Water turbine runner discharge noise monitoring and fault identification system and method

Abstract

The invention relates to a water turbine flow channel discharge noise monitoring and fault identification system and method, interference type optical fiber hydrophone arrays are arranged at a top cover between a rotating wheel and a movable guide vane, a volute mandoor and a draft tube straight cone section mandoor, discharge noise in a flow channel is directly measured, and after photoelectric conversion, signal amplification, analog-to-digital conversion and band-pass filtering, the discharge noise in the flow channel is detected. Blind source separation is carried out by using FastICA, features are extracted in combination with empirical mode decomposition (EEMD) and an energy method, audible sound and ultrasonic frequency bands are monitored at the same time, and feature type identification is carried out in combination with an existing fault feature database based on dynamic time warping (DTW) and a support vector machine (SVM) algorithm; the output signal-to-noise ratio and the anti-interference capability of the system are improved, multi-directional analysis of discharge noise characteristics is achieved, the monitoring range is wide, the fault recognition speed is high, the accuracy is high, and the system is mainly used for monitoring the running state of the water turbine in real time, finding and early warning faults in time and improving the running maintenance efficiency of the water turbine.

Description

Technical Field

[0001] The present invention relates to the technical field of motor drive systems and their radiated interference analysis, and particularly to a system and method for monitoring the discharge noise and fault identification in a water turbine flow passage. Background Art

[0002] In the operation and maintenance of hydropower stations, the water turbine is a key equipment, and the real-time monitoring and fault identification of its working state are crucial. The noise monitoring and fault identification technology in the water turbine flow passage is an important means to improve the operation efficiency of the equipment and prevent safety accidents. However, there are many deficiencies in the existing monitoring methods and urgent improvements are needed.

[0003] Traditional water turbine fault monitoring methods mainly rely on monitoring a single vibration signal from the outside, which belongs to contact monitoring and has problems such as low measurement accuracy, slow monitoring feedback, and large identification errors. In addition, due to the large scale and complex structure of the water turbine flow passage, it is difficult to repair, and traditional monitoring methods often fail to achieve real-time monitoring of the state of internal working components during the operation of the water turbine. Therefore, there is an urgent need for a technical solution that can efficiently and accurately monitor the noise in the water turbine flow passage and effectively identify faults.

[0004] Currently, some improved methods for water turbine noise monitoring and fault identification have been proposed. For example, CN119357790A discloses a method for identifying the operation faults of a water turbine guide vane mechanism based on deep learning. This method processes the vibration signal data of the guide vane mechanism, constructs a data set, and uses a deep learning model for training to achieve the identification of the operation faults of the guide vane mechanism. However, this method mainly focuses on the vibration signal of the guide vane mechanism, and the noise sources in the water turbine flow passage are diverse, including not only vibration signals but also various factors such as water flow, cavitation and erosion. Therefore, this method still needs to be improved in terms of comprehensiveness and accuracy.

[0005] Specifically, although the method disclosed in CN119357790A constructs a one-dimensional CNN network structure, integrates the time-domain and frequency-domain features of the time-history signal, and considers the influence of time on the signal, it may not be able to fully capture all key information when processing complex noise signals in the water turbine flow passage. Especially when the noise signal contains multiple frequency components and there is mutual interference between the components, the identification effect of this method may be limited.

[0006] In addition, existing noise monitoring and fault identification systems also have deficiencies in aspects such as the layout of measurement points in the water turbine runner passage, signal acquisition, and signal processing. For example, some systems use a single measurement point or have an unreasonable layout of measurement points, resulting in an inability to comprehensively reflect the noise characteristics in the runner passage; some systems fail to effectively filter out external interference during the signal acquisition and processing process, leading to a reduction in measurement accuracy; and some systems lack innovation in the fault identification algorithm, resulting in the identification speed and accuracy being unable to meet actual requirements.

[0007] In summary, existing water turbine noise monitoring and fault identification technologies have deficiencies in measurement accuracy, monitoring feedback speed, identification accuracy, as well as system layout, signal acquisition, and signal processing. Therefore, there is an urgent need for a technical solution that can overcome the above shortcomings and achieve efficient and accurate monitoring and fault identification of the leakage noise in the water turbine runner passage. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a system and method for monitoring the leakage noise and identifying faults in the water turbine runner passage, and to solve the technical problems existing in the real-time monitoring of the working component status in the water turbine runner passage. Especially in the context where the traditional method of externally monitoring a single vibration signal has low measurement accuracy, slow monitoring feedback, and large identification errors, as well as the problem of low efficiency in separately conducting ultrasonic detection or shutdown maintenance and troubleshooting.

[0009] To solve the above technical problems, the technical solution adopted by the present invention is: a system for monitoring the leakage noise and identifying faults in the water turbine runner passage, including a signal acquisition and conversion system, which is arranged at the top cover between the runner and the movable guide vane in the water turbine runner passage, at the access door of the spiral case, and at the access door of the straight cone section of the draft tube, receives the noise and converts it into an optical signal, the amplification and filtering system receives the optical signal and converts it into an electrical signal, amplifies and filters it, the signal separation and feature extraction system separates and extracts features from the filtered signal, and the fault identification system identifies the feature type based on the algorithm and database to achieve fault early warning.

