Turbine fault diagnosis method and related device

Abstract

The invention discloses a turbine fault diagnosis method and related device, and the method comprises the steps: collecting an original vibration signal of a turbine, carrying out the coding processing of the original vibration signal of the turbine through a denoising self-encoder, and achieving the signal dimension reduction compression through the hidden layer dimension constraint, and obtaining a coding signal; inputting the coded signals into a convolutional neural network, extracting local features through a convolutional layer, and combining the local features into a feature map with a higher depth dimension after the local features are screened by a pooling layer; and inputting the feature map with a higher depth dimension into a full connection layer, finely adjusting network parameters through a back propagation algorithm in a supervised learning mode, and outputting a classification result of turbine fault types and positions. According to the method, huge operation parameters of the turboset are trained through deep learning, and automatic output of fault feature type positions and intelligent diagnosis of operation health conditions are completed without depending on an existing expert diagnosis system.

Description

Technical Field

[0001] This invention belongs to the field of steam turbine fault diagnosis technology, and specifically relates to a steam turbine fault diagnosis method and related apparatus. Background Technology

[0002] With economic development and increasing electricity demand, the stable and safe operation of the power industry has become a key factor for stable social development. Steam turbines, as crucial energy conversion equipment in various forms of power generation such as thermal power, nuclear power, biomass power, and gas-fired power plants, play an indispensable role in power plants. Research on their fault diagnosis and health status assessment is of great significance, and extensive discussions and research have been conducted both domestically and internationally in this area. Under the dual-carbon goals, the National Energy Administration has issued relevant documents on the upgrading and transformation of thermal power units, making numerous requirements for the flexibility transformation of thermal power plants. Steam turbine fault diagnosis and operational status assessment can fully ensure the flexible operation of thermal power units. Currently, on-site fault diagnosis relies heavily on expert diagnosis, which depends entirely on experience, resulting in low diagnostic accuracy. Summary of the Invention

[0003] The purpose of this invention is to provide a method and related apparatus for diagnosing turbine faults in order to solve the above-mentioned problems.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for diagnosing turbine faults, comprising: The original vibration signal of the steam turbine is collected, and the original vibration signal of the steam turbine is encoded using a noise-reducing autoencoder. The signal is then reduced in dimension and compressed to obtain the encoded signal by the hidden layer dimension constraint. The encoded signal is input into a convolutional neural network, local features are extracted through convolutional layers, and then combined into a feature map with a higher depth dimension after being filtered by pooling layers. By inputting feature maps with higher depth dimensions into a fully connected layer, and fine-tuning the network parameters through backpropagation algorithm in supervised learning mode, the classification results of turbine fault type and location are output.

[0005] Furthermore, the process of acquiring the original vibration signal of the steam turbine and encoding it using a noise-reducing self-encoder includes: A denoising autoencoder is used as an unsupervised algorithm in the pre-training stage, and a backpropagation algorithm is used as a supervised algorithm in the fine-tuning stage. The autoencoder network takes the original signal as the input layer, reconstructs the signal, and outputs a high-level language expression. The encoder transforms the acquired non-stationary and unstable mechanical signal x into an encoded signal y, and the decoder converts the encoded y into an output signal. ,Right now:

[0006] Furthermore, the encoding signal obtained by signal dimensionality reduction and compression through hidden layer dimension constraint includes: The autoencoder reduces the dimension of the input signal in the hidden layer and decompresses it in the output layer, as shown in the following formula:

[0007] where L represents the loss function, which is the quadratic error or cross-entropy error depending on the different forms of the data; When w' = wT, a constraint function is added to the hidden layer to obtain a more accurate output signal.

[0008] Furthermore, the data dimension of the constraint function satisfies: n > p, for the hidden layer, the dimension of the constraint function is less than the dimension of the input data. The change from x' to y is a dimensionality reduction operation. Through this dimensionality reduction operation, irrelevant variables are eliminated and the main features of the fault signal are retained. When the relationship between variables in each two layers is a linear mapping relationship and the error of supervised training is a quadratic error, this network is equivalent to PCA; n < p, that is, the dimension of the hidden layer is greater than the dimension of the input data, and the expression sparsity of y is proposed. At this time, the encoder is a "sparse autoencoder".

