Water turbine vibration fault diagnosis method based on time-frequency analysis and residual network
Through time-frequency analysis and Resnet residual network processing of turbine vibration data, the problems of difficulty in judging vibration fault types and high misjudgment rate in the prior art are solved, and the accurate identification and reliability of turbine faults are achieved.
Patent Information
- Application Number
- CN202410839047.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2024-11-19
AI Technical Summary
The prior art cannot accurately determine the type of vibration failure of the turbine. It relies on manual parameters and experience to set the parameters and experience, and has poor generalization, resulting in complex vibration fault diagnosis and high misjudgment rate.
Using the method based on time-frequency analysis and Resnet residual network, the vibration data is processed through data splicing and short-time Fourier transform, a vibration fault diagnosis model is constructed, and the gradient disappearance problem is alleviated by batch normalization and activation functions to achieve fault-free , accurate identification of single and multiple faults.
It realizes unified processing of vibration data of different lengths, reduces the misjudgment rate, improves the credibility of diagnostic results, provides clear guidance for subsequent maintenance, and improves the accuracy and efficiency of turbine fault diagnosis.
Smart Images

Figure CN118981671A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fault diagnosis, in particular to a method for diagnosing vibration faults of a hydraulic turbine based on time-frequency analysis and a residual network. Background Art
[0002] As the core equipment of a large hydropower station, the operating status of the turbine is directly related to the operating efficiency and safety of the entire hydropower station. However, due to the particularity of the working environment of the turbine, such as long-term exposure to huge water pressure, high-speed rotation and harsh hot and humid environment, long-term operation will inevitably lead to problems such as component wear and material aging, which may in turn cause mechanical failures such as bearing failure, blade failure and seal failure, among which vibration failure is the most common and most serious one. Vibration failure will not only cause equipment efficiency to decrease and energy consumption to increase, but in severe cases it may also cause equipment damage and shutdown accidents, bringing major safety hazards and economic losses to the normal operation of the hydropower station. Therefore, it is important to detect and accurately diagnose vibration failures in a timely manner. The type of fault is very important. Targeted maintenance measures, such as replacing faulty parts and adjusting working parameters, can be taken to reduce the degree of equipment wear, extend the service life of the equipment, reduce the frequency of equipment replacement and maintenance costs, and improve the economic benefits of the hydropower station. In addition to economic considerations, timely detection and handling of vibration faults are also crucial to ensure the safe operation of the hydropower station. Once a serious vibration fault causes the turbine to shut down, it will not only cause power supply interruptions, but may also further damage other related equipment and even cause safety accidents. Therefore, establishing efficient and reliable vibration fault diagnosis technology is not only in the economic interests of the enterprise, but also helps to prevent problems before they occur and ensure the continued safe operation of large hydropower stations.
[0003] However, the common solutions currently available have many shortcomings, including: the existing technology mainly judges the presence of vibration faults based on simple indicators such as the maximum vibration displacement, maximum peak value and root mean square value of vibration velocity, and is unable to further determine the type of vibration fault and guide maintenance; the use of traditional signal analysis technology for vibration fault diagnosis, such as time domain analysis, frequency domain analysis, time-frequency analysis and wavelet analysis, usually requires manual setting of parameters, selection of analysis methods and interpretation of analysis results, which not only increases the complexity and time cost of the analysis, but also depends on the experience and skill level of the analyst, and is subjective and inaccurate; the use of neural network models to directly diagnose vibration faults on vibration signals has the problem of poor generalization. Summary of the invention
[0004] In view of the fact that when diagnosing vibration faults of turbines, the existing technology is unable to further determine the type of vibration faults and guide maintenance, and requires manual setting of parameters, selection of analysis methods and interpretation of analysis results, which relies on the experience and skill level of the analysts and has problems such as poor generalization, the present invention is proposed.