[0010] In a preferred solution, the signal acquisition and conversion system uses an interferometric fiber optic hydrophone array to receive the vibration of the noise sound wave, converts the noise signal into an optical signal and transmits it, and the interferometric fiber optic hydrophone has specific models and specifications, and its monitoring frequency range, dynamic range, and sensitivity all meet the preset standards.

[0011] In a preferred solution, the signal amplification and filtering system receives the optical signal from the signal acquisition and conversion system, converts it into an electrical signal, amplifies the signal through a differential amplifier, then converts the electrical signal into a digital signal through an analog-to-digital converter, and then retains the noise signals in the frequency bands of 20 Hz to 20 kHz and 20 kHz to 70 kHz through a band-pass filter, and suppresses the reverberation interference in the water turbine cavity through an inverse filter; In a preferred embodiment, the band-pass filter in the signal amplification and filtering system is a FIR (Finite Impulse Response) type or Butterworth type band-pass filter, and high-performance models are selected for both the differential amplifier and the analog-to-digital converter to ensure accurate amplification and conversion of the signal.

[0012] In a preferred embodiment, the signal separation and feature extraction system performs FastICA (Fast Independent Component Analysis) blind source separation on the filtered digital signal, extracts the time-frequency domain features of the noise signal, including Mel-Frequency Cepstral Coefficients (MFCC), center frequency, root mean square frequency, peak-to-peak value features, and outputs a visual Mel spectrogram.

[0013] In a preferred embodiment, the signal separation and feature extraction system further includes steps of performing Fast Fourier Transform (FFT) and Wavelet Packet Transform (WPT) on the noise signal to extract richer time-frequency domain features.

[0014] In a preferred embodiment, the fault identification system is based on the Dynamic Time Warping (DTW) and Support Vector Machine (SVM) algorithms, and combines with the existing fault feature database to identify the feature types of the monitored noise signal for fault early warning.

[0015] In a preferred embodiment, the fault identification system further includes a backpropagation algorithm for optimizing partial parameter weights or updating the matching algorithm according to the results of fault identification to improve the identification accuracy.

[0016] In a preferred embodiment, the system further includes a data storage and management system, which is connected to the signal separation and feature extraction system and the fault identification system, and is used to store the collected noise signals, the extracted feature data, and the fault identification results for subsequent analysis and processing.

[0017] A method for monitoring the discharge noise and identifying faults in a water turbine runner passage adopts a water turbine runner passage discharge noise monitoring and fault identification system described in any one of the above, and the method includes the following steps: Step1: Receive the vibration of the noise sound wave through an interferometric fiber optic hydrophone array arranged in the water turbine runner passage, and convert the noise signal into an optical signal for transmission; Step 2: Convert the optical signal into an electrical signal, amplify the signal, and convert it into a digital signal; Step 3: Perform band-pass filtering and inverse filtering on the digital signal, retain the noise signal in the specified frequency band, and suppress reverberation interference; Step 4: Perform FastICA blind source separation on the filtered signal to extract the characteristics of the noise signal; Step 5: Based on the dynamic time warping (DTW) and support vector machine (SVM) algorithms, combined with the fault feature database, identify the extracted features to achieve fault early warning.

[0018] In the preferred solution, the method further includes preprocessing the extracted feature data to improve the accuracy and efficiency of subsequent fault identification.

[0019] In the preferred solution, the method further includes regularly updating the fault feature database to reflect the latest fault types and characteristics, thereby improving the adaptability and accuracy of fault identification.

[0020] In the preferred solution, the method further includes classifying and prioritizing the identified fault types so as to take corresponding maintenance measures in a timely manner to ensure the safe and stable operation of the water turbine.

