Water turbine fault monitoring method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a water turbine fault monitoring method and device, computer equipment, a readable storage medium and a program product. The method comprises the steps that sound signals generated in the operation process of the water turbine are collected through a microphone array arranged in advance; converting the sound signal into a corresponding multi-channel digital signal; feature extraction is carried out on the multiple paths of digital signals, analysis is carried out based on extracted feature data, and the current running state of the water turbine is determined; and according to the current running state of the water turbine and the inherent characteristic parameters of the water turbine, performing health degree evaluation on the water turbine to obtain an evaluation result. Therefore, the abnormal sound source in the operation process of the water turbine can be spatially positioned in real time, the accuracy of fault part judgment is improved, a whole-process monitoring system covering signal acquisition, conversion, analysis and health assessment can be formed, the monitoring cost is well reduced, stable operation of the water turbine is guaranteed, and non-planned shutdown loss is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault monitoring, in particular to a method and device for monitoring the fault of a hydraulic turbine, a computer device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] As the core equipment of a hydroelectric power system, the stable and efficient operation of a hydraulic turbine is crucial to the entire power production. Due to the long-term operation of the hydraulic turbine in a complex and harsh water flow environment, it often faces many fault risks, and therefore the fault of the hydraulic turbine needs to be accurately monitored.

[0003] Among many monitoring methods, the monitoring of parameters such as vibration, temperature, pressure and sound is more common. The operating state information of the equipment is obtained by sensors, and with the help of data analysis and signal processing methods such as frequency spectrum analysis and time domain analysis, the fault phenomenon and cause are studied in depth. Among them, sound data as an important representation of the operating state of the equipment has great potential in the fault monitoring of hydroelectric equipment.

[0004] However, during the operation of the hydroelectric equipment, many sounds are not within the audible range of the human ear, and the environment in the key areas such as the water wheel chamber and the wind tunnel is complex. The existing acoustic monitoring methods are difficult to comprehensively and accurately obtain the sound information of the hydraulic turbine during operation and to effectively analyze it. SUMMARY

[0005] Therefore, it is necessary to provide a method and device for monitoring the fault of a hydraulic turbine, a computer device, a computer readable storage medium and a computer program product, which can spatially locate abnormal sound sources during the operation of the hydraulic turbine and improve the accuracy of fault location.

[0006] In a first aspect, the present application provides a method for monitoring the fault of a hydraulic turbine, which comprises:

[0007] acquiring sound signals generated during the operation of the hydraulic turbine by a pre-arranged microphone array;

[0008] converting the sound signals into corresponding multi-channel digital signals;

[0009] extracting features from the multi-channel digital signals and analyzing the extracted feature data to determine the current operating state of the hydraulic turbine;

[0010] performing health assessment on the hydraulic turbine according to the current operating state of the hydraulic turbine and the inherent characteristic parameters of the hydraulic turbine to obtain an assessment result.

[0011] In one embodiment, the conversion of the sound signals into corresponding multi-channel digital signals comprises:

[0012] amplify the collected sound signal through a low noise amplifier to obtain an amplified signal;

[0013] filter the amplified signal through a band-pass filter to obtain a signal after removing noise;

[0014] convert the signal after removing noise into a digital quantity in binary data format through a differential synchronous conversion mode;

[0015] convert the digital quantity in binary data format into a floating point number and pack it through a frame synchronization mechanism to obtain a plurality of digital signals.

[0016] In one embodiment, the plurality of digital signals are subjected to feature extraction, and based on the extracted feature data, the current operating state of the water turbine is determined, including:

[0017] write the plurality of digital signals into a cache and queue them in the form of data frames;

[0018] select any one of the digital signals from the cache for normalization and mean removal processing to obtain a pre-processed signal;

[0019] extract time-frequency features from the pre-processed signal to obtain time-frequency feature data, the time-frequency feature data including at least one of fundamental frequency, fundamental frequency energy proportion, odd-even multiple frequency, high-low frequency ratio, root mean square, peak value, zero-crossing rate, kurtosis, skewness, entropy, margin, instantaneous sound pressure level change rate, and instantaneous voiceprint component change rate;

[0020] determine Mahalanobis distance based on the time-frequency feature data, and determine the current operating state of the water turbine according to a Mahalanobis distance threshold, the operating state including normal or abnormal.

[0021] In one embodiment, when the current operating state of the water turbine is abnormal, the method further includes:

[0022] extract corresponding wavelet time-frequency graph features based on the pre-processed signal;

[0023] input the wavelet time-frequency graph features into an inference model to determine whether the water turbine has a fault through the inference model;

[0024] when there is a fault, use an array positioning method to determine the fault location of the water turbine based on the plurality of digital signals; wherein the array positioning method includes decomposing the plurality of digital signals into a plurality of sub-band signals through wavelet packet decomposition, extracting corresponding signal subspaces for each sub-band signal, and determining the azimuth and elevation of the sound source in combination with the geometric characteristics of the circular array.

[0025] In one of the embodiments, the health degree of the water turbine is evaluated according to the current operating state of the water turbine and inherent characteristic parameters of the water turbine, and an evaluation result is obtained, which includes:

[0026] The health degree of the water turbine is evaluated according to the current operating state of the water turbine, equipment parameters of the water turbine, and historical operating data of the water turbine, and an evaluation result is obtained.

[0027] According to the evaluation result, it is determined whether to send alarm information to a superior system or an operation and maintenance terminal device, and the alarm information includes a fault level and / or a risk degree.

[0028] In one of the embodiments, in the case of determining to send alarm information to the superior system or the operation and maintenance terminal device, the method further includes:

[0029] The sound signal corresponding to the alarm moment is stored and marked as fault data, and the fault data is used for backtracking analysis and iterative optimization of the inference model.

[0030] In a second aspect, the present application further provides a water turbine fault monitoring device, which includes:

[0031] A microphone array is arranged in the water turbine room and used to collect sound signals generated during the operation of the water turbine.

[0032] A signal conditioning module is used to convert the sound signals into corresponding multi-channel digital signals.

[0033] A signal processing module is used to extract features from the multi-channel digital signals and analyze the extracted feature data to determine the current operating state of the water turbine, and evaluate the health degree of the water turbine according to the current operating state of the water turbine and inherent characteristic parameters of the water turbine, and obtain an evaluation result.