[0005] Therefore, the problem to be solved by the present invention is how to provide a method for length judgment and data splicing of vibration data, realize unified processing of vibration data of different lengths, eliminate dimensional differences between data, alleviate the gradient vanishing problem in deep learning, enable the model to learn data features more fully, realize accurate identification of no faults, single faults and multiple faults, reduce the misjudgment rate, and provide clear guidance for subsequent maintenance and processing.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In the first aspect, an embodiment of the present invention provides a method for diagnosing turbine vibration faults based on time-frequency analysis and residual networks, which includes collecting turbine vibration data and preprocessing it to obtain preprocessed vibration data and construct a vibration fault data set; using the vibration fault data set to perform model training and construct a vibration fault diagnosis model; performing fault diagnosis based on the vibration fault diagnosis model to obtain a vibration fault diagnosis result.
[0008] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network described in the present invention, the turbine vibration data includes training vibration data, vibration fault types of the training vibration data and vibration data to be analyzed; the preprocessing includes the following steps: determining the data length L by using the time window size and the number of time windows of short-time Fourier transform; judging the length of the training vibration data based on the data length L, if the length of the training vibration data is less than the data length L, performing data splicing on the training vibration data, otherwise keeping the original training vibration data unchanged; combining the vibration data after data splicing and the vibration data without data splicing to obtain spliced vibration data; performing short-time Fourier transform on the spliced vibration data based on the set time window size and the number of time windows to obtain time-frequency data; and performing standardization on the time-frequency data to obtain standardized time-frequency data.
[0009] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network of the present invention, the calculation formula of the pre-processed vibration data is as follows:
[0010]
[0011] Among them, S(t, ω) is the time-frequency data after standardization; t is the time variable; ω is the frequency variable; n is the discrete time index of the signal; x*(n) is the vibration data after splicing; w(nt) is the time window function; j is the imaginary unit; μ is the mean; σ is the standard deviation; ε is the stability parameter.
[0012] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network described in the present invention, wherein: the vibration fault data set includes input data and output data; the input data is the time-frequency data after standardized processing; and the output data is the vibration fault type corresponding to the vibration data.
[0013] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network described in the present invention, the construction of the vibration fault diagnosis model includes the following steps: constructing a residual network model based on the Resnet residual network structure; combining batch normalization and activation function, performing convolution operation in each layer of the residual network model, the specific formula is as follows:
[0014]
[0015] Among them, Y res (l) is the output of the lth layer; σ is the activation function; BatchNorm is the batch normalization operation; is the kth convolution kernel of the lth layer; * is the convolution operation; is the output of the l-1th layer; b (l) is a bias term; the deviation between the model prediction result and the true label is calculated using the cross entropy loss function, and a regularization term is added; the residual network model is trained using the vibration fault data set, and the model parameters are optimized by minimizing the total loss function. The calculation formula for minimizing the total loss function is as follows:
[0016]
[0017] Where, Θ is the set of vibration fault diagnosis model parameters; N is the number of samples; C is the number of fault types; is the true fault type of the i-th sample; is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; λ is the regularization coefficient; L is the cross entropy loss function; ||W (l) || F is the F norm of the convolution kernel of the lth layer; a vibration fault diagnosis model is constructed based on the optimized model parameters.
[0018] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network of the present invention, the calculation formula of the vibration fault diagnosis model is as follows:
[0019]
[0020] in, is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; W cis the weight vector of the output layer corresponding to category c; X processed is the output feature of the last layer of the vibration fault diagnosis model; b c is the bias term of the output layer corresponding to category c; C is the number of fault types; W j is the weight vector of the output layer corresponding to category j; b j is the bias term of the output layer corresponding to category j.
