Hydroelectric generating set runner crack fault diagnosis method and device based on slt-ca-2dcnn

Abstract

The application discloses a kind of water turbine set runner crack fault diagnosis method and device based on SLT-CA-2DCNN, belong to water turbine set runner crack fault diagnosis technical field.The method comprises: collecting the multichannel vibration signal of water turbine set;The multichannel vibration signal is preprocessed, and the multichannel vibration signal after preprocessing is converted into hyperwave time-frequency diagram;The hyperwave time-frequency diagram is input into the CA-2DCNN water turbine set runner crack fault diagnosis model trained to obtain the fault diagnosis result of water turbine set runner crack;CA-2DCNN water turbine set runner crack fault diagnosis model includes channel attention mechanism and two-dimensional convolutional neural network, channel attention mechanism is used to allocate dynamic weight for different channels, and two-dimensional convolutional neural network is used to capture spatial correlation features by two-dimensional convolution.The method can fully exploit the fault information in different position signals by multichannel parallel feature extraction and feature level fusion, improve the accuracy and stability of water turbine set runner crack fault diagnosis.

Description

Technical Field

[0001] This application belongs to the field of hydropower turbine runner crack fault diagnosis technology, and in particular relates to a hydropower turbine runner crack fault diagnosis method and device based on SLT-CA-2DCNN. Background Technology

[0002] With the growth of global energy demand, hydropower, as a clean and renewable energy source, occupies an important position in the energy structure. Hydropower units, as the core equipment of hydropower stations, are crucial for ensuring power supply and energy security through safe, stable, and efficient operation. However, hydropower units have complex structures and operate in harsh environments. Runner cracking is a common and serious type of failure. If not diagnosed promptly and accurately, it can lead to the escalation of the fault, causing serious safety accidents and resulting in huge economic losses.

[0003] Traditional fault diagnosis methods for hydropower units often rely on single-channel signals acquired by a single sensor, such as vibration sensors. While these methods can detect some fault phenomena, they only acquire information from limited dimensions, making it difficult to fully capture the complex, multi-dimensional features of turbine runner cracks. This results in insufficient accuracy and reliability in fault diagnosis, and is prone to problems such as delayed fault identification or misjudgment. For example, some existing technologies use models based on time-shifted multi-scale attention entropy and random forests to diagnose faults in hydropower units using single vibration signals, but their diagnostic performance drops significantly when faced with complex, multi-site turbine runner crack faults. Multi-channel signal fusion diagnostic technology can acquire signals from different locations using multiple sensors, achieving information complementarity and extracting fault features more comprehensively, thereby improving diagnostic accuracy. Currently, multi-channel fault diagnosis methods have been applied in fields such as bearings and motors, such as deep residual convolutional neural networks based on multi-channel information fusion and deep learning models that fuse multiple resolutions and multiple sensors. However, when these models are applied to the fault diagnosis of turbine runner cracks in hydropower units, they still suffer from large diagnostic errors and insufficient stability, making it difficult to meet practical diagnostic needs.

[0004] Existing methods for diagnosing hydropower turbine runner crack faults suffer from low diagnostic accuracy, poor stability, and difficulty in accurately identifying such faults due to the limited range of sensors, incomplete information acquisition, and complex fault characteristics. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method and device for diagnosing hydropower turbine runner cracks based on SLT-CA-2DCNN. By utilizing ultrawavelet transform, channel attention mechanism, and two-dimensional convolutional neural network to construct a CA-2DCNN hydropower turbine runner crack fault diagnosis model, fault features of multi-channel vibration signals are extracted and feature-level fusion is performed, thereby improving the accuracy and stability of the hydropower turbine runner crack fault diagnosis system.

[0006] To address the aforementioned problems, according to the first aspect of this application, a method for diagnosing hydroelectric turbine runner cracks based on SLT-CA-2DCNN is provided, the method comprising: Collect multi-channel vibration signals from hydropower units; The multi-channel vibration signal is preprocessed, and the preprocessed multi-channel vibration signal is converted into a wavelet time-frequency diagram based on the super wavelet transform. The ultrawavelet time-frequency graph is input into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of hydropower turbine runner crack. The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution. The training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

[0007] According to one embodiment of this application, the step of acquiring vibration signals under different guide vane openings and different degrees of impeller cracks to obtain multi-channel vibration signals includes: Vibration signals were collected under different degrees of wheel cracking under optimal and small opening conditions to obtain multi-channel vibration signals under 10 types of conditions. The cracks under the optimal opening and small opening conditions include: normal, 1 2cm rotary wheel crack, 2 2cm rotary wheel cracks, 3 2cm rotary wheel cracks, and 3 4cm rotary wheel cracks, respectively.

[0008] According to one embodiment of this application, the step of converting the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on wavelet transform includes: The complex-valued Morlet wavelet was selected as the mother wavelet, and the sampling frequency of the preprocessed multi-channel vibration signal was matched to obtain the mother wavelet parameters. Based on the mother wavelet parameters, the initial number of periods and the super-resolution order interval are set, and the super-wavelet set is constructed using a multiplicative growth mode. The response of the superwavelet set to the preprocessed multi-channel vibration signal is calculated. The complex convolution results of individual wavelets in the superwavelet set are fused by the complex convolution operator. The power recovery factor is used to compensate for the power halving effect to obtain the fused multiwavelet response. The geometric mean of the fused multi-wavelet responses is taken to generate the core time-frequency representation of the superwavelet transform. The core time-frequency representation is transformed into a super wavelet scale map, and the super wavelet scale map is scaled and normalized to obtain the super wavelet time-frequency map.