[0021] A water turbine runner discharge noise monitoring and fault identification system and method provided by the present invention have the following beneficial effects: 1. The interferometric fiber optic hydrophone array adopted by the present invention significantly improves the signal-to-noise ratio and anti-interference ability of the system output, effectively reduces the crosstalk phenomenon between signals, and is particularly suitable for environmental monitoring in complex and extensive environments such as water turbine runners, ensuring the accuracy and reliability of monitoring data; 2. The present invention adopts advanced signal processing algorithms, and adopts a strategy of amplifying first and then filtering in signal processing to ensure that the target frequency signal will not be misfiltered. At the same time, an FIR or Butterworth type band-pass digital filter with small distortion and flexible design is introduced to further improve the control ability of the amplitude response and the fidelity effect of the signal waveform; 3. The present invention uses an efficient algorithm combination to achieve noise signal separation. By combining FastICA with advanced algorithms such as EEMD and STE, efficient separation and feature extraction of noise signals are achieved, which not only improves the accuracy of fault identification, but also significantly speeds up the identification speed, providing strong support for real-time monitoring and fault early warning; 4. The present invention adopts an improved algorithm combining DTW and SVM in fault matching. Compared with the traditional softmax algorithm, this combination has greater discrimination and faster processing speed in fault matching, providing more accurate results for fault early warning and diagnosis; 5. The present invention realizes the real-time monitoring of the leakage noise in the water turbine flow passage, and constructs a remote fault warning system based on feature extraction and fault identification algorithms, which can timely detect potential faults, provide warnings for maintenance and repair work, and greatly reduce the downtime and maintenance costs; 6. The present invention doubles the monitoring accuracy and feedback speed. By combining an interferometric fiber optic hydrophone array and advanced signal processing algorithms, it effectively solves the problems of traditional monitoring methods such as low accuracy and slow feedback speed. It not only improves the monitoring accuracy but also significantly speeds up the feedback speed, providing a strong guarantee for the stable operation of the water turbine; 7. The present invention scientifically plans the selection of monitoring points and arranges the monitoring points at positions with obvious noise and good analyzability of characteristics, which can truly reflect the characteristics of the leakage noise in the flow passage and provide more accurate data support for fault identification and warning; 7. The present invention has a wide frequency coverage range in fault identification, including 20 Hz to 20 kHz and 20 kHz to 70 kHz. It not only covers the identification of basic faults but also supports the monitoring of special fault phenomena such as cavitation and Karman vortices, providing strong support for the comprehensive operation and maintenance of the water turbine; 9. Through the application of the real-time monitoring and fault warning system, the present invention significantly improves the operation efficiency and reliability of the water turbine, reduces the downtime and maintenance costs caused by faults, and provides a strong guarantee for the stable operation and economic benefits of the hydropower station; 10. The present invention combines a variety of advanced algorithms and technologies to form a unique comprehensive technical advantage, which is not only reflected in the efficient processing of noise signals and the accuracy of fault identification, but also in the stability and scalability of the system. With the continuous development of technology and the continuous expansion of application fields, the present invention has broad application prospects and market potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flowchart of the working method of the identification system of the present invention; Figure 2 is a schematic diagram of the layout of the noise measurement system of the present invention; Figure 3 is a schematic diagram of the installation of the interferometric fiber optic hydrophone of the present invention; Figure 4 is a schematic diagram of the dimensions of the interferometric fiber optic hydrophone and its sleeve of the present invention; Figure 5 is an enlarged view of the hydrophone probe of the present invention; In the figure: signal acquisition and conversion system 1, signal amplification and filtering system 2, signal separation and feature extraction system 3, fault identification system 4, top cover 5, volute access door 6, draft tube straight cone section access door 7, sleeve 8, installation wall surface 9, sealant 10, rubber sleeve 11, fiber optic hydrophone 12, in-flow channel flange 13. Specific embodiments

[0023] The technical solutions in the present invention will be further described below in conjunction with the accompanying drawings and embodiments: Embodiment 1 As Figure 1 shown, a water turbine flow channel discharge noise monitoring and fault identification system includes a signal acquisition and conversion system 1, which is arranged at the top cover 5, volute access door 6, and draft tube straight cone section access door 7 between the runner and the movable guide vane of the water turbine flow channel, receives noise and converts it into an optical signal. The signal amplification and filtering system 2 receives the optical signal and converts it into an electrical signal, amplifies and filters it. The signal separation and feature extraction system 3 separates and extracts the features of the filtered signal. The fault identification system 4 identifies the feature type based on algorithms and databases to achieve fault warning.

[0024] In this embodiment, the signal acquisition and conversion system 1 uses an interferometric fiber optic hydrophone array to receive the vibration of the noise sound wave, converts the noise signal into an optical signal and transmits it. The interferometric fiber optic hydrophone has specific models and specifications, and its monitoring frequency range, dynamic range, and sensitivity all meet the preset standards.

[0025] Furthermore, the signal amplification and filtering system 2 receives the optical signal from the signal acquisition and conversion system 1, converts it into an electrical signal, amplifies the signal through a differential amplifier, then converts the electrical signal into a digital signal through an analog-to-digital converter, and then retains the noise signals in the frequency bands of 20 Hz to 20 kHz and 20 kHz to 70 kHz through a band-pass filter, and suppresses the reverberation interference in the water turbine cavity through an inverse filter.

[0026] Furthermore, the band-pass filter in the signal amplification and filtering system 2 is an FIR type or Butterworth type band-pass filter, and high-performance models are selected for both the differential amplifier and the analog-to-digital converter to ensure the accurate amplification and conversion of the signal.

[0027] Furthermore, the signal separation and feature extraction system 3 performs FastICA blind source separation on the filtered digital signal, extracts the time-frequency domain features of the noise signal, including Mel frequency cepstral coefficients MFCC, centroid frequency, root mean square frequency, peak-to-peak value features, and outputs a visualized Mel spectrogram.

[0028] Further, the signal separation and feature extraction system 3 further includes steps of performing fast Fourier transform (FFT) and wavelet packet transform (WPT) on the noise signal to extract richer time-frequency domain features.

[0029] Further, the fault identification system 4 is based on the dynamic time warping (DTW) and support vector machine (SVM) algorithms, and combines with the existing fault feature database to identify the feature types of the monitored noise signal to achieve fault warning.