[0034] In a third aspect, the present application further provides a computer device including a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0035] Sound signals generated during the operation of the water turbine are collected through a pre-arranged microphone array.

[0036] The sound signals are converted into corresponding multi-channel digital signals.

[0037] Features are extracted from the multi-channel digital signals, and the extracted feature data is analyzed to determine the current operating state of the water turbine.

[0038] According to the current operating state of the water turbine and inherent characteristic parameters of the water turbine, health degree of the water turbine is evaluated to obtain an evaluation result.

[0039] In a fourth aspect, the present application further provides a computer readable storage medium, which has a computer program stored thereon, and the computer program is executed by a processor to implement the following steps:

[0040] Collect sound signals generated in the operation of the water turbine through a pre-arranged microphone array;

[0041] Convert the sound signals into corresponding multi-channel digital signals;

[0042] Extract features of the multi-channel digital signals, and analyze the extracted feature data to determine the current operating state of the water turbine;

[0043] According to the current operating state of the water turbine and inherent characteristic parameters of the water turbine, health degree of the water turbine is evaluated to obtain an evaluation result.

[0044] In a fifth aspect, the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the following steps:

[0045] Collect sound signals generated in the operation of the water turbine through a pre-arranged microphone array;

[0046] Convert the sound signals into corresponding multi-channel digital signals;

[0047] Extract features of the multi-channel digital signals, and analyze the extracted feature data to determine the current operating state of the water turbine;

[0048] According to the current operating state of the water turbine and inherent characteristic parameters of the water turbine, health degree of the water turbine is evaluated to obtain an evaluation result.

[0049] The water turbine fault monitoring method, device, computer equipment, computer readable storage medium and computer program product can collect sound signals generated in the operation of the water turbine through the pre-arranged microphone array, so that sound signals of the water turbine from multiple different angles can be collected. The sound signals are converted into corresponding multiple digital signals, so that massive original sound signals do not need to be transmitted to the back end for processing, the communication pressure is reduced, the real-time performance of data processing is improved, and subsequent feature extraction and analysis of the signals can be facilitated to determine whether the operation state of the water turbine is normal. The current operation state of the water turbine is determined by performing feature extraction on the multiple digital signals and performing analysis based on the extracted feature data, so that the operation state of the water turbine can be accurately analyzed, the single sound source positioning ambiguity problem is solved, the fault position is accurately identified, abnormal conditions are found in time, and the maintenance efficiency is improved. The health degree of the water turbine is evaluated according to the current operation state of the water turbine and the inherent characteristic parameters of the water turbine, and an evaluation result is obtained. Therefore, the abnormal sound source in the operation of the water turbine can be spatially positioned in real time, the accuracy of fault position determination is improved, and a full-process monitoring system covering signal collection, conversion, analysis and health evaluation can be formed, the monitoring cost is greatly reduced, the stable operation of the water turbine is ensured, and the loss of unplanned shutdown is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 A flowchart of a water turbine fault monitoring method in an embodiment;

[0052] Figure 2 A flowchart of a water turbine fault monitoring method in another embodiment;

[0053] Figure 3 A structural diagram of a water turbine fault monitoring system in another embodiment;

[0054] Figure 4 A signal processing process diagram of a signal processing module in an embodiment;

[0055] Figure 5 A structural block diagram of a water turbine fault monitoring device in an embodiment;

[0056] Figure 6 An internal structure diagram of a computer equipment in an embodiment. DETAILED DESCRIPTION

[0057] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application and not to limit the present application.

[0058] It should be noted that the terms "comprising" and "having" and any variations thereof used herein are intended to cover non-exclusive inclusion. The term "a plurality of" used herein refers to two or more. The term "and / or" used herein refers to one of the options or any combination of the options.

[0059] As an important clean energy, hydropower has the advantages of being renewable, clean, low cost, and strong regulation capacity, and plays an important role in global energy supply. The water turbine converts the gravitational potential energy of water into kinetic energy, and then drives the generator to rotate to generate electric energy. At the same time, it also plays a key role in peak regulation, phase regulation and emergency backup in the power grid system. However, the water turbine is often subject to many failure risks due to its long-term operation in complex and harsh water flow conditions. For example, impurities such as silt in the water flow can cause erosion and wear on the surface of the blade, making the blade surface rough and reducing the power generation efficiency. Long-term wear can reduce the thickness of the blade, affect the structural strength, cause the blade to deform, and ultimately cause the water turbine to fail. Once the water turbine unit fails, not only will it cause downtime, but it will also affect the economic benefits of the power station, and in severe cases, it may cause significant property damage and casualties. Therefore, in order to be able to monitor water turbine failures in a timely manner, various monitoring technologies have been proposed.

[0060] In related technologies, methods for monitoring water turbine failures are covered, including neural networks, support vector machines (SVM), statistical analysis, and fuzzy diagnosis. These methods are suitable for fault diagnosis under various operating conditions and complex structures of water turbines, and have the advantages of high precision, strong robustness, and strong generalization ability. However, data-driven methods represented by deep learning methods have "black box" characteristics due to their multi-layer nonlinear mapping, and the models lack interpretability. At the same time, the computational models constructed by such methods are usually large, with parameter scales of millions or even tens of millions, requiring high hardware computing power. In practical applications, they often rely on backend servers for data processing and analysis. This not only causes significant real-time delays in fault monitoring (from data acquisition, transmission to backend processing, and result feedback, the entire process may take several seconds or even tens of seconds) and cannot respond to sudden failures of the water turbine in a timely manner; but also causes huge communication pressure, with massive raw monitoring data continuously transmitted to the backend server, which is prone to network congestion, increasing the risk of data loss, and also significantly increasing the communication cost and server load of the power station.

[0061] Among the many monitoring means, the monitoring of parameters such as vibration, temperature, pressure, sound, etc. is relatively common. The device operating state information is obtained through sensors, and in-depth research is conducted with the aid of data analysis and signal processing methods, such as frequency spectrum analysis, time domain analysis, etc. to identify fault phenomena and causes. Among them, sound data as an important representation of the device operating state has great potential in the fault monitoring of hydroelectric equipment. However, in the process of hydroelectric equipment operation, many sounds are not within the audible range of the human ear, and the environment in key areas such as the water wheel chamber and the wind tunnel is complex. The existing acoustic monitoring means cannot comprehensively and accurately obtain the sound information of the water turbine in operation and conduct effective analysis.