[0009] Furthermore, inputting the encoding signal into a convolutional neural network, extracting local features through the convolutional layer, and combining them into a feature map with a higher depth dimension after screening by the pooling layer includes: The convolutional neural network consists of an input layer, an output layer, and multiple hidden layers. The convolution and pooling methods are used to reduce the number of input parameters of the fully connected layer, extract data features through the convolution step, and combine them into data with a deeper depth and higher dimension. Finally, output prediction classification is performed through the fully connected layer; the pooling layer is used to pool the output of the convolutional layer, screen and filter the feature values obtained by the convolutional layer, screen the useful information in the layer, reduce the model size, and output it to the next layer for analysis; the maximum pooling method is adopted.

[0010] Furthermore, inputting the feature map with a higher depth dimension into the fully connected layer, and fine-tuning the network parameters through the backpropagation algorithm in the supervised learning mode to output the classification results of the steam turbine fault type and location includes: This neural network uses the steam turbine rotor vibration signal as the input, performs operations such as downsampling and multi-layer convolution on the original mechanical signal for feature extraction; uses the extracted feature parameters as the input signal of the fully connected layer.

[0011] In the second aspect, the present invention provides a steam turbine fault diagnosis system, including: The data acquisition module is used to acquire the original vibration signal of the steam turbine, encode the original vibration signal of the steam turbine using a noise-reducing autoencoder, and obtain the encoded signal by dimensionality reduction and compression through hidden layer dimensionality constraints. The feature map acquisition module is used to input the encoded signal into the convolutional neural network, extract local features through the convolutional layer, and combine them into a feature map with a higher depth dimension after being filtered by the pooling layer. The classification output module is used to input feature maps with higher depth dimensions into the fully connected layer, fine-tune the network parameters through the backpropagation algorithm in supervised learning mode, and output the classification results of turbine fault type and location.

[0012] Furthermore, the step of inputting the encoded signal into a convolutional neural network, extracting local features through convolutional layers, and combining them into a feature map with a higher depth dimension after filtering by pooling layers includes: Convolutional neural networks (CNNs) consist of input and output layers, as well as various hidden layers. Convolution and pooling reduce the number of input parameters in fully connected layers. Convolution steps extract data features and combine them into deeper, higher-dimensional data. Finally, fully connected layers perform output prediction and classification. Pooling layers pool the output of convolutional layers, filtering the feature values ​​obtained from the convolutional layers, selecting useful information from the layers, reducing the model size, and outputting it to the next layer for analysis. Max pooling is used. The process of inputting a feature map with a higher depth dimension into a fully connected layer, fine-tuning the network parameters through backpropagation algorithm in supervised learning mode, and outputting the classification results of turbine fault type and location includes: the neural network uses the turbine rotor vibration signal as input, and performs feature extraction on the original mechanical signal through operations such as downsampling and multi-layer convolution; the extracted feature parameters are used as the input signal of the fully connected layer.

[0013] Thirdly, the present invention provides a computer device, 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 steps of the steam turbine fault diagnosis method.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the turbine fault diagnosis method.

[0015] Compared with the prior art, the present invention has the following technical effects: This invention trains a deep learning method on the vast amount of operating parameters of a steam turbine unit, achieving automatic output of fault feature types and locations, and intelligent diagnosis of operational health status without relying on existing expert diagnostic systems. First, an autoencoder network is used to reconstruct the original signal output. Then, a convolutional neural network is used, employing convolution and pooling methods to reduce the number of input parameters to the fully connected layer. Data features are extracted through the convolution step and combined into deeper, higher-dimensional data. Finally, the fully connected layer performs output prediction and classification. Engineering applications demonstrate that this method achieves adaptive extraction of fault features and accurate identification of health status for different fault locations and types in steam turbine systems under various operating conditions, with a significantly higher accuracy than currently widely used expert diagnostic systems. This method can accurately identify common steam turbine fault types and pinpoint fault locations. Finally, verification shows that this method has high robustness. Attached Figure Description

[0016] Figure 1 Self-encoding principle diagram.