[0030] Further, the fault identification system 4 further includes a backpropagation algorithm for optimizing some parameter weights or updating the matching algorithm according to the results of fault identification to improve the identification accuracy.

[0031] Further, the system further includes a data storage and management system, which is connected to the signal separation and feature extraction system 3 and the fault identification system 4, and is used to store the collected noise signals, the extracted feature data, and the fault identification results for subsequent analysis and processing.

[0032] Embodiment 2 As Figures 2 to 5 shown, a method for monitoring the discharge noise and identifying faults in a water turbine runner adopts the water turbine runner discharge noise monitoring and fault identification system described in Embodiment 1, and includes the following steps: Step1: Receive the vibration of the noise sound wave through the interferometric fiber optic hydrophone array arranged in the water turbine runner, and convert the noise signal into an optical signal and transmit it out; Step2: Convert the optical signal into an electrical signal, amplify the signal, and convert it into a digital signal; Step3: Perform band-pass filtering and inverse filtering on the digital signal to retain the noise signal in the specified frequency band and suppress the reverberation interference; Step4: Perform FastICA blind source separation on the filtered signal to extract the features of the noise signal; Step5: Based on the dynamic time warping (DTW) and support vector machine (SVM) algorithms, and in combination with the fault feature database, identify the extracted features to achieve fault warning.

[0033] In this embodiment, the method further includes preprocessing the extracted feature data to improve the accuracy and efficiency of subsequent fault identification.

[0034] Further, the method further includes regularly updating the fault feature database to reflect the latest fault types and features, thereby improving the adaptability and accuracy of fault identification.

[0035] Further, the method further includes classifying and prioritizing the identified fault types so as to take corresponding maintenance measures in a timely manner to ensure the safe and stable operation of the water turbine.

[0036] Embodiment 3 In another preferred embodiment, on the basis of the above Embodiments 1 and 2, as Figures 1 to 5 shown, a water turbine runner passage discharge noise monitoring and fault identification system provided in this embodiment mainly includes four major parts: a signal acquisition and conversion system 1, a signal amplification and filtering system 2, a signal separation and feature extraction system 3, and a fault identification system 4.

[0037] I. Signal acquisition and conversion system 1 The signal acquisition system 1 is arranged at key positions in the water turbine runner passage, specifically including at the top cover 5 between the runner and the movable guide vane, at the volute access door 6, and at the draft tube straight cone section access door 7. At these positions, we installed an interferometric fiber optic hydrophone array for collecting noise signals. The fiber optic hydrophone model is XARION's Eta 250 L Ultra, with a sensing head diameter of 6.5 mm, a length of 34 mm. The monitoring frequency range of the fiber optic hydrophone 12 is 10 Hz to 2 MHz, the dynamic range is 1 mPa to 10 kPa, the sensitivity is 3.4 mV / Pa (0 dB gain, impedance 50 Ω), the power < 50 W, and the operating temperature is 5°C to 60°C (40°F to 140°F).

[0038] The installation method of the fiber optic hydrophone is as follows: First, put a sleeve 8 made of cast steel with a diameter of 10 mm and a length of 105 mm on the upper part of the fiber optic hydrophone probe. A trapezoidal hole slightly larger than the outer shape of the fiber optic hydrophone is opened in the sleeve 8 for fixing the fiber optic hydrophone. After the fiber optic hydrophone is put on the rubber sleeve 11, it is stuffed into the sleeve 8 for sealing and fixing. Then, a hole with a diameter slightly larger than the diameter of the sleeve 8 is opened on the installation wall surface 9 of the fiber optic hydrophone, the sleeve 8 is inserted into the opening, and a special metal sealant 10 is used to fill the gap between the outer wall of the sleeve 8 and the inner wall of the opening on the installation wall surface 9. Finally, the fixed flange outside the runner and the inner flange 13 in the runner on the sleeve 8 are welded, and the weld seam is brushed with a special sealant 10 to ensure the sealing performance.

[0039] At the top cover 5, we randomly arranged four monitoring points to form an array to improve the comprehensiveness and accuracy of noise collection. At the volute access door 6 and the draft tube straight cone section access door 7, one monitoring point is arranged respectively.

[0040] The collected noise signal is transmitted out to the signal conversion system in the central control room in the form of an optical signal through an optical fiber. The signal conversion system consists of an optoelectronic converter and an analog-to-digital converter (ADC, Analog-to-Digital Converter). The optoelectronic converter converts the optical signal into an electrical signal. The differential amplifier selects Texas Instruments INA597e-trim applicable to the optoelectronic module to suppress the common-mode signal and improve the output quality. The analog-to-digital converter selects Texas Instruments ADS127L1x 512kSPS eight-channel synchronous sampling 24-bit ADC to convert the electrical signal into discrete digital signals for subsequent processing.