[0062] Traditional acoustic monitoring mostly uses single sound source collection methods, that is, only a single microphone or a simple acoustic sensor is used. This method can only obtain sound signals at a single position and cannot determine the spatial position of the sound source through the differences in sound signals at multiple positions, such as distinguishing abnormal vibration sounds at different parts of the blade, friction sounds at different components of the bearing, etc. Therefore, it is difficult to accurately determine the specific position of the fault, which brings great inconvenience to the troubleshooting and maintenance (often requiring a comprehensive detection after shutdown, further expanding the economic losses caused by the fault). At the same time, under the interference of complex environmental noise, the extraction and identification ability of fault characteristic sound is limited, and misjudgment or omission is likely to occur.

[0063] In summary, the existing water turbine fault monitoring technology has problems such as insufficient real-time performance, low accuracy, poor adaptability, and lack of explainability when facing the complex operating environment of the water turbine and various fault types, and it is difficult to meet the demand for efficient and reliable monitoring of the water turbine.

[0064] In view of the problems in the related art, the present application aims to provide a water turbine fault monitoring method, which can solve the problem that the traditional single acoustic monitoring means cannot accurately locate the fault position, and realize spatial positioning of abnormal sound sources in the operation process of the water turbine through a sound array, thereby improving the accuracy of fault position judgment.

[0065] In addition, the water turbine fault monitoring method provided by the present application can also solve the problem that the existing data-driven diagnosis method has a large calculation model and is difficult to deploy on a front-end device. By integrating edge computing into the sensor end, the device has the functions of collection, processing and diagnosis, reduces the data transmission pressure, improves the real-time performance of fault monitoring, and responds to sudden faults of the water turbine in a timely manner.

[0066] The method in the embodiments of the present application can realize accurate, real-time and reliable monitoring of water turbine faults, reduce the cost of monitoring, ensure the stable operation of the hydroelectric power generation system, and reduce the economic losses caused by faults.

[0067] In an exemplary embodiment, as shown inFigure 1 As shown, a water turbine fault monitoring method is provided, and the method in the embodiment can include the following steps S101 to S104. Among them:

[0068] Step S101, collecting the sound signal generated in the water turbine operation process through the pre-arranged microphone array.

[0069] In the embodiment, a microphone array can be pre-arranged in the water turbine room to be monitored, and the microphone array can also integrate signal conditioning, processing and other functional modules. Among them, the microphone array is directed towards the water turbine (pump group), that is, it is arranged directionally (for example, it is installed on the wall of the side facing the water turbine, and the height is about 2-3 meters), so as to ensure that the microphone array can effectively collect the sound signal generated by all components in the water turbine operation process.

[0070] As an optional example, 8 Micro-Electro-Mechanical Systems (MEMS) microphones can be selected to form a uniform circular surface array, and the array element radius is 5 centimeters, so as to form a two-dimensional spatial acoustic perception range. Optionally, in order to enhance the forward sound receiving effect, the microphone sound receiving structure can be set to a horn mouth shape with a 45-degree cutting angle.

[0071] Step S102, converting the sound signal into corresponding multi-channel digital signal.

[0072] In the embodiment, the sound signal collected by the microphone array is converted into a corresponding digital signal, and the sound signal includes the acoustic characteristics of the water turbine under normal and abnormal working conditions. Converting the analog domain sound signal into a digital domain digital signal can facilitate subsequent feature extraction and analysis.

[0073] As an optional example, multiple processing links can be set during the conversion of the sound signal into a digital signal. For example, first, the collected sound signal is amplified by a low-noise amplifier to obtain an amplified signal; the amplified signal is filtered by a band-pass filter to obtain a signal after removing noise; then the signal after removing noise is converted into a digital quantity in binary data format through a differential synchronous conversion method; finally, the digital quantity in binary data format is converted into a floating point number, and is packaged through a frame synchronization mechanism to obtain a multi-channel digital signal.

[0074] Exemplarily, a low-noise amplifier (such as TI LMV822) is used to amplify a weak signal of 0.1-10 mV to 0-3 V (gain 40-60 dB) to ensure that the signal dynamic range is completely retained; then, an 8-stage Butterworth band-pass filter (20 Hz-20 kHz) is used to filter out 50 / 60 Hz power frequency and high-frequency environmental noise, and focus on the fault feature frequency of the water turbine; a 16-bit sigma-delta analog-to-digital converter (such as ADS1255) is used to realize differential synchronous conversion (CMRR>100 dB) at a sampling rate of 48 kHz, and convert the analog signal into digital quantity; finally, the binary data is formatted into IEEE 754 floating-point numbers, and is packaged and output through a frame synchronization mechanism, to provide a high signal-to-noise ratio and phase-consistent digital signal for subsequent algorithms.

[0075] In step S103, feature extraction is performed on the multiple digital signals, and analysis is performed based on the extracted feature data to determine the current operating state of the water turbine.

[0076] In this embodiment, first, the multiple digital sound signals are analyzed. Then, feature extraction is performed on the multiple digital signals, and the current operating state (normal / abnormal) of the water turbine is determined based on the extracted features. Further, when it is determined that the current operating state of the water turbine is abnormal, the fault occurrence position (such as a blade, a bearing, or the like) needs to be accurately located, and state information is output.

[0077] Exemplarily, the multiple digital signals are written into a cache and queued in the form of data frames; any one of the digital signals in the cache is selected for normalization and de-meaning processing to obtain a preprocessed signal; time-frequency feature extraction is performed on the preprocessed signal to obtain time-frequency feature data, the time-frequency feature data including at least one of a fundamental frequency, a fundamental frequency energy proportion, an odd-even multiple frequency, a high-low frequency ratio, a root mean square, a peak value, a zero-crossing rate, a kurtosis, a skewness, an entropy, a margin, an instantaneous sound pressure level change rate, and an instantaneous voiceprint component change rate; a Mahalanobis distance is determined based on the time-frequency feature data, and the current operating state of the water turbine is determined according to a Mahalanobis distance threshold, the operating state including normal or abnormal.