[0017] Figure 2 A schematic diagram of a fully connected neural network with a single hidden layer.

[0018] Figure 3 A schematic diagram of a convolutional neural network.

[0019] Figure 4 Unit shaft system structure diagram.

[0020] Figure 5 Comparison chart of the accuracy of different fault detection methods.

[0021] Figure 6 Flowchart of this invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings: Example 1, please refer to Figure 6 This invention provides a method for diagnosing turbine faults, comprising: The original vibration signal of the steam turbine is collected, and the original vibration signal of the steam turbine is encoded using a noise-reducing autoencoder. The signal is then reduced in dimension and compressed to obtain the encoded signal by the hidden layer dimension constraint. The encoded signal is input into a convolutional neural network, local features are extracted through convolutional layers, and then combined into a feature map with a higher depth dimension after being filtered by pooling layers. By inputting feature maps with higher depth dimensions into a fully connected layer, and fine-tuning the network parameters through backpropagation algorithm in supervised learning mode, the classification results of turbine fault type and location are output.

[0023] Example 2: This invention provides a method for diagnosing turbine faults, comprising: This invention proposes a novel method for steam turbine fault diagnosis, leveraging the advantages of deep learning. First, an autoencoder network is used to reconstruct the original signal output. Then, a convolutional neural network (CNN) is employed, using convolution and pooling to reduce the number of input parameters to the fully connected layers. Data features are extracted through the convolutional steps and combined into deeper, higher-dimensional data. Finally, the output is predicted and classified using fully connected layers. The method is then applied in engineering, and results show that it achieves adaptive extraction of fault features and accurate identification of health status for steam turbine systems under various operating conditions, with a significantly higher accuracy than widely used expert diagnostic systems.

[0024] The specific technical solutions and implementation methods are as follows: (1) In the unsupervised learning mode, the DNN is pre-trained layer by layer to improve its ability to discover fault features of unstable and non-stationary mechanical signals. Then, in the supervised learning mode, the DNN is trained more accurately to improve its real-time monitoring and intelligent fault diagnosis capabilities. A denoising autoencoder (DAE) is used as the unsupervised algorithm in the pre-training stage, and the BP algorithm is used as the supervised algorithm in the fine-tuning stage. Auto-encoder (AE) networks are a typical example of adaptive learning. They take the original signal as input, reconstruct the signal, and inductively output a high-level language expression. The basic principle is as follows: Figure 1 As shown.

[0025] Figure 1 In the image, the area within the dashed blue box represents a self-encoder black box model. This black box integrates the encoder and decoder into a single algorithm, both essentially performing some form of transformation on the input signal. The encoder transforms the acquired non-stationary and unstable mechanical signal x into a encoded signal y, and the decoder converts the encoded y into an output signal. .Right now:

[0026] (2) The autoencoder performs dimensionality reduction and compression on the input signal in the hidden layer and decompresses it in the output layer. Information loss occurs during the dimensionality reduction and inverse decoding process, but this can be avoided through training. The model structure is as follows: Figure 2 As shown.

[0027] The specific formula is as follows: y=f(x)=s(wx+b)(3) x' = g(y) = s(w'y + b') (4) L(x, x') = L(x, g(f(x))) (5) Here, L represents the loss function, which can be the squared error loss or the cross entropy loss depending on the different forms of the data.