[0009] According to one embodiment of this application, the step of inputting the ultrawavelet time-frequency image into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis result of the hydropower turbine runner crack includes: The super wavelet time-frequency plot is input into the channel module group for feature extraction to obtain the first output feature, the second output feature and the third output feature; The first, second, and third output features are input into the feature fusion module, and the fusion of multi-channel features is achieved through feature concatenation and channel attention mechanisms to obtain fused features. The fused features are input into the flattening layer for dimensionality transformation to obtain a one-dimensional feature vector. One-dimensional feature vectors are input into a fully connected layer for classification and identification, resulting in fault diagnosis of hydropower turbine runner cracks.

[0010] According to one embodiment of this application, the step of inputting the ultrawavelet time-frequency image into the channel module group for feature extraction to obtain a first output feature, a second output feature, and a third output feature includes: The ultrawavelet time-frequency plot is input into the first channel module for feature extraction to obtain the first output feature; The ultrawavelet time-frequency plot is input into the second channel module for feature extraction to obtain the second output feature. The ultrawavelet time-frequency plot is input into the third channel module for feature extraction, resulting in the third output feature. The channel module group includes a first channel module, a second channel module, and a third channel module.

[0011] According to one embodiment of this application, the step of inputting the ultrawavelet time-frequency image into the first channel module for feature extraction to obtain the first output feature includes: The ultrawavelet time-frequency map is input into the first channel convolutional layer for feature extraction to obtain the first channel convolutional features; The first channel convolutional features are input into the first channel max pooling layer for dimensionality reduction to obtain the first channel pooling features; The pooling features of the first channel are input into the convolutional layer of the second channel for feature extraction, resulting in the convolutional features of the second channel. The second channel convolutional features are input into the second channel max pooling layer for dimensionality reduction to obtain the first output features; The parameters of the first channel convolutional layer are 16×3×3, the parameters of the second channel convolutional layer are 32×3×3, both the first and second channel convolutional layers are configured with batch normalization and ReLU activation functions, the parameters of the first channel max pooling layer are 3×3, and the parameters of the second channel max pooling layer are 2×2.

[0012] According to one embodiment of this application, the step of inputting the first output feature, the second output feature, and the third output feature into the feature fusion module, and fusing multi-channel features through feature concatenation and attention mechanisms to obtain fused features includes: The first output feature, the second output feature, and the third output feature are input into the feature concatenation layer for channel-dimensional concatenation to obtain the initial fused feature. The initial fused features are input into the first convolutional layer for feature extraction, resulting in the first convolutional features; The first convolutional features are input into the second convolutional layer for feature extraction, resulting in the second convolutional features; The second convolutional feature is input into the first max pooling layer for dimensionality reduction to obtain the first pooling feature. The first pooling feature is input into the first channel attention layer for adaptive adjustment of channel weights to obtain the first attention feature; The first attention feature is input into the third convolutional layer for feature extraction, resulting in the third convolutional feature; The third convolutional feature is input into the second max pooling layer for dimensionality reduction to obtain the second pooling feature. The second pooling feature is input into the second channel attention layer for adaptive adjustment of channel weights to obtain the fused feature. The parameters of the first convolutional layer are 64×3×3, the parameters of the second convolutional layer are 32×3×3, the parameters of the first max pooling layer are 2×2, the parameters of the third convolutional layer are 32×3×3, and the parameters of the second max pooling layer are 2×2.

[0013] According to one embodiment of this application, the training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

[0014] According to a second aspect of this application, a hydropower turbine runner crack fault diagnosis device based on SLT-CA-2DCNN is provided, the device comprising: The acquisition module is used to acquire multi-channel vibration signals from the hydropower unit; The processing module is used to preprocess the multi-channel vibration signal and convert the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on the wavelet transform. The diagnostic module is used to input the ultrawavelet time-frequency map into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of the hydropower turbine runner crack. The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution. The training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

[0015] According to a third aspect of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method as described in the first aspect above.

[0016] According to a fourth aspect of this application, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method as described in the first aspect above.

[0017] According to a fifth aspect of this application, a chip is provided, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method as described in the first aspect.

[0018] According to a sixth aspect of this application, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method as described in the first aspect above.

[0019] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application.

[0020] This application provides a method for diagnosing hydropower turbine runner cracks based on SLT-CA-2DCNN, which has the following advantages over existing technologies: (1) This application obtains multi-channel vibration signals by acquiring signals from multiple key locations of the hydropower unit, thereby achieving information complementarity and reducing the problem of incomplete information acquisition by a single sensor; by introducing wavelet transform to convert the vibration signal into a wavelet time-frequency diagram, it breaks through the resolution limitation of traditional time-frequency analysis, fully preserves the time-domain and frequency-domain fault characteristics of the signal, and provides a high-quality data foundation for more accurate diagnosis; by constructing a CA-2DCNN hydropower unit runner crack fault diagnosis model with a fusion channel attention mechanism, the attention mechanism is used to adaptively highlight key features and suppress redundant information, thereby achieving efficient fusion of multi-channel features and improving the model's feature discrimination ability and diagnostic accuracy.