[0078] Optionally, a double-buffer design (1 MB per buffer, supporting 8 channels of data) is used for the cache, and the parallel execution of data writing and processing is realized through ping-pong operation.

[0079] Optionally, each data frame contains 4800 sampling points (corresponding to 100 ms of audio, with a sampling rate of 48 kHz); a circular queue is used to manage the data frames, and when the queue is full (threshold 100 frames), overload protection is triggered, and old frames are preferentially discarded to ensure real-time performance.

[0080] Wherein, the normalization refers to using maximum minimum equalization and each data subtracting mean value and then dividing by maximum value; the mean value elimination refers to eliminating direct current bias through high pass filtering (cutoff frequency 10 Hz), thereby avoiding the influence of baseline drift on feature extraction.

[0081] Wherein, the odd-even frequency multiplication refers to calculating the proportion of the energy sum of the odd times frequency multiplication of the fundamental frequency and the energy sum of the even times frequency multiplication.

[0082] Wherein, the high-low frequency ratio refers to comparing the energy distribution of the low frequency band (0-5 kHz, such as bearing vibration) and the high frequency band (5-20 kHz, such as cavitation noise), thereby improving the sensitivity to fault types.

[0083] In the embodiment, the multi-channel digital signals obtained by real-time conversion are sequentially written into the cache to form a data stream to be processed, and the data stream is queued in the form of data frames and waits for subsequent processing. Optionally, in order to alleviate the computing pressure, one channel of data can be selected for normalization and mean value elimination operation (i.e. one data line is selected for endpoint detection to determine whether the voiceprint component has changed), and then the time-frequency domain features of the data are extracted. The Mahalanobis distance is calculated for the time-frequency domain features, and if the obtained Mahalanobis distance exceeds the preset Mahalanobis distance threshold, it is determined that the digital signal is abnormal. Correspondingly, if the obtained Mahalanobis distance does not exceed the preset Mahalanobis distance threshold, it is determined that the digital signal is normal.

[0084] For example, 1000 frames of data after the device is started are used as a reference distribution model to calculate the covariance matrix Σ and the mean vector μ of the feature vector, and then a dynamic threshold is set; the threshold T (confidence 99.7%) is determined based on χ² distribution, and the calculation formula is:

[0085]

[0086] In the formula, D represents the Mahalanobis distance, and x represents the feature value.

[0087] Optionally, every 1000 frames of data are processed, and the sliding window (the last 500 frames) is used to re-estimate Σ and μ to adapt to the feature drift caused by device aging.

[0088] As another example, in the case that the current operating state of the hydraulic turbine is abnormal, the corresponding wavelet time-frequency graph features are extracted based on the preprocessed signals; the wavelet time-frequency graph features are input into the inference model to determine whether the hydraulic turbine has a fault through the inference model; in the case that there is a fault, the array positioning method is used to determine the fault position of the hydraulic turbine based on the multi-channel digital signals; wherein, the array positioning method comprises: decomposing the multi-channel digital signals into a plurality of sub-band signals through wavelet packet, extracting the corresponding signal subspace for each sub-band signal, and combining the geometric characteristics of the circular array to determine the azimuth and pitch angle of the sound source.

[0089] In this embodiment, the array positioning method is a rotation-invariant technique signal parameter estimation method introducing multi-resolution analysis. By using wavelet packet decomposition, the wide frequency is divided into each sub-band analysis, thereby improving the positioning accuracy of different frequency fault sound sources.

[0090] For example, the wavelet decomposition parameters include: selecting db4 wavelet basis, 5-layer decomposition, covering the frequency band range (20-20KHz); wherein, the approximate coefficient A5: 0-1.5kHz (bearing low-frequency fault); the detail coefficient D5: 1.5-3kHz (loose blade); the detail coefficient D4: 3-6kHz (cavitation noise); the detail coefficient D3: 6-12kHz (high-frequency cavitation).

[0091] For example, the time-frequency diagram construction refers to arranging the energy of each frequency band in time sequence to generate a 224x224 pixel grayscale image with a pixel value range of 0-255 (after normalization). The time-frequency diagram features are input into the inference model for model inference. If a fault is detected, the fault location is located. If no fault is detected, the next frame of data is processed and calculated.

[0092] In this embodiment, the inference model can use an improved Molilenet-V3 model, which improves the network depth to adapt to the time-domain continuity of acoustic signals, and improves the convolution kernel size to enhance the ability to extract specific fault feature frequencies.

[0093] For example, the structure of the improved Molilenet-V3 model is shown in Table 1, which can include 9 modules, and is composed of an input layer (224x224x1), a feature extraction backbone network, a global average pooling, a classification head, and an output layer in hierarchical level. The improved Molilenet-V3 model is adapted to the input size and optimized in structure, while maintaining the lightweight advantage, fully utilizing the high-resolution information of the 224x224 time-frequency diagram, thereby significantly improving the recognition ability of weak fault features (such as early bearing pitting).

[0094] Table 1

[0095]

[0096] Optionally, the inference model can be a lightweight model after compression and quantization to reduce the pressure of hardware calculation. The model training process is briefly introduced as follows:

[0097] 1) First, train a high-precision baseline model that adapts to 224x224 input, ensuring fault diagnosis accuracy ≥ 97.8% in full-precision state, and save the complete parameters as the quantization benchmark. Select 500 representative acoustic data to construct a calibration set, covering typical working conditions such as normal operation, cavitation, and bearing wear. Generate 224x224 wavelet time-frequency graphs for each frame to ensure consistent data distribution with actual monitoring scenarios. Simultaneously configure the RK3588 dedicated toolchain, install RKNN-Toolkit2 (v1.6.0), verify the compatibility of the Neural Processing Unit (NPU) driver with hardware, and confirm the support for INT8 / INT16 quantization modes through test scripts to provide an environmental foundation for subsequent quantization.