[0021] As a preferred solution of the turbine vibration fault diagnosis method based on time-frequency analysis and residual network described in the present invention, wherein: the fault diagnosis is to preprocess the vibration data to be analyzed, input the preprocessed vibration data to be analyzed into the vibration fault diagnosis model and analyze it to obtain the vibration fault diagnosis result; the vibration fault diagnosis result includes: if all the predicted probabilities are all less than the first threshold, the i-th sample is judged to be a non-fault type; if there is only one category of predicted probability If it is greater than or equal to the first threshold, it is judged as a single fault type, and the i-th sample is the corresponding fault type; when there are multiple categories of predicted probabilities Greater than or equal to the first threshold, judging the highest prediction probability The number of categories, if the highest predicted probability If there is only one fault type, it is determined to be a single fault type, and the i-th sample is the corresponding fault type; if the highest predicted probability If there are multiple numbers, then the highest prediction probability is determined. Whether there are multiple cases of the number of consecutive N times, if it occurs N times in a row, it is determined to be a multiple fault type, and the i-th sample is the highest prediction probability of multiple Corresponding fault type; if it does not occur N times in a row, re-collect the vibration data to be analyzed and perform vibration fault diagnosis.
[0022] In the second aspect, in order to further solve the safety problems existing in the diagnosis of turbine vibration faults, the embodiment of the present invention provides a turbine vibration fault diagnosis system based on time-frequency analysis and residual network, which includes: a vibration data module, which is used to collect turbine vibration data and perform data splicing, short-time Fourier transform and standardization processing to obtain preprocessed vibration data and construct a vibration fault data set; a diagnosis model module, which is used to train the residual network using the vibration fault data set to construct a vibration fault diagnosis model; a vibration fault module, which is used to preprocess the vibration data to be analyzed, and perform fault diagnosis using the vibration fault diagnosis model to obtain a vibration fault diagnosis result.
[0023] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of a method for diagnosing turbine vibration faults based on time-frequency analysis and residual network as described in the first aspect of the present invention is implemented.
[0024] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of a method for diagnosing turbine vibration faults based on time-frequency analysis and residual networks as described in the first aspect of the present invention is implemented.
[0025] The beneficial effects of the present invention are as follows: the present invention uses data length to perform length judgment and data splicing on training vibration data, thereby realizing unified processing of vibration data of different lengths, effectively solving the problem of inconsistent length of vibration data in actual engineering, improving data utilization and representativeness, and utilizing short-time Fourier transform and normalization processing to extract time-frequency characteristics of signals and eliminate dimensional differences between data; utilizing the Resnet residual network structure to construct a model, and combining batch normalization with activation functions for convolution operations in each layer to effectively alleviate the gradient vanishing problem in deep learning, so that the model can learn data features more fully; realizing accurate identification of no faults, single faults and multiple faults, reducing the misjudgment rate, improving the credibility of diagnosis results, and providing clear guidance for subsequent maintenance and processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0027] Figure 1 This is a schematic diagram of the network structure of the 18-layer Resnet in Example 1.
[0028] Figure 2 This is a flow chart of vibration fault diagnosis in Example 1. DETAILED DESCRIPTION
[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0032] Example 1
[0033] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for diagnosing turbine vibration faults based on time-frequency analysis and residual network, comprising the following steps:
[0034] S1: Collect turbine vibration data and preprocess it to obtain preprocessed vibration data and construct a vibration fault data set.
[0035] Preferably, the turbine vibration data includes training vibration data, vibration fault types of the training vibration data, and vibration data to be analyzed.
[0036] Specifically, the preprocessing includes the following steps: determining the data length L by using the time window size and the number of time windows of the short-time Fourier transform.
[0037] The length of the training vibration data is judged based on the data length L. If the length of the training vibration data is less than the data length L, the training vibration data is spliced. Otherwise, the original training vibration data is kept unchanged. The specific formula for data splicing is as follows:
[0038]
[0039] in, is the vibration data for training after data splicing; x train (t) is the vibration data for training.
[0040] The spliced vibration data is combined with the unspliced vibration data to obtain the spliced vibration data.
[0041] Based on the set time window size and number of time windows, the spliced vibration data is subjected to short-time Fourier transform to obtain time-frequency data. The specific formula of short-time Fourier transform is as follows:
[0042]
[0043] Among them, x(t) is the time domain signal; w(nt) is the time window function, which is used to analyze the signal within a specific time window; t is the time variable, which is used to indicate the time point of analysis; ω is the frequency variable; n is the discrete time index of the signal; and j is the imaginary unit.