[0028] If w' = wT, it is generally called tied weights. By adding a constraint function to the hidden layer, a more accurate output signal can be obtained. In terms of data dimensions, the following two common situations exist: n > p, that is, the dimension of the hidden layer's constraint function is less than the dimension of the input data. This means that the change from x' to y is a dimensionality reduction operation. Through this dimensionality reduction operation, irrelevant variables are eliminated while the main features of the fault signal are retained. When the relationship between variables in each two layers is a linear mapping and the error of supervised training is a quadratic error, this network is equivalent to PCA.

[0029] n < p, that is, the dimension of the hidden layer is greater than the dimension of the input data. It is proposed to make the expression of y as sparse as possible (a large number of dimensions are 0 and not activated). At this time, the encoder is a "sparse autoencoder". From the perspective of features, a sparse expression means that the system is trying to perform feature selection to find out several truly important dimensions among a large number of dimensions.

[0030] (3) Convolutional Neural Networks (CNNs) consist of an input layer, an output layer, and multiple hidden layers. The method of convolution and pooling is used to reduce the number of input parameters in the fully connected layer, thereby reducing the computational amount while maintaining the original features of the data and preventing the overfitting phenomenon. Data features are extracted through the convolution step and combined into data with a deeper depth and higher dimension. Finally, output prediction classification is performed through the fully connected layer. The role of the pooling layer is to pool the output of the convolutional layer. It is a process of screening and filtering the feature values obtained from the convolutional layer, which can screen out useful information in the layer, reduce the model size, and output it for analysis by the next layer. Generally, the max-pooling method and the average-pooling method are used, and the max-pooling method is mostly adopted. In the field of vibration fault diagnosis of steam turbines, taking the vibration time-domain signal as an example, the input signal of the convolutional neural network is one-dimensional data. Figure 3 It is the schematic diagram of the convolutional neural network used for steam turbine fault diagnosis.

[0031] The encoding signal is input into the convolutional neural network, and local features are extracted through the convolutional layer. After being screened by the pooling layer, they are combined into a feature map with a deeper depth and higher dimension, including: Convolutional neural networks (CNNs) consist of input and output layers, as well as various hidden layers. Convolution and pooling reduce the number of input parameters in fully connected layers. Convolution steps extract data features and combine them into deeper, higher-dimensional data. Finally, fully connected layers perform output prediction and classification. Pooling layers pool the output of convolutional layers, filtering the feature values ​​obtained from the convolutional layers, selecting useful information from the layers, reducing the model size, and outputting it to the next layer for analysis. Max pooling is used.

[0032] This neural network uses the vibration signal of a steam turbine rotor as input and extracts features from the original mechanical signal through operations such as downsampling and multi-layer convolution. The extracted feature parameters are then used as input signals for the fully connected layers. This significantly reduces the number of input parameters to the fully connected layers and thus reduces the computational load while ensuring the extraction of vibration features.

[0033] Example 3, taking a Siemens 1000MW unit as the research object, the shaft system structure is as follows: Figure 4 As shown.

[0034] This steam turbine is a 1000MW ultra-supercritical steam turbine manufactured by Shanghai Turbine Co., Ltd., which introduced technology from Siemens in Germany. The model number is CZK1034-28 / 0.5 / 600 / 620.

[0035] This 1000MW ultra-supercritical steam turbine generator unit adopts a novel single-support shaft system structure, which boasts advantages such as compact structure, space saving, and stable and reliable operation under varying conditions. It has become the fastest-growing model among megawatt-class ultra-supercritical units. However, due to the single-support structure of the main bearing and the influence of vibration modes between adjacent rotors, many units already in operation have experienced excessive bearing vibration during commissioning and operation. This steam turbine has generated a large amount of operational data during commercial operation, and the method proposed in this invention was experimentally verified on this turbine. Vibration signals from bearings #1 to #9 were extracted, with each bearing containing x- and y-directions, totaling 18 channels. Turbine faults are categorized into ordinary forced vibration and self-excited vibration. Common fault types of ordinary forced vibration and self-excited vibration, such as imbalance, airflow excitation, and dynamic-static rubbing, were selected for verification. 10,000 sets of rotor fault simulation data were generated, with 80% randomly selected as the training set and the remaining 20% ​​as the test set.