[0098] 2) First, analyze the quantization sensitivity of each layer of the model to identify key layers such as the SE attention module (Squeeze-and-Excitation Module) and classification head that are sensitive to quantization, which directly affect the identification accuracy of high-frequency cavitation features and low-frequency bearing features. Adopt a mixed precision quantization strategy: keep the INT16 precision for core sensitive layers to reduce errors, and use INT8 quantization for the remaining depth separable convolution layers to adapt to the native computing advantages of RK3588 NPU. Calibrate the quantization parameters using the KL divergence algorithm (Kullback-Leibler divergence), set a segmented threshold for different frequency band features, relax the quantization interval for low-frequency features below 1kHz, and tighten the precision for high-frequency features above 8kHz to ensure that key fault signals are not lost.

[0099] 3) Combine quantization with channel pruning to prune redundant channels with a contribution degree <0.05, reducing the number of parameters by 30% while compensating for accuracy loss through knowledge distillation. When deploying, enable the NPU operator fusion function of RK3588 to merge Conv+BN + activation layers into a single computing unit, and use dynamic batch processing (maximum batch size=8) to improve throughput.

[0100] As a further example, if the result identified by the inference model is that there is a fault, the multi-path digital signal is calculated using an array positioning method to locate the position of the fault point. For example, a 20Hz-20kHz wide frequency signal is first divided into 16 sub-bands (each sub-band with a bandwidth of 1.25kHz) by 4-layer wavelet packet decomposition, realizing multi-resolution frequency division: the low frequency band (20Hz-2kHz) corresponds to bearing wear and other faults, and the high frequency band (10kHz-20kHz) corresponds to blade cavitation characteristics. Then, the improved ESPRIT algorithm is applied to each sub-band signal to extract the signal subspace by constructing a rotation invariant matrix, and the azimuth and elevation angles of the sound source are solved in combination with the geometric characteristics of the circular array.

[0101] In step S104, the health degree of the hydraulic turbine is evaluated according to the current operating state of the hydraulic turbine and the inherent characteristic parameters of the hydraulic turbine, and an evaluation result is obtained.

[0102] In this embodiment, the health degree of the hydraulic turbine is evaluated according to the current operating state of the hydraulic turbine, the equipment parameters of the hydraulic turbine, and the historical operating data of the hydraulic turbine, and an evaluation result is obtained. According to the evaluation result, it is determined whether to send alarm information to a superior system or an operation and maintenance end device. The alarm information includes a fault level and / or a risk degree.

[0103] As an optional example, a health degree evaluation model can be constructed, which can quantitatively evaluate the current operating state of the hydraulic turbine, the equipment parameters of the hydraulic turbine, and the historical operation data of the hydraulic turbine, and determine the health degree score based on the corresponding weight value. Specifically, the hydraulic turbine fault monitoring can be divided into three levels, such as the equipment layer, the component layer, and the index layer, wherein the weight of the equipment layer can be set to 40%, and each equipment parameter will be analyzed in combination with the historical operation data of the hydraulic turbine, so as to determine the threshold range of each equipment parameter in normal operation (which can be dynamic). For example, when the equipment parameter is not within the threshold range, 0 points are obtained; when the equipment parameter is within the threshold range, 1 point is obtained. The weight of the component layer can be set to 15%, and the component layer can be further refined to include runner, bearing, sealing system, guide vane mechanism, etc., when a component fails (whether a component fails can be analyzed based on the historical operation data of the component), 0 points are obtained; when the component is in normal operation, 1 point is obtained. The index layer can include real-time monitored vibration, temperature, pressure, efficiency, leakage, etc. data, which are mainly used to judge whether the current operating state of the hydraulic turbine is normal (the weight of the index layer can be set to 45%). For example, when the operating state is normal, 1 point is obtained; when the operating state is abnormal, 0 point is obtained. Of course, a score within the range of 0-1 can also be set according to the monitored data, for example, when the score is greater than 0.5, it is considered that the operating state is normal, and when the score is less than 0.5, it is considered that the operating state is abnormal. Further, the scores of the three levels are multiplied by the weight of the level respectively, to obtain a comprehensive score. The health degree of the hydraulic turbine is determined based on the comprehensive score.

[0104] As another optional example, the health degree can also be analyzed from different evaluation dimensions respectively, such as the operating state dimension (the current operating state of the hydraulic turbine), the equipment parameter dimension (the equipment parameters of the hydraulic turbine), and the historical data dimension (the historical operation data of the hydraulic turbine). For example, the evaluation parameters of the operating state dimension can include: vibration amplitude, bearing temperature, and main bearing sealing temperature, etc. The evaluation parameters of the equipment parameter dimension can include: design efficiency and actual efficiency, working water head adaptability, etc. The evaluation parameters of the historical data dimension can include: fault downtime frequency (times / year), and maintenance cycle compliance rate, etc. For different evaluation parameters, a health degree score can be obtained by manual method (expert evaluation) or automatic method (intelligent agent), and the higher the health degree score, the lower the risk level.

[0105] For example, when the health degree score is 0.9-1, it is determined to be healthy; when the health degree score is 0.8-0.9, it is determined to be sub-healthy; when the health degree score is 0.7-0.8, it is determined to be slightly ill; and when the health degree score is less than 0.7, it is determined to be moderate or above failure.

[0106] It should be understood that in the embodiments of the present application, the score range of the health degree can be flexibly set according to actual conditions, and the embodiments do not limit the specific value range.

[0107] Optionally, the analysis result of the current data frame signal (time stamp of the current frame, fault feature, diagnosis result, and positioning information summary) can also be arranged. In order to reduce communication pressure, the analysis result of the frame information is sent to the host program through the bus only when a fault occurs, for the host program to make the next step of processing.

[0108] In the water turbine fault monitoring method, the sound signals generated in the operation of the water turbine are collected through the pre-arranged microphone array; so that sound signals of the water turbine at multiple different angles can be collected. The sound signals are converted into corresponding multi-channel digital signals; so that the massive original sound signals do not need to be transmitted to the back end for processing, the communication pressure is reduced, the real-time performance of data processing is improved, and the subsequent feature extraction and analysis of the signals can also be facilitated to determine whether the operation state of the water turbine is normal. The current operation state of the water turbine is determined by feature extraction on the multi-channel digital signals and analysis based on the extracted feature data; so that the operation state of the water turbine can be accurately analyzed, the single sound source positioning ambiguity problem is solved, the fault position is accurately identified, abnormal conditions are discovered in time, and the maintenance efficiency is improved. The health degree of the water turbine is evaluated according to the current operation state of the water turbine and the inherent characteristic parameters of the water turbine, and an evaluation result is obtained. So that the spatial positioning of the abnormal sound source in the operation of the water turbine can be realized in real time, the accuracy of fault position judgment is improved, and a full-process monitoring system covering signal collection, conversion, analysis and health evaluation can be formed, which greatly reduces the monitoring cost, ensures the stable operation of the water turbine, and reduces the loss of unplanned shutdown.