[0041] II. Signal Amplification and Filtering System 2 The signal amplification and filtering system 2 is connected to the differential amplifier and the analog-to-digital converter (ADC). We use a fourth-order FIR band-pass filter of the DigiKey IP4252CZ16-8-TTL series to filter the digital signal. This filter can retain the noise signals in the frequency bands of 20Hz - 20kHz and 20kHz - 70kHz, while weakening the interference signals in other frequency bands. In addition, a high-order inverse filter is also connected to weaken the reverberation interference in the turbine cavity. The filtered signal is transmitted to a computer for further processing.

[0042] III. Signal Separation and Feature Extraction System 3 The signal separation and feature extraction system 3 implemented on the computer uses an improved algorithm to process the filtered signal. First, the FastICA algorithm is used for blind source separation of the signal. The FastICA algorithm includes steps such as mean removal, signal normalization, and reduction of the main frequency components in dimension. When the non-Gaussianity measure during the operation reaches the maximum, it is considered that the separation has been completed and the original signal is output.

[0043] Then, the separated signal is further processed by combining the empirical mode decomposition (EMD) algorithm and the short-term energy feature (STE). Vectors with similar intrinsic mode functions (IMFs) are grouped into a vector set for secondary feature aggregation and feature extraction. By calculating and extracting multiple features such as Mel-frequency cepstral coefficients (MFCC), center frequency, root mean square of frequency, and peak-to-peak value of the monitored discharge noise signal, the characteristics of the noise signal can be more comprehensively characterized. At the same time, a Mel spectrogram can also be output to visualize the corresponding discharge noise characteristics.

[0044] IV. Fault Identification System 4 The fault identification system 4 is implemented by a computer based on an improved algorithm that combines dynamic time warping (DTW) and support vector machine (SVM). This algorithm identifies features such as MFCC, center frequency, root mean square frequency, and peak-to-peak value extracted in the signal separation and feature extraction system. Based on the existing fault signal feature database, the support vector machine can find an optimal hyperplane to separate the feature space composed of abnormal noise feature vectors and noise feature vectors during normal operation, thereby determining whether there is a fault in this section of noise and the type of fault.

[0045] Finally, the system model is gradually updated through the backpropagation algorithm to improve the recognition accuracy. According to the actual solution measures finally taken, some parameter weights are optimized or the matching algorithm is updated to continuously improve the performance of the system.

[0046] V. The working method of this embodiment includes the following steps: 1. Collect noise signals through the fiber optic hydrophone 12 array arranged at key positions in the water turbine flow passage; 2. Transmit the collected noise signals in the form of optical signals through the optical fiber to the signal conversion system in the central control room; 3. In the signal conversion system, convert the optical signal into an electrical signal, and perform differential amplification and analog-to-digital conversion; 4. In the signal amplification and filtering system 2, perform band-pass filtering and inverse filtering on the converted digital signal; 5. In the signal separation and feature extraction system 3 implemented on the computer, perform blind source separation and feature extraction on the filtered signal; 6. In the fault identification system 4, perform fault identification on the extracted features based on an improved algorithm that combines dynamic time warping (DTW) and support vector machine (SVM); 7. Take corresponding solution measures according to the recognition results, and optimize the system model to improve the recognition accuracy.

[0047] Through the above steps, this embodiment can realize real-time monitoring and fault identification of the leakage flow noise in the water turbine flow passage, and provide help for the maintenance work and long-term good operation of the water turbine.

[0048] Embodiment 4 In another preferred embodiment, based on the above Embodiment 3, this embodiment further details the implementation of a water turbine flow passage leakage flow noise monitoring and fault identification system and its method, which is specifically described as follows: I. Arrangement of the signal acquisition and conversion system 1: As Figure 2As shown in the figure, the signal acquisition system is mainly arranged at three key positions in the water turbine flow passage: at the top cover 5 between the runner and the movable guide vane, at the volute access door 6, and at the draft tube straight cone section access door 7. These positions are selected as monitoring points because they are located in areas with obvious water flow noise, good analyzability of noise characteristics, and high representativeness, and can more accurately reflect the discharge noise characteristics in the flow passage.

[0049] At the top cover 5, we randomly arranged four monitoring points, and these monitoring points form an array to improve the comprehensiveness and accuracy of noise acquisition. At the volute access door 6 and the draft tube cone section access door, we arranged one monitoring point respectively.

[0050] The interferometric fiber optic hydrophone 12 used at the monitoring point is of the XARION's Eta 250 L Ultra model, which has high precision and high sensitivity and can capture weak noise signals in the flow passage. The installation of the hydrophone is as Figures 3 to 5 shown, and it is fixed on the inner wall of the water turbine flow passage through a special sleeve 8 and sealant 10 to ensure the stability and reliability of signal transmission.

[0051] The signal conversion system consists of an optical-electric converter and an analog-to-digital converter (ADC), and these devices are installed in the central control room. The optical-electric converter converts the optical signal transmitted by the optical fiber into an electrical signal, then the signal is amplified through a differential amplifier, and finally the analog electrical signal is converted into discrete digital signals by the analog-to-digital converter for subsequent processing and analysis.