[0036] The proposed method is compared with classic algorithms in traditional machine learning methods, including K-nearest neighbor regression, support vector regression, random forest regression, and boosting tree regression (XGBoostR). Traditional machine learning methods preprocess the raw signal and use a fixed-window Fast Fourier Transform function to obtain time-frequency features. The proposed method, however, performs filtering to remove high-frequency components above 200Hz from non-stationary signals, selecting the top 200 feature values ​​in the frequency space as input variables. The diagnostic accuracy is compared using common forced vibration—imbalance—as the detection task. The proposed method achieves a diagnostic accuracy of 98.89%, with almost no impact on accuracy in multi-task diagnostics, significantly higher than traditional machine learning methods.

[0037] Compared with traditional machine learning methods, the method proposed in this invention has a diagnostic accuracy of 98.89%, and its accuracy is almost unaffected in multi-task diagnosis, which is much higher than that of traditional machine learning methods.

[0038] The accuracy of the proposed method is further demonstrated through confusion matrix analysis and principal component analysis. This method can accurately identify common turbine fault types and pinpoint fault locations.

[0039] Finally, the robustness of the proposed method was verified. By changing the signal-to-noise ratio and channel variables, the impact on the judgment results was summarized. The results show that even after adding different signal-to-noise ratios to the test set, the method's output results remain accurate.

[0040] In another embodiment of the present invention, a turbine fault diagnosis system is provided, which can be used to implement the above-mentioned method. Specifically, the system includes: The data acquisition module is used to acquire the original vibration signal of the steam turbine, encode the original vibration signal of the steam turbine using a noise-reducing autoencoder, and obtain the encoded signal by dimensionality reduction and compression through hidden layer dimensionality constraints. The feature map acquisition module is used to input the encoded signal into the convolutional neural network, extract local features through the convolutional layer, and combine them into a feature map with a higher depth dimension after being filtered by the pooling layer. The classification output module is used to input feature maps with higher depth dimensions into the fully connected layer, fine-tune the network parameters through the backpropagation algorithm in supervised learning mode, and output the classification results of turbine fault type and location.

[0041] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0042] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a steam turbine fault diagnosis method.

[0043] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the turbine fault diagnosis method in the above embodiments.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will 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 program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] 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 scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing steam turbine faults, characterized in that, It includes: Collect the original vibration signals of the steam turbine, encode the original vibration signals of the steam turbine using a denoising autoencoder, and perform signal dimensionality reduction and compression through hidden layer dimension constraints to obtain encoded signals; Input the encoded signals into a convolutional neural network, extract local features through the convolutional layer, and combine them into a feature map with a higher depth dimension after screening by the pooling layer; Input the feature map with a higher depth dimension into the fully connected layer, fine-tune the network parameters through the backpropagation algorithm in the supervised learning mode, and output the classification results of the steam turbine fault types and positions.

2. The turbine fault diagnosis method according to claim 1, characterized in that, The step of collecting the original vibration signals of the steam turbine and encoding the original vibration signals of the steam turbine using a denoising autoencoder includes: Use the denoising autoencoder as an unsupervised algorithm in the pre-training stage, and use the BP algorithm as a supervised algorithm in the fine-tuning stage. The denoising autoencoder uses the original signal as the input layer, performs signal reconstruction and induction, and outputs a high-level language expression; The encoder transforms the acquired non-stationary and unstable mechanical signal x into an encoded signal y, and the decoder converts the encoded y into an output signal. ,Right now:

3. The turbine fault diagnosis method according to claim 1, characterized in that, The step of performing signal dimensionality reduction and compression through hidden layer dimension constraints to obtain encoded signals includes: The autoencoder reduces the dimension of the input signal in the hidden layer and decompresses it in the output layer, as shown in the following formula: where L represents the loss function, which is the quadratic error or cross-entropy error according to different forms of data; When w' = wT, add a constraint function to the hidden layer to obtain a more accurate output signal.