[0021] (2) This application enables the model to fully mine fault information in signals at different locations through multi-channel parallel feature extraction and feature-level fusion. It has high diagnostic accuracy and good stability, and is suitable for actual hydropower unit runner crack fault diagnosis scenarios. It realizes high-precision and high-reliability diagnosis of hydropower unit runner crack faults, and ensures the safe and stable operation of hydropower units.

[0022] (3) This application uses ultrawavelet transform technology in the signal preprocessing stage to construct a multi-scale ultrawavelet set and perform geometric averaging fusion, transforming the one-dimensional vibration signal into a high-resolution time-frequency map. This breaks through the inherent limitation of the mutual constraint between time and frequency resolution in traditional wavelet transform, rather than having to make a trade-off between time resolution and frequency resolution. It can simultaneously optimize the time-frequency characterization capability, fully preserve the time-domain and frequency-domain fault characteristics in the vibration signal, and is extremely sensitive to the weak feature anomalies caused by early cracks in the turbine runner that are hidden in the background of strong noise. It can comprehensively capture the vibration response differences in different parts, and improve the early identification capability and all-condition adaptability of turbine runner crack fault diagnosis. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is one of the flowcharts illustrating the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method provided in this application embodiment; Figure 2 These are the wavelet time-frequency diagrams of various operating conditions at the guide bearing provided in the embodiments of this application; Figure 3 These are vibration waveform diagrams of normal operating conditions at the optimal opening degree at each position provided in the embodiments of this application; Figure 4 This is a schematic diagram of the channel attention mechanism provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a two-dimensional convolutional neural network provided in an embodiment of this application; Figure 6 This is a confusion matrix diagram of each multi-channel model provided in the embodiments of this application; Figure 7 These are feature visualization results of the various multi-channel models provided in the embodiments of this application; Figure 8 This is the second flowchart of the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN provided in the embodiments of this application; Figure 9 This is a schematic diagram of the structure of the hydropower unit runner crack fault diagnosis device based on SLT-CA-2DCNN provided in the embodiments of this application; Figure 10 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0025] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate to allow embodiments of this application to be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0026] The following description, in conjunction with the accompanying drawings, details the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method, SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis device, electronic equipment, and readable storage medium provided in this application, through specific embodiments and application scenarios.

[0027] Among them, the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN can be applied to the terminal, which can be executed by the hardware or software in the terminal.

[0028] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0029] The following embodiments describe a terminal including a display and a touch-sensitive surface. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, mouse, and joystick.

[0030] The hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN provided in this application embodiment can be executed by an electronic device or a functional module or entity in an electronic device that can implement the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The following uses an electronic device as the execution subject to illustrate the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN provided in this application embodiment.

[0031] Figure 1 This is one of the flowcharts illustrating the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method provided in this application embodiment, such as... Figure 1 As shown, the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN includes steps 110, 120 and 130.

[0032] Step 110: Collect multi-channel vibration signals from the hydropower unit; It is easy to understand that high-frequency vibration sensors are installed at key locations such as the guide bearing, volute, and tailrace pipe of the hydroelectric generator model. Data is collected using high-frequency vibration sensors with a sampling frequency of 20,000 Hz to obtain multi-channel vibration signals.

[0033] Step 120: Preprocess the multi-channel vibration signal and convert the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on the wavelet transform. In some embodiments, the step of converting the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on wavelet transform includes: The complex-valued Morlet wavelet was selected as the mother wavelet, and the sampling frequency of the preprocessed multi-channel vibration signal was matched to obtain the mother wavelet parameters. Based on the mother wavelet parameters, the initial number of periods and the super-resolution order interval are set, and the super-wavelet set is constructed using a multiplicative growth mode. The response of the superwavelet set to the preprocessed multi-channel vibration signal is calculated. The complex convolution results of individual wavelets in the superwavelet set are fused by the complex convolution operator. The power recovery factor is used to compensate for the power halving effect to obtain the fused multiwavelet response. The geometric mean of the fused multi-wavelet responses is taken to generate the core time-frequency representation of the superwavelet transform. The core time-frequency representation is transformed into a super wavelet scale map, and the super wavelet scale map is scaled and normalized to obtain the super wavelet time-frequency map.

[0034] It is easy to understand that time-frequency analysis is a commonly used signal processing method in fault diagnosis. However, traditional time-frequency analysis methods have resolution limitations, making it difficult to accurately characterize the nonlinear and non-stationary signal characteristics of turbine crack faults. SLT (Superlet Transform), as a novel time-frequency analysis method, can overcome the resolution limitations of traditional time-frequency analysis through flexible basis function design and adaptive mechanisms, achieving high-precision characterization of signal time-frequency features and providing new technical support for turbine crack fault feature extraction.