[0044] The time-frequency data is standardized to obtain the standardized time-frequency data. The specific formula for the standardized processing is as follows:
[0045]
[0046] Among them, X is the time-frequency data; μ is the mean; σ is the standard deviation; ε is the stability parameter, which is used to avoid the denominator being zero and enhance stability.
[0047] Specifically, the calculation formula of the preprocessed vibration data is as follows:
[0048]
[0049] Among them, S(t,ω) is the time-frequency data after normalization. The higher the time-frequency data, the greater the vibration intensity of the corresponding frequency in a specific time window. t is the time variable, which is used to indicate the time point of analysis. ω is the frequency variable. n is the discrete time index of the signal. x * (n) is the vibration data after splicing processing; w(nt) is the time window function, which is used to analyze the signal within a specific time window; j is the imaginary unit; μ is the mean; σ is the standard deviation; ε is the stability parameter, which is used to avoid the denominator being zero and enhance stability.
[0050] Preferably, the vibration fault data set includes input data and output data.
[0051] Specifically, the input data is time-frequency data after standardization.
[0052] Specifically, the output data is the vibration fault type corresponding to the vibration data.
[0053] Specifically, by introducing the data splicing operation, data splicing is performed when the length of the vibration data for training is less than the required data length L, which not only ensures the consistency of the input data, but also enhances the adaptability of the model to data of different lengths, which is beneficial to improving the generalization ability of the diagnostic model; by adopting short-time Fourier transform to convert the time domain signal into time-frequency data, the model can simultaneously capture the time domain and frequency domain characteristics of the signal, revealing the changes in the frequency characteristics of the fault in different time periods, thereby providing richer information for fault diagnosis.
[0054] S2: Use the vibration fault dataset to train the model and build a vibration fault diagnosis model.
[0055] Preferably, constructing a vibration fault diagnosis model comprises the following steps: Figure 1 The figure shows the network structure of the 18-layer Resnet. The residual network model is built based on the 18-layer Resnet residual network structure.
[0056] Combined with batch normalization and activation function, convolution operation is performed in each layer of the residual network model. The specific formula is as follows:
[0057]
[0058] Among them, Y res (l) is the output of the lth layer; σ is the activation function; BatchNorm is the batch normalization operation; is the kth convolution kernel of the lth layer; * is the convolution operation; is the output of the l-1th layer; b (l) is the bias term.
[0059] The cross entropy loss function is used to calculate the deviation between the model prediction result and the true label, and a regularization term is added to prevent overfitting. The specific formula is as follows:
[0060]
[0061] Where L is the cross entropy loss function; C is the number of fault types; y c is the actual fault type corresponding to the vibration data; is the prediction probability of the model for a certain category; λ is the regularization coefficient, which is used to control the weight of the regularization term to prevent the model from overfitting; ||W (l) || F is the F-norm of the convolution kernel of the lth layer.
[0062] The residual network model is trained using the vibration fault data set, and the model parameters are optimized by minimizing the total loss function. The calculation formula for minimizing the total loss function is as follows:
[0063]
[0064] Where, Θ is the set of vibration fault diagnosis model parameters; N is the number of samples; C is the number of fault types; is the true fault type of the i-th sample; is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; λ is the regularization coefficient, which is used to control the weight of the regularization term to prevent the model from overfitting; L is the cross entropy loss function; ||W (l) || F is the F-norm of the convolution kernel of the lth layer.
[0065] A vibration fault diagnosis model is constructed based on the optimized model parameters.
[0066] Specifically, the calculation formula of the vibration fault diagnosis model is as follows:
[0067]
[0068] in, is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; W c is the weight vector of the output layer corresponding to category c; X processed is the output feature of the last layer of the vibration fault diagnosis model; b c is the bias term of the output layer corresponding to category c; C is the number of fault types; W j is the weight vector of the output layer corresponding to category j; b j is the bias term of the output layer corresponding to category j.
