images with corresponding time labels into different color channels and superimpose them to obtain a feature fusion image. The extraction module is used to input the feature fusion image into a convolutional neural network to achieve fault feature extraction.

11. A fault diagnosis device for the main circulating pump in a converter valve, characterized in that: The device includes a processor and a storage medium; wherein... The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-9.

Images