[0035] In some embodiments, the preprocessed multi-channel vibration signal is converted into a wavelet time-frequency diagram. The specific steps are as follows: The complex-valued Morlet wavelet was selected as the mother wavelet, matched with a signal sampling frequency of 20000Hz, and the frequency analysis range was set to 10~10000Hz to generate 100 frequency points. Construct an ultrawavelet set, set the initial number of periods c1=3, the super-resolution order interval is (1,5), and adopt the multiplicative growth mode; The response of the superwavelet set to the vibration signal is calculated. The complex convolution results of a single wavelet are fused by the complex convolution operator. The power recovery factor is used to compensate for the halving effect of the complex wavelet on the power of the real signal. The geometric mean of multiple wavelet responses is taken to generate the core time-frequency representation of the super wavelet transform, achieving synchronous optimization of time resolution and frequency resolution; A wavelet scaling map is generated to characterize the energy intensity of the time-frequency distribution. The transformed amplitude spectrum is then normalized to a size of 100×100. Finally, the data is classified and saved as a wavelet time-frequency map according to the wheel crack condition label. Figure 2 These are the wavelet time-frequency diagrams of various operating conditions at the guide bearing provided in the embodiments of this application, such as... Figure 2 As shown, this ensures that the time-frequency diagram clearly reflects the time-domain and frequency-domain distribution of the fault characteristics.

[0036] Step 130: Input the super wavelet time-frequency graph into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of the hydropower turbine runner crack. In some embodiments, the training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

[0037] In some embodiments, the step of acquiring vibration signals under different guide vane openings and different degrees of impeller cracks to obtain multi-channel vibration signals includes: Vibration signals were collected under different degrees of wheel cracking under optimal and small opening conditions to obtain multi-channel vibration signals under 10 types of conditions. The cracks under the optimal opening and small opening conditions include: normal, 1 2cm rotary wheel crack, 2 2cm rotary wheel cracks, 3 2cm rotary wheel cracks, and 3 4cm rotary wheel cracks, respectively.

[0038] In some embodiments, firstly, high-frequency vibration sensors are installed at key locations such as the guide bearing, volute, and tailrace pipe of the hydroelectric generator model test bench. Data is acquired using high-frequency vibration sensors with a sampling frequency of 20000Hz, collecting vibration signals under different guide vane openings and different degrees of impeller cracking. This experiment designed two optimal guide vane opening conditions and a small opening condition. Figure 3 These are vibration waveforms under normal operating conditions at the optimal opening degree at various positions, as provided in the embodiments of this application. Figure 3 As shown, each working condition includes five types of wheel crack states: normal, one 2cm wheel crack, two 2cm wheel cracks, three 2cm wheel cracks, and three 4cm wheel cracks, for a total of 10 working conditions.

[0039] The experimental method is as follows: The test bench is adjusted to the specified guide vane opening and impeller crack degree working condition point. After stable operation, the vibration signal of the hydropower unit is collected. The vibration signal of the hydropower unit under different guide vane opening and different impeller crack degree is collected respectively. 50 samples are collected for each working condition, and the length of each sample is 10,000 data points to ensure that the sample can fully reflect the fault characteristics under the working condition.

[0040] Furthermore, the preprocessed signal is converted into a wavelet time-frequency graph and divided into training and testing sets to prepare for model training and validation.

[0041] For example, the acquired multi-channel vibration signals are divided into samples, with each working condition divided into 50 samples, and each sample having a length of 10,000 data points. The samples from 10 operating conditions were divided into training and testing sets in a 4:1 ratio. The training set contained 400 samples, and the testing set contained 100 samples, forming a total dataset of 500 samples and 10 categories of rotor crack faults. The classification labels were represented by 0-9, and the specific correspondences are as follows: 0 for optimal guide vane opening and normal operating condition; 1 for one 2cm rotor crack at optimal opening; 2 for two 2cm rotor cracks at optimal opening; 3 for three 2cm rotor cracks at optimal opening; 4 for three 4cm rotor cracks at optimal opening; 5 for small opening and normal operating condition; 6 for one 2cm rotor crack at small opening; 7 for two 2cm rotor cracks at small opening; 8 for three 2cm rotor cracks at small opening; and 9 for three 4cm rotor cracks at small opening.

[0042] Table 1 shows an example dataset provided in this application embodiment. The collected multi-channel vibration signals were processed, and 50 samples from each working condition were divided into a training set and a test set in a 4:1 ratio, i.e., 40 training samples and 10 test samples per working condition. This resulted in a wheel crack fault dataset containing 500 samples across 10 working conditions, with 400 samples in the training set and 100 samples in the test set. The 10 working conditions were labeled using a range of 0-9, and the specific label correspondence is shown in Table 1 below. Table 1 The multi-channel ultrawavelet time-frequency images in the training set are input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract and fuse runner crack fault features, and perform classification and identification. Through continuous feedback and adjustment, the model achieves the optimal effect.

[0043] The multi-channel ultrawavelet time-frequency graphs in the test set are input into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model, ultimately achieving accurate diagnosis of hydropower turbine runner crack faults.

[0044] In some embodiments, the network structure of the CA-2DCNN hydropower turbine runner crack fault diagnosis model is shown in Table 2 below. The model consists of three parallel channel modules, one feature fusion module, a flattening layer, and a fully connected layer. Each channel module extracts features from the wavelet time-frequency map at different locations. The feature fusion module achieves efficient fusion of multi-channel features through feature splicing and attention mechanisms, and finally outputs the classification result through the fully connected layer.