[0069] Specifically, by introducing batch normalization and activation functions, not only the gradient vanishing problem of deep networks is alleviated, but also the convergence speed and generalization ability of the model are improved. By introducing regularization terms, overfitting is effectively prevented, the performance of the model on unseen data is improved, the incidence of misdiagnosis is reduced, and economic losses are reduced.
[0070] S3: Perform fault diagnosis based on the vibration fault diagnosis model to obtain a vibration fault diagnosis result.
[0071] Preferably, the fault diagnosis is to pre-process the vibration data to be analyzed, input the pre-processed vibration data to be analyzed into a vibration fault diagnosis model and analyze it to obtain a vibration fault diagnosis result.
[0072] Further, such as Figure 2 The following is the implementation process of vibration fault diagnosis. The vibration fault diagnosis results include: if all predicted probabilities are all smaller than the first threshold, the i-th sample is determined to be a non-fault type.
[0073] If there is only one category of predicted probability If it is greater than or equal to the first threshold, it is determined to be a single fault type, and the i-th sample is the corresponding fault type.
[0074] When there are multiple categories of predicted probabilities Greater than or equal to the first threshold, judging the highest prediction probability The number of categories, if the highest predicted probability If there is only one fault type, it is determined to be a single fault type, and the i-th sample is the corresponding fault type.
[0075] If the highest predicted probability If there are multiple numbers, then the highest prediction probability is determined. Whether there are multiple cases of the number of consecutive N times, if it occurs N times in a row, it is determined to be a multiple fault type, and the i-th sample is the highest prediction probability of multiple Corresponding fault type; if it does not occur N times in a row, re-collect the vibration data to be analyzed and perform vibration fault diagnosis.
[0076] Furthermore, based on the particularity of turbine vibration fault diagnosis and actual engineering needs, for each fault type, the probability distribution of the correct prediction of the statistical model is used to set the initial threshold. According to the severity of the turbine fault, a cost matrix is introduced, and different types of misjudgments are assigned different cost weights. The optimal threshold is determined by minimizing the overall expected cost, and then the best first threshold is obtained.
[0077] Specifically, by proposing a multi-level judgment mechanism and using thresholds to filter low-confidence predictions, it helps to reduce false alarms. When multiple high-confidence predictions appear, considering the complexity and instantaneousness of turbine vibration signals, multiple sampling is used to improve the reliability of diagnosis and the ability to identify complex faults.
[0078] The present embodiment also provides a turbine vibration fault diagnosis system based on time-frequency analysis and residual network, including: a vibration data module, used to collect turbine vibration data and perform data splicing, short-time Fourier transform and standardization processing to obtain preprocessed vibration data and construct a vibration fault data set; a diagnosis model module, used to train the residual network using the vibration fault data set to construct a vibration fault diagnosis model; a vibration fault module, used to preprocess the vibration data to be analyzed, and perform fault diagnosis using the vibration fault diagnosis model to obtain a vibration fault diagnosis result.
[0079] This embodiment also provides a computer device, which is applicable to a method for diagnosing turbine vibration faults based on time-frequency analysis and residual networks, and includes: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a method for diagnosing turbine vibration faults based on time-frequency analysis and residual networks as proposed in the above embodiment.
[0080] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.
[0081] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, a method for diagnosing turbine vibration faults based on time-frequency analysis and residual networks as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.
[0082] In summary, the present invention uses data length to perform length judgment and data splicing on training vibration data, thereby realizing unified processing of vibration data of different lengths, effectively solving the problem of inconsistent length of vibration data in actual engineering, improving data utilization and representativeness, and using short-time Fourier transform and normalization processing to extract the time-frequency characteristics of the signal and eliminate the dimensional differences between the data; the Resnet residual network structure is used to build a model, and the gradient vanishing problem in deep learning is effectively alleviated by combining batch normalization and activation function for convolution operations in each layer, so that the model can learn data features more fully; accurate identification of no fault, single fault and multiple faults is achieved, the misjudgment rate is reduced, the credibility of the diagnosis results is improved, and clear guidance is provided for subsequent maintenance and processing.