[0109] In another exemplary embodiment, as shown in Figure 2 A water turbine fault monitoring method is provided, and the method in the embodiment can include the following steps S201 to S205. Wherein:

[0110] Step S201, collecting sound signals generated in the operation of the water turbine through a pre-arranged microphone array.

[0111] Step S202, converting the sound signals into corresponding multi-channel digital signals.

[0112] Step S203, feature extraction on the multi-channel digital signals, and analysis based on the extracted feature data to determine the current operation state of the water turbine.

[0113] Step S204, evaluating the health degree of the water turbine according to the current operation state of the water turbine and the inherent characteristic parameters of the water turbine, and obtaining an evaluation result.

[0114] In this embodiment, the specific implementation process and technical effects of steps S201-S204 are described above, and are not repeated here. Figure 1 The related descriptions of steps S101-S104 in the method embodiment shown in FIG. 1 are referred to, and are not repeated here.

[0115] In step S205, the sound signal corresponding to the alarm time is stored and marked as fault data.

[0116] The fault data is used for backtracking analysis and iterative optimization of the inference model.

[0117] In this embodiment, the sound signal can be collected, converted, analyzed, and health degree evaluated in the whole process, and the sound signal corresponding to the alarm time is stored and marked, which is used for iterative optimization of the inference model, so that a closed-loop fault monitoring system can be formed, the inference model can be continuously optimized, and the performance of the inference model can be improved.

[0118] In one exemplary embodiment, as shown in FIG. 2, a water turbine fault monitoring system is provided, which can include a power module, a data transmission module, a control processing module, a storage module, a microphone array, a signal conditioning module, a signal processing module, and a field alarm module. Figure 3 The microphone array is used to collect water turbine room sound signals, and is a uniform surface array. The signal conditioning module is used to convert analog signals (sound signals) into digital signals. The signal processing module is used to analyze the operating condition state of the water turbine. The control processing module is used to control the flow operation of the whole device, coordinate the operation of each module, and integrate intelligent decision logic. The storage module is used for local storage of data, meets the whole-process tracing needs of the water turbine fault monitoring, is used for storage of device program code, and meets the device operation needs. The field alarm module is used for field alarm warning. The data transmission module is used for data transmission between the edge device and the cloud and the back-end service platform, and reporting of alarm information. The power module is used to ensure stable power supply of the device. The system in this embodiment can use RK3588 as a data processing platform and support NPU model inference.

[0119] In this embodiment, the signal conditioning module completes three-level processing: program-controlled gain amplification (0-60dB adjustable, suitable for weak cavitation noise); band-pass filtering (20Hz-20kHz, suppresses power frequency 50Hz interference); analog-to-digital conversion (16bit resolution, 48kHz sampling rate). The signal processing module is used to analyze the operating condition of the water turbine; the control processing module is used to control the flow operation of the entire system, coordinate the operation of other modules, and integrate intelligent decision logic, bearing data fusion and decision functions. The storage module is used for local storage of data and programs, which can be configured with eMMC flash memory, supporting 64GB circular storage, capable of retaining nearly 30 days of fault data. The field alarm module is used for field alarm warning, such as implementing hierarchical response: 1-2 level fault triggers yellow warning light, 3-5 level fault starts red flashing + 85dB buzzer, suitable for operation and maintenance priority. The power module is used to ensure stable power supply of the system. For example, wide voltage design (input 9-36V DC) is adopted, and through isolated DCDC conversion, independent power supply is provided for analog circuits (5V), digital cores (3.3V) and peripherals (12V), with overvoltage / overcurrent protection; electromagnetic compatibility design (compliant with IEC 61000-4 standard) ensures stable operation of the system in strong electromagnetic interference of the water and power plant. The data transmission module is used for data transmission between the system and the cloud, back-end service platform, and alarm information reporting. For example, industrial Ethernet + SPI dual interface can be used to support Modbus / TCP protocol, so that structured data containing timestamp, fault type, and positioning result can be uploaded to the main control system.

[0120] Exemplary, Figure 4 The signal processing process of the signal processing module in one embodiment is shown in the schematic diagram, taking an 8-microphone array as an example. The steps in this embodiment can include:

[0121] Step S401, acquire sound data (8-way digital signal).

[0122] Step S402, pre-process (1-way digital signal normalization / de-meaning).

[0123] Step S403, extract time-frequency domain features.

[0124] Step S404, determine whether the time-frequency domain features are abnormal (endpoint detection), if yes, execute step S405, if no, return to execute step S401.

[0125] Step S405, extract wavelet time-frequency graph features.

[0126] Step S406, the wavelet time-frequency map features are identified by the inference model (fault identification), whether a fault occurs is judged, if yes, step S407 is executed, if not, step S401 is returned to execute.

[0127] Step S407, the array orientation algorithm is used to determine the fault position.

[0128] Step S408, the device state information is summarized.

[0129] Optionally, the abnormal data and normal data of step S404 and the fault identification data of step S406 are all summarized by step S408.

[0130] In the above embodiment, the multi-channel digital sound signal is analyzed by the signal processing module. Through feature extraction and algorithm analysis of the sound signal, the current running state of the water turbine (normal / abnormal) is determined, and the fault occurrence position (such as blade, bearing, etc. Component) is accurately positioned, and finally the state information is output.

[0131] For example, the multi-channel digital signal is written into the cache for the signal processing module to analyze, and the data stream is queued to perform signal processing in the form of data frames; Select one-way data N1 to do normalization and mean removal operation; Extract time-frequency domain features (mainly 13 time-frequency domain features such as fundamental frequency, fundamental frequency energy ratio, odd-even frequency, high-low frequency ratio, root mean square, peak value, zero-crossing rate, kurtosis, skewness, entropy, margin, instantaneous sound pressure level change rate, and instantaneous voiceprint component change rate) for N1 data; Determine whether an abnormal change is detected by Mahalanobis distance, if an abnormality occurs, extract the wavelet time-frequency map features as the input of the inference model, and output the fault identification result; When a fault occurs, use the array positioning method to calculate the position of the fault point; Finally, the analysis results of the current frame signal are sorted and sent to the main control program through the bus for the main control program to do the next step processing.