[0045] This experiment used Python 3.8.19 and PyTorch 2.3.1 to build the model, and utilized GPU acceleration for training on a hardware platform equipped with an AMD Ryzen 9 7945HX processor and an NVIDIA GeForce RTX 4060 Laptop GPU. The initial learning rate was set to 0.0005 and automatically adjusted based on the training progress using a learning rate scheduler. The training iterations were 100, and the experiment was repeated 10 times. The average value was used as the fault diagnosis accuracy, as shown in Table 2 below. Table 2 The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution.

[0046] In some embodiments, the step of inputting the ultrawavelet time-frequency image into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis result of the hydropower turbine runner crack includes: The super wavelet time-frequency plot is input into the channel module group for feature extraction to obtain the first output feature, the second output feature and the third output feature; In some embodiments, the step of inputting the ultrawavelet time-frequency graph into the channel module group for feature extraction to obtain a first output feature, a second output feature, and a third output feature includes: The ultrawavelet time-frequency plot is input into the first channel module for feature extraction to obtain the first output feature; In some embodiments, the step of inputting the ultrawavelet time-frequency image into the first channel module for feature extraction to obtain the first output feature includes: The ultrawavelet time-frequency map is input into the first channel convolutional layer for feature extraction to obtain the first channel convolutional features; The first channel convolutional feature is input into the first max pooling layer for dimensionality reduction to obtain the first channel pooling feature. The pooling features of the first channel are input into the convolutional layer of the second channel for feature extraction, resulting in the convolutional features of the second channel. The second channel convolutional features are input into the second max pooling layer for dimensionality reduction to obtain the first output features.

[0047] The ultrawavelet time-frequency plot is input into the second channel module for feature extraction to obtain the second output feature. The ultrawavelet time-frequency plot is input into the third channel module for feature extraction, resulting in the third output feature. The channel module group includes a first channel module, a second channel module, and a third channel module.

[0048] It is easy to understand that the first, second, and third channel modules have the same structure. Each channel module contains two convolutional layers and two max-pooling layers. The convolutional layer parameters are 16×3×3 and 32×3×3, respectively. They are all configured with batch normalized (BN) and ReLU activation functions. The max-pooling layer kernel sizes are 3×3 and 2×2, respectively. Each channel module receives a 3×100×100 ultrawavelet time-frequency map as input and outputs a 32×16×16 feature map after processing.

[0049] The first, second, and third output features are input into the feature fusion module, and the fusion of multi-channel features is achieved through feature concatenation and channel attention mechanisms to obtain fused features. In some embodiments, the input of the first output feature, the second output feature, and the third output feature into the feature fusion module, and the fusion of multi-channel features through feature concatenation and attention mechanisms to obtain fused features, includes: The first output feature, the second output feature, and the third output feature are input into the feature concatenation layer for channel-dimensional concatenation to obtain the initial fused feature. The initial fused features are input into the first convolutional layer for feature extraction, resulting in the first convolutional features; The first convolutional features are input into the second convolutional layer for feature extraction, resulting in the second convolutional features; The second convolutional feature is input into the first max pooling layer for dimensionality reduction to obtain the first pooling feature. The first pooling feature is input into the first channel attention layer for adaptive adjustment of channel weights to obtain the first attention feature; The first attention feature is input into the third convolutional layer for feature extraction, resulting in the third convolutional feature; The third convolutional feature is input into the second max pooling layer for dimensionality reduction to obtain the second pooling feature. The second pooling feature is input into the second channel attention layer for adaptive adjustment of channel weights to obtain the fused feature.

[0050] Attention mechanisms can adaptively highlight key features and suppress redundant information. When combined with convolutional neural networks, they can further enhance the feature extraction capabilities of the model. Figure 4This is a schematic diagram of the channel attention mechanism provided in this application embodiment. The CA (Channel Attention) mechanism is a feature adaptive enhancement technique used to improve the performance of convolutional neural networks. Its core idea is to model the importance of each channel in the feature map and assign dynamic weights to different channels to highlight key information channels. The specific implementation process is as follows: First, global max pooling and global average pooling are simultaneously performed on the feature map with an input size of (W, H, C) to obtain two feature vectors of size (1, 1, C). Then, these two vectors are fed into a shared fully connected layer and added together, and activated by the Sigmoid function to generate channel attention weights of size (1, 1, C). Finally, these weights are multiplied element-wise with the original feature map to enhance the features of important channels, suppress unimportant channels, and improve the model's ability to capture useful features.

[0051] Figure 5 This is a schematic diagram of the structure of the two-dimensional convolutional neural network provided in this application embodiment. 2DCNN (Two-Dimensional Convolutional Neural Network) is a deep learning model for two-dimensional time-frequency image data. Its core consists of convolutional layers, activation functions, batch normalization, pooling layers, flattening layers, fully connected layers, and an output layer. It captures spatial correlation features through two-dimensional convolutional kernels, introduces non-linearity through ReLU activation to overcome the limitations of linear expression, normalizes the feature distribution through BN layers to accelerate training convergence, reduces dimensionality and enhances robustness through pooling layers, and flattens the features after multiple rounds of layer processing before inputting them into the fully connected layer, finally outputting the classification result. This network can automatically extract deep features from the bottom to the top layers without requiring manual feature design, and it exhibits fast training convergence and strong stability, making it suitable for fault diagnosis.