[0052] II. Realization of the signal amplification and filtering system 2: The signal amplification and filtering system 2 is connected after the signal conversion system and is used to amplify and filter the acquired digital signals. This system adopts a fourth-order FIR type band-pass filter of the DigiKey IP4252CZ16-8-TTL series. This filter has the advantages of small distortion and flexible design, and can effectively retain the noise signals in the frequency bands of 20Hz - 20kHz and 20kHz - 70kHz, while suppressing the interference signals in other frequency bands.

[0053] In addition, a reverse filter is also connected to the system to weaken the reverberation interference in the water turbine cavity and further improve the purity and signal-to-noise ratio of the signal.

[0054] III. Realization of the signal separation and feature extraction system 3: The signal separation and feature extraction system 3 is implemented on a computer. This system first uses the FastICA algorithm to perform blind source separation on the acquired noise signals to remove the interference of background noises such as mechanical, electrical, and resonance noises. The FastICA algorithm calculates the non-Gaussianity measure of the signals, gradually iteratively optimizes the separation matrix, and finally obtains mutually independent source signals.

[0055] The separated signal will be subjected to feature extraction processing. This system adopts an improved algorithm combining ensemble empirical mode decomposition (EEMD), short-time energy (STE) and energy ratio to calculate and extract multiple features of the signal, such as mel-frequency cepstral coefficients (MFCC), center frequency, root mean square of frequency, peak-to-peak value, etc. At the same time, the system can also output a visual mel spectrogram so that the staff can more intuitively understand the characteristics of the noise signal.

[0056] IV. Implementation of the fault identification system 4: The fault identification system 4 is also implemented on a computer. Based on an improved algorithm combining dynamic time warping (DTW) and support vector machine (SVM), it conducts fault identification and classification on the extracted noise features. First, the SVM algorithm finds an optimal hyperplane to separate the feature space composed of abnormal noise feature vectors and noise feature vectors during normal operation according to the existing fault feature database. Then, the DTW algorithm is used to calculate the similarity between the noise signal to be identified and each fault type in the fault feature database, so as to determine the fault type of the noise signal to be identified.

[0057] Finally, the system uses the backpropagation algorithm to gradually update the model parameters and weights to improve the accuracy and robustness of fault identification. When an abnormal noise signal is identified, the system will issue a warning prompt in time so that the staff can take corresponding measures for repair and maintenance.

[0058] Through the above specific implementation manners, the water turbine runner discharge noise monitoring and fault identification system of the present invention can achieve high-precision and high-efficiency noise monitoring and fault identification functions, providing a strong guarantee for the long-term stable operation of the water turbine.

[0059] In a preferred solution, the signal acquisition and conversion system 1 uses an interferometric fiber optic hydrophone 12 array to receive the vibration of the noise sound wave, converts the noise signal into an optical signal and transmits it. The interferometric fiber optic hydrophone 12 has specific models and specifications, and its monitoring frequency range, dynamic range and sensitivity all meet the preset standards; the above settings ensure that the system can accurately capture underwater noise and improve the accuracy of data acquisition; at the same time, the system is equipped with a high-performance optoelectronic converter that can quickly convert the optical signal into an electrical signal for subsequent signal processing and analysis.

[0060] In a preferred solution, the signal amplification and filtering system 2 receives the optical signal from the signal acquisition and conversion system 1, converts it into an electrical signal, amplifies the signal through a differential amplifier, then converts the electrical signal into a digital signal through an analog-to-digital converter. After that, a band-pass filter is used to retain the noise signals in the frequency bands of 20 Hz to 20 kHz and 20 kHz to 70 kHz, and a reverse filter is used to suppress the reverberation interference in the turbine cavity. With the above settings, the signal quality is effectively improved, ensuring the accuracy of the monitoring data. Then, the digital signal processor further processes and analyzes the filtered signal, identifies the key operating parameters, and provides strong support for the condition monitoring and fault diagnosis of the turbine.

[0061] In a preferred solution, the band-pass filter in the signal amplification and filtering system 2 is an FIR-type or Butterworth-type band-pass filter, and high-performance models are selected for both the differential amplifier and the analog-to-digital converter to ensure the accurate amplification and conversion of the signal. With the above settings, it is aimed at effectively suppressing noise interference and improving signal quality. At the same time, the system adopts advanced digital signal processing algorithms to further optimize the signal characteristics and ensure the stability and reliability of data transmission.

[0062] In a preferred solution, the signal separation and feature extraction system 3 performs FastICA blind source separation on the filtered digital signal, extracts the time-frequency domain features of the noise signal, including Mel-frequency cepstral coefficients MFCC, center frequency, root mean square frequency, peak-to-peak feature, and outputs a visual Mel spectrogram. With the above settings, the accuracy and efficiency of signal processing are further improved, enabling the system to accurately identify the target signal in a complex noise environment, providing solid data support for subsequent signal processing and decision-making analysis, and significantly enhancing the robustness and practicality of the entire system.

[0063] In a preferred solution, the signal separation and feature extraction system 3 also includes steps of performing fast Fourier transform FFT and wavelet packet transform WPT on the noise signal to extract richer time-frequency domain features. With the above settings, the system can more accurately identify and separate the target signal and the noise signal, improve the accuracy and efficiency of signal processing, and provide more reliable data support for subsequent signal analysis and decision-making.