4. The turbine fault diagnosis method according to claim 3, characterized in that, The data dimension of the constraint function satisfies: n > p, for the hidden layer, the dimension of the constraint function is less than the dimension of the input data. The change from x' to y is a dimensionality reduction operation. Through the dimensionality reduction operation, irrelevant variables are eliminated and the main features of the fault signal are retained; when the relationship between every two layers of variables is a linear mapping relationship and the error of the supervised training is a quadratic error, this network is equivalent to PCA; n < p, that is, the dimension of the hidden layer is greater than the dimension of the input data, and the expression sparsity of y is proposed. At this time, the encoder is a "sparse autoencoder".

5. The turbine fault diagnosis method according to claim 1, characterized in that, The step of inputting the encoded signals into a convolutional neural network, extracting local features through the convolutional layer, and combining them into a feature map with a higher depth dimension after screening by the pooling layer includes: The convolutional neural network consists of an input layer, an output layer, and multiple hidden layers. The method of convolution and pooling is used to reduce the number of input parameters of the fully connected layer, extract data features through the convolution step, and combine them into data with a deeper depth and a higher dimension. Finally, output prediction classification through the fully connected layer; the pooling layer is used to pool the output of the convolutional layer, screen and filter the feature values obtained by the convolutional layer, screen the useful information in the layer, reduce the model size, and output it to the next layer for analysis; the maximum pooling method is adopted.

6. The turbine fault diagnosis method according to claim 1, characterized in that, The step of inputting the feature map with a higher depth dimension into the fully connected layer, fine-tuning the network parameters through the backpropagation algorithm in the supervised learning mode, and outputting the classification results of the steam turbine fault types and positions includes: This neural network uses the vibration signals of the steam turbine rotor as the input, and performs feature extraction on the original mechanical signals through operations such as downsampling and multi-layer convolution; the extracted feature parameters are used as the input signals of the fully connected layer.

7. A steam turbine fault diagnosis system, characterized in that, It includes: The data acquisition module is used to acquire the original vibration signal of the steam turbine, encode the original vibration signal of the steam turbine using a noise-reducing autoencoder, and obtain the encoded signal by dimensionality reduction and compression through hidden layer dimensionality constraints. The feature map acquisition module is used to input the encoded signal into the convolutional neural network, extract local features through the convolutional layer, and combine them into a feature map with a higher depth dimension after being filtered by the pooling layer. The classification output module is used to input feature maps with higher depth dimensions into the fully connected layer, fine-tune the network parameters through the backpropagation algorithm in supervised learning mode, and output the classification results of turbine fault type and location.

8. A turbine fault diagnosis system according to claim 7, characterized in that, The process of inputting the encoded signal into a convolutional neural network, extracting local features through convolutional layers, and combining them into a feature map with higher depth dimensions after filtering by pooling layers includes: Convolutional neural networks (CNNs) consist of input and output layers, as well as various hidden layers. Convolution and pooling reduce the number of input parameters in fully connected layers. Convolution steps extract data features and combine them into deeper, higher-dimensional data. Finally, fully connected layers perform output prediction and classification. Pooling layers pool the output of convolutional layers, filtering the feature values ​​obtained from the convolutional layers, selecting useful information from the layers, reducing the model size, and outputting it to the next layer for analysis. Max pooling is used. The process of inputting a feature map with a higher depth dimension into a fully connected layer, fine-tuning the network parameters through backpropagation algorithm in supervised learning mode, and outputting the classification results of turbine fault type and location includes: the neural network uses the turbine rotor vibration signal as input, and performs feature extraction on the original mechanical signal through operations such as downsampling and multi-layer convolution; the extracted feature parameters are used as the input signal of the fully connected layer.

9. A computer 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 computer program, it implements the steps of the turbine fault diagnosis method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the turbine fault diagnosis method as described in any one of claims 1 to 7.

Images