[0132] In this embodiment, there is no need to transmit massive original acoustic data to the back end, which can greatly reduce the communication pressure and realize real-time fault diagnosis (millisecond-level response). Due to the directional collection of the microphone array and the collaborative processing of multiple modules, the single sound source positioning ambiguity problem can be solved, the fault position can be accurately identified, and the maintenance efficiency can be improved. In addition, this embodiment covers signal collection, conversion, analysis and health evaluation processes, which can form a closed-loop fault monitoring system to ensure the stable operation of the water turbine and reduce unplanned downtime losses.

[0133] It should be understood that, although the steps in the flowcharts involved in the above embodiments are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the above embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential, but can be alternately executed with other steps or steps or stages in other steps. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by the combination are within the scope of protection of the present application.

[0134] Based on the same inventive concept, the embodiments of the present application also provide a hydraulic turbine fault monitoring device for implementing the above-mentioned hydraulic turbine fault monitoring method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more hydraulic turbine fault monitoring device embodiments provided below can refer to the limitations of the hydraulic turbine fault monitoring method described above, which will not be repeated here.

[0135] In an exemplary embodiment, as shown in Figure 5 A hydraulic turbine fault monitoring device is provided, comprising: a microphone array 501, a signal conditioning module 502, and a signal processing module 503, wherein:

[0136] The microphone array 501 is arranged in the hydraulic turbine room for collecting sound signals generated during the operation of the hydraulic turbine;

[0137] The signal conditioning module 502 is configured to convert the sound signals into corresponding multi-channel digital signals;

[0138] The signal processing module 503 is configured to extract features from the multi-channel digital signals and analyze the extracted feature data to determine the current operating state of the hydraulic turbine; based on the current operating state of the hydraulic turbine and the inherent characteristic parameters of the hydraulic turbine, the health of the hydraulic turbine is evaluated to obtain an evaluation result.

[0139] The signal conditioning module 502 is configured to amplify the collected sound signal through a low-noise amplifier to obtain an amplified signal, filter the amplified signal through a band-pass filter to obtain a signal after noise removal, convert the signal after noise removal into a digital quantity in binary data format through a differential synchronous conversion mode, convert the digital quantity in binary data format into a floating point number, and obtain a plurality of digital signals through frame synchronization mechanism packet processing.

[0140] The signal processing module 503 is configured to write the plurality of digital signals into a cache and queue process in a data frame manner, select any one of the digital signals from the cache for normalization and mean removal processing to obtain a preprocessed signal, perform time-frequency feature extraction on the preprocessed signal to obtain time-frequency feature data, and determine the Mahalanobis distance based on the time-frequency feature data and determine the current operating state of the water turbine according to the Mahalanobis distance threshold. The operating state includes normal or abnormal.

[0141] The signal processing module 503 is further configured to extract corresponding wavelet time-frequency graph features based on the preprocessed signal when the current operating state of the water turbine is abnormal, input the wavelet time-frequency graph features into an inference model to determine whether the water turbine has a fault through the inference model, and determine the fault position of the water turbine based on the plurality of digital signals using an array positioning method when there is a fault. The array positioning method includes decomposing the plurality of digital signals into a plurality of sub-band signals through wavelet packet decomposition, extracting a corresponding signal subspace for each sub-band signal, and determining the azimuth and pitch angle of the sound source in combination with the geometric characteristics of the circular array.

[0142] The signal processing module 503 is configured to evaluate the health of the water turbine according to the current operating state of the water turbine, the equipment parameters of the water turbine, and the historical operation data of the water turbine to obtain an evaluation result, and determine whether to send an alarm information to a superior system or an operation and maintenance device according to the evaluation result. The alarm information includes a fault level and / or a risk level.

[0143] The device can further include a storage module 504 configured to store the sound signal corresponding to the alarm time and mark it as fault data. The fault data is used for backtracking analysis and iterative optimization of the inference model.

[0144] The various modules in the water turbine fault monitoring device can be implemented by software, hardware, or a combination thereof. The various modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so that the processor can call and execute the operations corresponding to the various modules.

[0145] In an exemplary embodiment, a computer device, which can be a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 6 The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. The processor of the computer device is configured 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 running the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, mobile cellular network, Near Field Communication (NFC), or other technologies. The computer program is executed by the processor to implement a water turbine fault monitoring method. The display unit of the computer device is configured to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, a trackball, or a touchpad arranged on the shell of the computer device, or an external keyboard, a touchpad, a mouse, or the like.

[0146] Those skilled in the art can understand that Figure 6 The structure shown in the above description is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0147] In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the following steps:

[0148] The sound signal generated in the operation process of the water turbine is collected through a pre-arranged microphone array; the sound signal is converted into corresponding multi-channel digital signals; feature extraction is performed on the multi-channel digital signals, and analysis is performed based on the extracted feature data to determine the current operating state of the water turbine; and the health degree of the water turbine is evaluated according to the current operating state of the water turbine and the inherent characteristic parameters of the water turbine to obtain an evaluation result.

[0149] In one embodiment, the processor, when executing the computer program, also implements the following steps:

[0150] The collected sound signal is amplified by a low-noise amplifier to obtain an amplified signal; the amplified signal is filtered by a band-pass filter to obtain a signal after noise removal; the signal after noise removal is converted into a binary data format digital quantity by a differential synchronous conversion method; and the binary data format digital quantity is converted into a floating point number and packaged by a frame synchronization mechanism to obtain a multi-channel digital signal.

[0151] In one embodiment, the processor, when executing the computer program, also implements the following steps:

[0152] The multi-channel digital signal is written into a cache and queued in a data frame manner; any one of the digital signals in the cache is selected for normalization and mean removal to obtain a pre-processed signal; time-frequency feature extraction is performed on the pre-processed signal to obtain time-frequency feature data, the time-frequency feature data including at least one of fundamental frequency, fundamental frequency energy proportion, odd-even multiple frequency, high-low frequency ratio, root mean square, peak value, zero-crossing rate, kurtosis, skewness, entropy, margin, instantaneous sound pressure level change rate, and instantaneous voiceprint component change rate; the Mahalanobis distance is determined based on the time-frequency feature data, and the current operating state of the water turbine is determined according to the Mahalanobis distance threshold, the operating state including normal or abnormal.