[0064] In a preferred solution, the fault identification system 4 is based on the dynamic time warping DTW and support vector machine SVM algorithms, combines with the existing fault feature database, and identifies the feature types of the monitored noise signal to achieve fault early warning. With the above settings, the accuracy and timeliness of fault identification can be greatly improved, reducing false alarms and missed alarms, providing strong technical support for the stable operation of the equipment, and also providing an important reference basis for subsequent fault handling and maintenance work.

[0065] In a preferred embodiment, the fault identification system 4 further includes a backpropagation algorithm, which is used to optimize some parameter weights or update the matching algorithm according to the results of fault identification to improve the identification accuracy. With the above settings, the fault identification system 4 can continuously learn and evolve by itself to adapt to more complex fault situations. At the same time, the system also has a real-time monitoring function. Once an anomaly is detected, the alarm mechanism will be triggered immediately to ensure that the problem is handled in a timely manner.

[0066] In a preferred embodiment, the system further includes a data storage and management system, which is connected to the signal separation and feature extraction system 3 and the fault identification system 4, and is used to store the collected noise signals, the extracted feature data, and the fault identification results for subsequent analysis and processing. With the above settings, not only the integrity and traceability of the data are ensured, but also the efficiency and accuracy of data analysis are greatly improved, providing strong data support for the preventive maintenance of the equipment and the rapid response to faults.

[0067] In a preferred embodiment, the method further includes preprocessing the extracted feature data to improve the accuracy and efficiency of subsequent fault identification. With the above settings, the feature data is optimized by means of noise removal, normalization, and dimensionality reduction to ensure the quality of the model input, thereby enhancing the robustness and real-time response ability of the fault classification algorithm and laying a solid foundation for subsequent intelligent diagnosis.

[0068] In a preferred embodiment, the method further includes regularly updating the fault feature database to reflect the latest fault types and features, thereby improving the adaptability and accuracy of fault identification. With the above settings, it can be ensured that when the system faces new types of faults, it can quickly match and respond, reducing the losses caused by the delay of fault identification. At the same time, by continuously optimizing the algorithm, the efficiency and intelligent level of fault handling are further improved.

[0069] In a preferred embodiment, the method further includes classifying and prioritizing the identified fault types so as to take corresponding maintenance measures in a timely manner to ensure the safe and stable operation of the water turbine. With the above settings, the efficiency and accuracy of fault handling can be effectively improved. At the same time, the system can automatically trigger a warning to notify relevant personnel, realizing a rapid response to fault handling and further ensuring the overall operation efficiency and safety of the hydropower station.

[0070] In summary, the present invention proposes an innovative system and method for monitoring the discharge noise and identifying faults in the water turbine flow channel. This solution addresses the technical problems faced by the real-time monitoring of the working component status in the water turbine flow channel, especially the low measurement accuracy, slow monitoring feedback, large identification errors in traditional monitoring methods, and the low efficiency of separate ultrasonic detection or shutdown inspection and troubleshooting. It provides a new solution.

[0071] First, the present invention first introduces an interferometric fiber optic hydrophone 12 array as a monitoring tool in the field of turbine runner monitoring. The application of this technology not only significantly improves the monitoring accuracy and anti-interference ability, but also effectively reduces the crosstalk phenomenon between signals, providing reliable technical support for the real-time monitoring of the operating state of the turbine. Through this innovative monitoring tool, the present invention successfully breaks through the limitations of traditional monitoring methods and realizes the accurate capture and real-time monitoring of the leakage flow noise in the turbine runner.

[0072] Secondly, in terms of signal processing, the present invention introduces a variety of advanced algorithms and technologies, such as the combination of amplification first and then filtering, FIR or Butterworth type band-pass digital filters, FastICA and EEMD, STE and other algorithms, as well as the improved algorithms of DTW and SVM. These algorithms and technologies show significant advantages in noise signal separation, feature extraction and fault matching, and can achieve the efficient processing of noise signals and the improvement of the accuracy of fault identification. Through these innovative algorithms and technologies, the present invention not only improves the speed and accuracy of fault identification, but also reduces the complexity and cost of the system, providing strong guarantee for the stable operation of the turbine.

[0073] In addition, the present invention further improves the monitoring accuracy and reliability by scientifically planning the monitoring point positions and widely covering the fault identification frequency range. The scientific layout of the monitoring points can truly reflect the characteristics of the leakage flow noise in the runner, providing more accurate data support for fault identification and early warning. At the same time, the widely covered fault identification frequency range enables the present invention to achieve accurate identification of various faults, including basic faults and special fault phenomena such as cavitation erosion and Karman vortices, providing strong support for the comprehensive operation and maintenance of the turbine.

[0074] Generally speaking, the present invention successfully breaks through the limitations of traditional turbine runner monitoring methods by introducing an interferometric fiber optic hydrophone 12 array and advanced signal processing algorithms, as well as scientifically planning the monitoring point positions and widely covering the fault identification frequency range, and realizes the efficient processing of noise signals and the improvement of the accuracy of fault identification. This innovative solution not only improves the monitoring accuracy and reliability of the turbine operating state, but also provides strong support for the comprehensive operation and maintenance of the turbine, having important application value and promotion prospects.