[0052] It's easy to understand that the feature maps output from the three channel modules are first stitched together into a 96×16×16 fused feature map through a feature stitching layer. Then, it sequentially passes through a convolutional layer with a kernel of 64×3×3, a convolutional layer with a kernel of 32×3×3, a 2×2 max pooling layer, the first CA attention layer, another convolutional layer with a kernel of 32×3×3, a 2×2 max pooling layer, and the second CA attention layer. Each convolutional layer is followed by batch normalization (BN) and ReLU activation functions, and finally, a 32×4×4 fused feature map is output.

[0053] The fused features are input into the flattening layer for dimensionality transformation to obtain a one-dimensional feature vector. One-dimensional feature vectors are input into a fully connected layer for classification and identification, resulting in fault diagnosis of hydropower turbine runner cracks.

[0054] For example, the flattening layer transforms the 32×4×4 fused feature map into a 512-dimensional feature vector, and the fully connected layer maps this feature vector into a 10-dimensional classification probability vector, corresponding to 10 types of wheel crack fault conditions.

[0055] In some embodiments, to verify the diagnostic performance of the proposed model, single-channel model comparison experiments and multi-channel model comparison experiments were designed. The single-channel comparison models included CNN, GoogleNet, and MK-resCNN. The proposed single-channel model is the SLT-CA-2DCNN model, which integrates a wavelet processing stage based on the CA-2DCNN hydropower turbine runner crack fault diagnosis model of this application, retaining the structure of a single-channel input. The multi-channel comparison models included MCNN, DRCNN, and MRSFN. The proposed model is the SLT-CA-2DCNN model, which integrates a wavelet processing stage based on the CA-2DCNN hydropower turbine runner crack fault diagnosis model of this application.

[0056] Figure 6 This is a confusion matrix diagram of each multi-channel model provided in the embodiments of this application. Figure 7 These are feature visualization results of various multi-channel models provided in the embodiments of this application, such as... Figure 6 and Figure 7 As shown, experimental results indicate that the accuracies of the single-channel contrast models CNN, GoogleNet, MK-resCNN, and the proposed single-channel model are 81.90%, 84.67%, 86.07%, and 90.77%, respectively, with the proposed single-channel model achieving the highest accuracy. For multi-channel contrast models MCNN, DRCNN, MRSFN, and the proposed model, the accuracies are 84.50%, 93.00%, 97.80%, and 98.90%, respectively, with the proposed model achieving the highest accuracy and the smallest error. Confusion matrix and T-SNE feature visualization results show that the proposed model has very few misclassifications, and the extracted fault features exhibit compact clustering and strong discriminative power, demonstrating excellent fault classification and recognition performance.

[0057] The results show that under different experimental conditions, the proposed model maintains high diagnostic accuracy and good stability, effectively suppresses interference factors, and extracts runner crack fault features more accurately, providing stronger support for early warning and accurate diagnosis of hydropower unit runner crack faults. This method can effectively distinguish fault types under different crack degrees and operating conditions. Compared with other model methods, it has extremely high diagnostic accuracy and good stability, and can more effectively solve the problem of difficulty in accurately identifying hydropower unit runner crack faults due to single sensors, incomplete information collection, and complex fault features. It has important practical reference value for the actual application of hydropower unit runner crack fault diagnosis.

[0058] Figure 8This is the second flowchart illustrating the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method provided in this application embodiment. Figure 8 As shown, firstly, a simulated test of turbine runner cracking fault was conducted on a hydropower unit model test bench, and multi-channel vibration signals were collected. After preprocessing the signals, a dataset of hydropower unit turbine runner cracking faults was constructed. Then, the vibration signals were converted into a wavelet time-frequency map through wavelet transform, fully preserving the time-domain and frequency-domain fault characteristics of the signals. Finally, a CA-2DCNN hydropower unit turbine runner cracking fault diagnosis model was constructed by combining channel attention mechanism and two-dimensional convolutional neural network. The model was then used to train and test hydropower unit turbine runner cracking fault samples to obtain fault diagnosis results.

[0059] The specific process of this method includes the following steps: Step 1: Build a hydroelectric generator model test bench and install high-frequency vibration sensors at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of impeller cracks.

[0060] Step 2: Preprocess the collected signals, divide the samples and construct a dataset, dividing it into training set and test set according to the proportion.

[0061] Step 3: Convert the vibration signals in the training and test sets into wavelet time-frequency graphs to provide data support for model training and testing.

[0062] Step 4: Input the multi-channel ultrawavelet time-frequency map from the training set into the CA-2DCNN hydropower turbine runner crack fault diagnosis model, extract and fuse the runner crack fault features, and perform classification and identification. Adjust the model parameters through continuous feedback to achieve the optimal model effect.

[0063] Step 5: Input the multi-channel ultrawavelet time-frequency graphs from the test set into the trained model to obtain the fault diagnosis results and achieve accurate diagnosis of hydropower turbine runner crack faults.