[0153] In one embodiment, the processor, when executing the computer program, also implements the following steps:

[0154] In the case that the current operating state of the water turbine is abnormal, the wavelet time-frequency graph feature corresponding to the pre-processed signal is extracted; the wavelet time-frequency graph feature is input into an inference model to determine whether the water turbine has a fault through the inference model; in the case that there is a fault, an array positioning method is used to determine the fault position of the water turbine based on the multi-channel digital signal; wherein the array positioning method includes: decomposing the multi-channel digital signal into a plurality of sub-band signals by wavelet packet, extracting a corresponding signal subspace for each sub-band signal, and determining the azimuth and pitch angle of the sound source in combination with the geometric characteristics of the circular array.

[0155] In one embodiment, the processor, when executing the computer program, also implements the following steps:

[0156] According to the current operating state of the water turbine, the equipment parameters of the water turbine, and the historical operation data of the water turbine, the health degree of the water turbine is evaluated to obtain an evaluation result; and according to the evaluation result, it is determined whether to send an alarm information to a superior system or an operation and maintenance terminal device, the alarm information including a fault level and / or a risk degree.

[0157] In one embodiment, the processor also implements the following steps when executing the computer program:

[0158] The sound signal corresponding to the alarm moment is stored and marked as fault data; wherein the fault data is used for backtracking analysis and iterative optimization of the inference model.

[0159] In one embodiment, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method steps in each of the above embodiments.

[0160] In one embodiment, a computer program product is provided, comprising a computer program, and the computer program is executed by a processor to implement the method steps in each of the above embodiments.

[0161] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., but is not limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but is not limited thereto.

[0162] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, it should be considered as the scope of the present application.

[0163] The above embodiments only express several implementation ways of the present application, and the description is specific and detailed, but it should not be understood as a limitation to the patent scope of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method of monitoring a water turbine for failure, the method comprising: The method comprises: acquiring a sound signal generated during operation of a hydraulic turbine through a pre-arranged microphone array; converting the sound signal into corresponding multi-channel digital signals; writing the multi-channel digital signals into a cache and queuing them in the form of data frames; selecting any one of the digital signals from the cache for normalization and mean removal processing to obtain a pre-processed signal; performing time-frequency feature extraction on the pre-processed signal to obtain time-frequency feature data; determining a Mahalanobis distance based on the time-frequency feature data and determining a current operating state of the hydraulic turbine according to a Mahalanobis distance threshold; performing health degree evaluation on the hydraulic turbine according to the current operating state of the hydraulic turbine and inherent characteristic parameters of the hydraulic turbine to obtain an evaluation result.

2. The method of claim 1, wherein, The conversion of the sound signal into corresponding multi-channel digital signals comprises: amplifying the acquired sound signal through a low-noise amplifier to obtain an amplified signal; filtering the amplified signal through a band-pass filter to obtain a signal with removed noise; converting the signal with removed noise into a digital quantity in binary data format through a differential synchronous conversion mode; converting the digital quantity in binary data format into a floating-point number and performing packet processing through a frame synchronization mechanism to obtain multi-channel digital signals.

3. The method of claim 1, wherein, The time-frequency feature data comprises at least one of a fundamental frequency, a fundamental frequency energy proportion, an odd-even multiple frequency, a high-low frequency ratio, a root mean square, a peak value, a zero-crossing rate, a kurtosis, a skewness, an entropy, a margin, an instantaneous sound pressure level change rate, and an instantaneous soundprint component change rate. The operating state comprises normal or abnormal.

4. The method of claim 3, wherein, In the case where the current operating state of the hydraulic turbine is abnormal, the method further comprises: extracting a wavelet time-frequency graph feature corresponding to the pre-processed signal; inputting the wavelet time-frequency graph feature into an inference model to determine whether the hydraulic turbine has a fault through the inference model; in the case where there is a fault, determining a fault position of the hydraulic turbine based on the multi-channel digital signals using an array positioning method; wherein the array positioning method comprises decomposing the multi-channel digital signals into a plurality of sub-band signals through wavelet packet decomposition, extracting a signal subspace corresponding to each sub-band signal, and determining an azimuth angle and a pitch angle of a sound source in combination with the geometric characteristics of a circular array.

5. The method according to any one of claims 1 to 4, characterized in that, The health degree evaluation on the hydraulic turbine according to the current operating state of the hydraulic turbine and the inherent characteristic parameters of the hydraulic turbine to obtain an evaluation result comprises: performing health degree evaluation on the hydraulic turbine according to the current operating state of the hydraulic turbine, device parameters of the hydraulic turbine, and historical operation data of the hydraulic turbine to obtain an evaluation result; determining whether to send alarm information to a superior system or an operation and maintenance end device according to the evaluation result, wherein the alarm information comprises a fault level and / or a risk degree.

6. The method of claim 5, wherein, In the case where it is determined to send alarm information to the superior system or the operation and maintenance end device, the method further comprises: storing a sound signal corresponding to an alarm time and marking it as fault data; wherein the fault data is used for backtracking analysis and iterative optimization of the inference model.

7. A device for monitoring the failure of a hydraulic turbine, characterized in that The device comprises: A microphone array is arranged in a water turbine room for collecting sound signals generated during operation of the water turbine; a signal conditioning module for converting the sound signals into corresponding multi-channel digital signals; a signal processing module for writing the multi-channel digital signals into a cache and processing them in the form of data frames; selecting any one of the digital signals from the cache for normalization and mean removal to obtain a pre-processed signal; performing time-frequency feature extraction on the pre-processed signal to obtain time-frequency feature data; determining Mahalanobis distance based on the time-frequency feature data and determining the current operating state of the water turbine according to a Mahalanobis distance threshold; and performing health degree evaluation on the water turbine according to the current operating state of the water turbine and inherent characteristic parameters of the water turbine to obtain an evaluation result.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images