Claims

1. A turbine flow passage discharge noise monitoring and fault identification system, characterized in that: The invention comprises a signal acquisition and conversion system (1), which is arranged at the top cover (5) between the runner and the movable guide vane of the turbine flow channel, the volute entry door (6) and the tailwater pipe straight cone section entry door (7), receives noise and converts it into an optical signal, an amplification and filtering system (2) receives the optical signal and converts it into an electrical signal, amplifies and filters it, a signal separation and feature extraction system (3) separates and extracts features from the filtered signal, and a fault identification system (4) identifies the feature type based on an algorithm and a database to realize fault warning.

2. A turbine flow passage leakage noise monitoring and fault identification system according to claim 1, characterized in that: The signal acquisition and conversion system (1) uses an array of interference-type optical fiber hydrophones (12) to receive the vibration of noise sound waves, convert the noise signals into optical signals and transmit them out. The interference-type optical fiber hydrophones (12) have specific models and specifications, and their monitoring frequency range, dynamic range and sensitivity all meet preset standards.

3. A turbine flow passage leakage noise monitoring and fault identification system according to claim 2, characterized in that: The signal amplification and filtering system (2) receives the optical signal from the signal acquisition and conversion system (1), converts the optical signal into an electrical signal, amplifies the signal through a differential amplifier, and then converts the electrical signal into a digital signal through an analog-to-digital converter. The noise signal in the frequency bands of 20 Hz to 20 kHz and 20 kHz to 70 kHz is then retained through a bandpass filter, and the reverberation interference in the turbine cavity is suppressed through an inverse filter. The bandpass filter in the signal amplification and filtering system (2) is a FIR type or a Butterworth type bandpass filter.

4. A turbine flow passage leakage noise monitoring and fault identification system according to claim 3, characterized in that: The signal separation and feature extraction system (3) performs FastICA blind source separation on the filtered digital signal, extracts the time-frequency domain features of the noise signal, including Mel-frequency cepstral coefficients MFCC, centroid frequency, frequency root mean square, and peak-to-peak features, and outputs a visualized Mel-frequency spectrum.

5. A turbine flow passage leakage noise monitoring and fault identification system according to claim 4, characterized in that: The signal separation and feature extraction system (3) further comprises the steps of performing fast Fourier transform (FFT) and wavelet packet transform (WPT) on the noise signal to extract richer time-frequency domain features.

6. A turbine flow passage leakage noise monitoring and fault identification system according to claim 5, characterized in that: The fault identification system (4) is based on dynamic time warping (DTW) and support vector machine (SVM) algorithms, combined with an existing fault feature database, to identify the feature type of the monitored noise signal to achieve fault early warning.

7. A turbine flow passage leakage noise monitoring and fault identification system according to claim 6, characterized in that: The fault identification system (4) also includes a back propagation algorithm, which is used to optimize some parameter weights or update the matching algorithm according to the result of the fault identification, so as to improve the identification accuracy.

8. A turbine flow passage leakage noise monitoring and fault identification system according to claim 7, characterized in that: The system also includes a data storage and management system connected to the signal separation and feature extraction system (3) and the fault identification system (4) and used to store the collected noise signals, extracted feature data and fault identification results for subsequent analysis and processing.

9. A method for monitoring the flow noise of a turbine flow passage and identifying faults, characterized in that: A system for monitoring flow noise and identifying faults in a turbine flow passage according to any one of claims 1 to 8 is used, and the method comprises the following steps: Step 1: Receive the vibration of noise sound waves through an interference type optical fiber hydrophone (12) array arranged in the turbine flow channel, and convert the noise signal into an optical signal for transmission; Step 2: Convert the optical signal into an electrical signal, amplify the signal, and convert it into a digital signal; Step 3: Perform bandpass filtering and reverse filtering on the digital signal to retain the noise signal in the specified frequency band and suppress reverberation interference; Step 4: Perform FastICA blind source separation on the filtered signal to extract the characteristics of the noise signal; Step 5: Based on dynamic time warping (DTW) and support vector machine (SVM) algorithms, combined with the fault feature database, the extracted features are identified to achieve fault warning.

10. A method for monitoring flow leakage noise and identifying faults in a turbine flow passage according to claim 9, characterized in that: The method also includes preprocessing the extracted characteristic data to improve the accuracy and efficiency of subsequent fault identification.

11. A method for monitoring flow leakage noise and identifying faults in a turbine flow passage according to claim 10, characterized in that: The method also includes regularly updating the fault feature database to reflect the latest fault types and features, thereby improving the adaptability and accuracy of fault identification.

12. A method for monitoring flow leakage noise and identifying faults in a turbine flow passage according to claim 11, characterized in that: The method also includes classifying and prioritizing the identified fault types so as to take corresponding maintenance measures in time to ensure the safe and stable operation of the turbine.

Images