[0064] The hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN provided in this application embodiment can be executed by a hydropower turbine runner crack fault diagnosis device based on SLT-CA-2DCNN. This application embodiment uses the example of a hydropower turbine runner crack fault diagnosis device based on SLT-CA-2DCNN executing the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN to illustrate the hydropower turbine runner crack fault diagnosis device based on SLT-CA-2DCNN provided in this application embodiment.

[0065] This application also provides a hydroelectric turbine runner crack fault diagnosis device based on SLT-CA-2DCNN, such as... Figure 9As shown, the hydropower turbine runner crack fault diagnosis device based on SLT-CA-2DCNN includes: an acquisition module 910, a processing module 920, and a diagnosis module 930.

[0066] The acquisition module 910 is used to acquire multi-channel vibration signals from hydropower units. Processing module 920 is used to preprocess multi-channel vibration signals and convert the preprocessed multi-channel vibration signals into a wavelet time-frequency diagram based on wavelet transform. The diagnostic module 930 is used to input the ultrawavelet time-frequency map into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis result of the hydropower turbine runner crack. The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution. The training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

[0067] The hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN provided in this application obtains multi-channel vibration signals by acquiring signals from multiple key locations of the hydropower turbine, achieving information complementarity and reducing the problem of incomplete information acquisition from a single sensor. By introducing ultrawavelet transform to convert the vibration signals into ultrawavelet time-frequency maps, it breaks through the resolution limitations of traditional time-frequency analysis, fully preserves the time-domain and frequency-domain fault characteristics of the signals, and provides a high-quality data foundation for more accurate diagnosis. By constructing a CA-2DCNN hydropower turbine runner crack fault diagnosis model with a fusion channel attention mechanism, the attention mechanism adaptively highlights key features and suppresses redundant information, achieving efficient fusion of multi-channel features and improving the model's feature discrimination ability and diagnostic accuracy.

[0068] The hydro-generator turbine runner crack fault diagnosis device based on SLT-CA-2DCNN provided in this application embodiment can achieve… Figure 1-8 The various processes implemented in the embodiment of the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN will not be described in detail here to avoid repetition.

[0069] In some embodiments, such as Figure 10 As shown, this application embodiment also provides an electronic device 1000, including a processor 1001, a memory 1002, and a computer program stored in the memory 1002 and executable on the processor 1001. When the program is executed by the processor 1001, it implements the various processes of the above-described embodiment of the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0070] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0071] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described embodiment of the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0072] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0073] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for diagnosing hydropower turbine runner cracks based on SLT-CA-2DCNN.

[0074] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0075] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described embodiment of the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0076] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a device-level chip, device chip, chip device, or on-chip device chip, etc.

[0077] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0078] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the SLT-CA-2DCNN-based hydropower turbine runner crack fault diagnosis method of various embodiments of this application.

[0079] In the description of this application, "first feature" and "second feature" may include one or more of the features.

[0080] In the description of this application, "multiple" means two or more.

[0081] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0082] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0083] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for diagnosing hydropower unit runner crack faults based on SLT-CA-2DCNN, characterized in that, The method includes: Collect multi-channel vibration signals from hydropower units; The multi-channel vibration signal is preprocessed, and the preprocessed multi-channel vibration signal is converted into a wavelet time-frequency diagram based on the super wavelet transform. The ultrawavelet time-frequency graph is input into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of hydropower turbine runner crack. The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution. The training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

2. The method for diagnosing hydropower unit runner cracks based on SLT-CA-2DCNN according to claim 1, characterized in that, The vibration signals were collected under different guide vane openings and different degrees of impeller cracking to obtain multi-channel vibration signals, including: Vibration signals were collected under different degrees of wheel cracking under optimal and small opening conditions to obtain multi-channel vibration signals under 10 types of conditions. The cracks under the optimal opening and small opening conditions include: normal, 1 2cm rotary wheel crack, 2 2cm rotary wheel cracks, 3 2cm rotary wheel cracks, and 3 4cm rotary wheel cracks, respectively.

3. The method for diagnosing hydropower unit runner cracks based on SLT-CA-2DCNN according to claim 1, characterized in that, The process of converting the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on wavelet transform includes: The complex-valued Morlet wavelet was selected as the mother wavelet, and the sampling frequency of the preprocessed multi-channel vibration signal was matched to obtain the mother wavelet parameters. Based on the mother wavelet parameters, the initial number of periods and the super-resolution order interval are set, and the super-wavelet set is constructed using a multiplicative growth mode. The response of the superwavelet set to the preprocessed multi-channel vibration signal is calculated. The complex convolution results of individual wavelets in the superwavelet set are fused by the complex convolution operator. The power recovery factor is used to compensate for the power halving effect to obtain the fused multiwavelet response. The geometric mean of the fused multi-wavelet responses is taken to generate the core time-frequency representation of the superwavelet transform. The core time-frequency representation is transformed into a super wavelet scale map, and the super wavelet scale map is scaled and normalized to obtain the super wavelet time-frequency map.