[0083] Example 2
[0084] Referring to Table 1 and Table 2, which are the second embodiment of the present invention, this embodiment is different from the first embodiment in that, in order to verify its beneficial effects, experimental comparison data between the present invention and the prior art are provided.
[0085] This example uses different types of turbines that have been in use for many years and have vibration problems to varying degrees. Vibration sensors are installed on the turbines with a sampling frequency of 1024 Hz and a collection time of 60 minutes to obtain initial vibration data and perform preprocessing. The vibration fault diagnosis model of the present invention is used to perform vibration fault diagnosis on the vibration data, wherein the time window size of the short-time Fourier transform is 256 and the number of time windows is 64, so the data length L is 16384. The vibration fault diagnosis results are shown in Table 1.
[0086] Table 1 Vibration fault diagnosis result data table
[0087]
[0088] It can be seen from the above table that for different turbines, the corresponding fault diagnosis accuracy rate reaches more than 95%, indicating that the present invention performs well in dealing with different types of faults and effectively improves the accuracy and robustness of vibration fault diagnosis; at the same time, through the diagnosis time, false alarm rate and missed alarm rate, it can be seen that the present invention can effectively control the fault diagnosis time so that maintenance measures can be taken in time, avoiding the loss of production efficiency caused by unnecessary shutdown and maintenance, and improving the calculation efficiency and reliability of fault diagnosis.
[0089] For the fault diagnosis results of the present invention, data comparison is performed using the prior art and the present invention, and the specific data is shown in Table 2.
[0090] Table 2 Data comparison table of the present invention and the prior art
[0091] Experimental methods Average diagnostic accuracy (%) Average diagnosis time (ms) Multiple fault identification rate (%) Method of the present invention 97.4 40.3 92.5 Conventional CNN algorithm 94.1 68.5 85.7
[0092] It can be seen from the comparison between the method of the present invention and the prior art that the present invention effectively improves the accuracy of fault diagnosis, shortens the diagnosis time, and effectively improves the recognition rate of multiple fault types in complex fault scenarios.
[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for diagnosing turbine vibration faults based on time-frequency analysis and residual network, characterized in that: include: Collect turbine vibration data and preprocess it to obtain preprocessed vibration data and construct a vibration fault data set; Using the vibration fault data set to perform model training and build a vibration fault diagnosis model; Fault diagnosis is performed based on the vibration fault diagnosis model to obtain a vibration fault diagnosis result.
2. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network according to claim 1, characterized in that: The turbine vibration data includes training vibration data, vibration fault types of the training vibration data and vibration data to be analyzed; The pre-processing comprises the following steps: The data length L is determined by using the time window size and the number of time windows of the short-time Fourier transform; The length of the training vibration data is judged based on the data length L. If the length of the training vibration data is less than the data length L, the training vibration data is spliced, otherwise the original training vibration data is kept unchanged; Combining the vibration data after data splicing and the vibration data before data splicing to obtain the spliced vibration data; Based on the set time window size and the number of time windows, short-time Fourier transform is performed on the spliced vibration data to obtain time-frequency data; The time-frequency data is standardized to obtain standardized time-frequency data.
3. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network as claimed in claim 2, characterized in that: The calculation formula of the preprocessed vibration data is as follows: Among them, S(t,ω) is the time-frequency data after normalization; t is the time variable; ω is the frequency variable; n is the discrete time index of the signal; x * (n) is the vibration data after splicing processing; w(nt) is the time window function; j is the imaginary unit; μ is the mean; σ is the standard deviation; ε is the stability parameter.
4. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network as claimed in claim 3, characterized in that: The vibration fault data set includes input data and output data; The input data is time-frequency data after standardization; The output data is the vibration fault type corresponding to the vibration data.
5. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network as claimed in claim 4, characterized in that: The construction of the vibration fault diagnosis model comprises the following steps: Build a residual network model based on the Resnet residual network structure; Combined with batch normalization and activation function, convolution operation is performed in each layer of the residual network model. The specific formula is as follows: Among them, Y res (l) is the output of the lth layer; σ is the activation function; BatchNorm is the batch normalization operation; is the kth convolution kernel of the lth layer; * is the convolution operation; is the output of the l-1th layer; b (l) is the bias term; The cross entropy loss function is used to calculate the deviation between the model prediction result and the true label, and a regularization term is added; The residual network model is trained using the vibration fault data set, and the model parameters are optimized by minimizing the total loss function. The calculation formula of the minimization total loss function is as follows: Where, Θ is the set of vibration fault diagnosis model parameters; N is the number of samples; C is the number of fault types; is the true fault type of the i-th sample; is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; λ is the regularization coefficient; L is the cross entropy loss function; ||W (l) || F is the F norm of the convolution kernel of the lth layer; A vibration fault diagnosis model is constructed based on the optimized model parameters.
6. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network as claimed in claim 5, characterized in that: The calculation formula of the vibration fault diagnosis model is as follows: in, is the predicted probability of the vibration fault diagnosis model for the i-th sample belonging to category c; W c is the weight vector of the output layer corresponding to category c; X processed is the output feature of the last layer of the vibration fault diagnosis model; b c is the bias term of the output layer corresponding to category c; C is the number of fault types; W j is the weight vector of the output layer corresponding to category j; b j is the bias term of the output layer corresponding to category j.
7. The method for diagnosing hydraulic turbine vibration faults based on time-frequency analysis and residual network as claimed in claim 6, characterized in that: The fault diagnosis is to pre-process the vibration data to be analyzed, input the pre-processed vibration data to be analyzed into a vibration fault diagnosis model and analyze it to obtain a vibration fault diagnosis result; The vibration fault diagnosis results include: If all predicted probabilities are all less than the first threshold, then the i-th sample is determined to be a non-fault type; If there is only one category of predicted probability If it is greater than or equal to the first threshold, it is determined to be a single fault type, and the i-th sample is the corresponding fault type; When there are multiple categories of predicted probabilities Greater than or equal to the first threshold, judging the highest prediction probability The number of categories, if the highest predicted probability If there is only one fault type, it is determined to be a single fault type, and the i-th sample is the corresponding fault type; If the highest predicted probability If there are multiple numbers, then the highest prediction probability is determined. Whether there are multiple cases of the number of consecutive N times, if it occurs N times in a row, it is determined to be a multiple fault type, and the i-th sample is the highest prediction probability of multiple Corresponding fault type; if it does not occur N times in a row, re-collect the vibration data to be analyzed and perform vibration fault diagnosis.
8. A hydraulic turbine vibration fault diagnosis system based on time-frequency analysis and residual network, based on a hydraulic turbine vibration fault diagnosis method based on time-frequency analysis and residual network according to any one of claims 1 to 7, characterized in that: include, The vibration data module is used to collect turbine vibration data and perform data splicing, short-time Fourier transform and standardization processing to obtain pre-processed vibration data and construct a vibration fault data set; The diagnostic model module is used to train the residual network using the vibration fault data set and build a vibration fault diagnosis model; The vibration fault module is used to pre-process the vibration data to be analyzed, and to perform fault diagnosis using a vibration fault diagnosis model to obtain a vibration fault diagnosis result.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for diagnosing turbine vibration faults based on time-frequency analysis and residual network as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for diagnosing turbine vibration faults based on time-frequency analysis and residual network as described in any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Deep learning-based mechanical bearing vibration fault detection method, computer device and readable storage medium
CN117786368A
Wind turbine generator gearbox fault diagnosis method, system and equipment and storage medium
CN118150157A
Cited By
Fault diagnosis method and device for electro-hydraulic servo valve of aircraft brake system
CN120180310A