4. The method for diagnosing hydropower unit runner cracks based on SLT-CA-2DCNN according to claim 1, characterized in that, The step of inputting the ultrawavelet time-frequency image into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of the hydropower turbine runner crack includes: The super wavelet time-frequency plot is input into the channel module group for feature extraction to obtain the first output feature, the second output feature and the third output feature; The first, second, and third output features are input into the feature fusion module, and the fusion of multi-channel features is achieved through feature concatenation and channel attention mechanisms to obtain fused features. The fused features are input into the flattening layer for dimensionality transformation to obtain a one-dimensional feature vector. One-dimensional feature vectors are input into a fully connected layer for classification and identification, resulting in fault diagnosis of hydropower turbine runner cracks.

5. The method for diagnosing hydropower unit runner cracks based on SLT-CA-2DCNN according to claim 4, characterized in that, The step of inputting the ultrawavelet time-frequency graph into the channel module group for feature extraction to obtain the first output feature, the second output feature, and the third output feature includes: The ultrawavelet time-frequency plot is input into the first channel module for feature extraction to obtain the first output feature; The ultrawavelet time-frequency plot is input into the second channel module for feature extraction to obtain the second output feature. The ultrawavelet time-frequency plot is input into the third channel module for feature extraction, resulting in the third output feature. The channel module group includes a first channel module, a second channel module, and a third channel module.

6. The method for diagnosing hydropower unit runner cracks based on SLT-CA-2DCNN according to claim 5, characterized in that, The step of inputting the ultrawavelet time-frequency graph into the first channel module for feature extraction to obtain the first output feature includes: The ultrawavelet time-frequency map is input into the first channel convolutional layer for feature extraction to obtain the first channel convolutional features; The first channel convolutional features are input into the first channel max pooling layer for dimensionality reduction to obtain the first channel pooling features; The pooling features of the first channel are input into the convolutional layer of the second channel for feature extraction, resulting in the convolutional features of the second channel. The second channel convolutional features are input into the second channel max pooling layer for dimensionality reduction to obtain the first output features; The parameters of the first channel convolutional layer are 16×3×3, the parameters of the second channel convolutional layer are 32×3×3, both the first and second channel convolutional layers are configured with batch normalization and ReLU activation functions, the parameters of the first channel max pooling layer are 3×3, and the parameters of the second channel max pooling layer are 2×2.

7. The hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN according to claim 4, characterized in that, The first, second, and third output features are input into the feature fusion module, and the multi-channel features are fused through feature concatenation and attention mechanisms to obtain fused features, including: The first output feature, the second output feature, and the third output feature are input into the feature concatenation layer for channel-dimensional concatenation to obtain the initial fused feature. The initial fused features are input into the first convolutional layer for feature extraction, resulting in the first convolutional features; The first convolutional features are input into the second convolutional layer for feature extraction, resulting in the second convolutional features; The second convolutional feature is input into the first max pooling layer for dimensionality reduction to obtain the first pooling feature. The first pooling feature is input into the first channel attention layer for adaptive adjustment of channel weights to obtain the first attention feature; The first attention feature is input into the third convolutional layer for feature extraction, resulting in the third convolutional feature; The third convolutional feature is input into the second max pooling layer for dimensionality reduction to obtain the second pooling feature. The second pooling feature is input into the second channel attention layer for adaptive adjustment of channel weights to obtain the fused feature. The parameters of the first convolutional layer are 64×3×3, the parameters of the second convolutional layer are 32×3×3, the parameters of the first max pooling layer are 2×2, the parameters of the third convolutional layer are 32×3×3, and the parameters of the second max pooling layer are 2×2.

8. A hydropower unit runner crack fault diagnosis device based on SLT-CA-2DCNN, implemented using the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN as described in any one of claims 1 to 7, characterized in that, The device includes: The acquisition module is used to acquire multi-channel vibration signals from the hydropower unit; The processing module is used to preprocess the multi-channel vibration signal and convert the preprocessed multi-channel vibration signal into a wavelet time-frequency diagram based on the wavelet transform. The diagnostic module is used to input the ultrawavelet time-frequency map into the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model to obtain the fault diagnosis results of the hydropower turbine runner crack. The CA-2DCNN hydropower turbine runner crack fault diagnosis model includes a channel attention mechanism and a two-dimensional convolutional neural network. The channel attention mechanism is used to assign dynamic weights to different channels, and the two-dimensional convolutional neural network is used to capture spatial correlation features through two-dimensional convolution. The training process of the CA-2DCNN hydropower turbine runner crack fault diagnosis model includes: Construct a pre-defined CA-2DCNN model for diagnosing hydropower unit runner cracks; A test bench for a hydroelectric generator was built, and high-frequency vibration sensors were installed at the guide bearing, volute, and tailrace pipe to collect vibration signals under different guide vane openings and different degrees of runner cracks, thus obtaining multi-channel vibration signals. The multi-channel vibration signals are preprocessed and converted into ultrawavelet time-frequency maps to obtain the training dataset. The training dataset is input into the preset CA-2DCNN hydropower turbine runner crack fault diagnosis model to extract, fuse and classify the runner crack fault features. The model parameters are adjusted through continuous feedback to achieve the optimal effect and obtain the trained CA-2DCNN hydropower turbine runner crack fault diagnosis model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the hydropower unit runner crack fault diagnosis method based on SLT-CA-2DCNN as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the hydropower turbine runner crack fault diagnosis method based on SLT-CA-2DCNN as described in any one of claims 1 to 7.

Images