Water pump water turbine sound and vibration cooperative monitoring and signal fusion system and method

Abstract

This invention relates to a system and method for coordinated monitoring and signal fusion of acoustic and vibration signals in pump-turbine units, belonging to the field of condition monitoring technology for pumped storage units. It aims to solve the problems of existing technologies, such as different sources of acoustic and vibration measurement points, fixed acquisition strategies, lack of physical basis for fusion, and low accuracy of condition assessment. The system includes acoustic and vibration signal sensing and adaptive acquisition, signal preprocessing and multi-dimensional feature extraction, acoustic and vibration homogeneous coupling and feature-level fusion, a condition assessment and decision-making application layer, and a data storage unit. The method synchronously acquires acoustic and vibration signals from paired measurement points of the same source, adaptively switches acquisition strategies according to operating conditions, and constructs a multi-dimensional fusion feature system through differentiated processing and feature filtering. After dimensionality reduction, the system is input into the assessment model and outputs health status and hydraulic stability scores. This invention ensures the physical homogeneity of acoustic and vibration signals, balances data integrity and storage efficiency, achieves feature-level fusion driven by physical essence, significantly improves the accuracy of unit condition assessment, and provides support for the safe operation and condition-based maintenance of pumped storage units.

Description

Technical Field

[0001] This invention relates to the field of pumped storage unit condition monitoring technology, and in particular to a system and method for coordinated monitoring and signal fusion of sound and vibration of water pumps and turbines. Background Technology

[0002] As the core power equipment of pumped storage power stations, the pump-turbine's operating status directly affects the overall performance and safety of the station. Pump-turbines operate in both directions, frequently switching between different conditions, and undergo complex and varied operating scenarios. These scenarios encompass steady-state operation of both pump and turbine modes, as well as various transient processes such as start-up, shutdown, load shedding, and mode switching. Under different operating conditions, the hydraulic excitation characteristics within the flow channel differ significantly. Flow channel noise and structural vibration response are both excited by the same hydraulic phenomenon, exhibiting natural physical homogeneity and information complementarity. Therefore, collaboratively monitoring acoustic and vibration signals and exploring their inherent correlations becomes a crucial approach to achieving accurate assessment of the unit's hydraulic stability and precise perception of its operating status.

[0003] Currently, several technical solutions have been disclosed regarding the condition monitoring and fault identification of hydro-turbine units. For example, CN117932525A discloses a method, system, equipment, and medium for acoustic-optical fiber fault identification of hydro-generator units, which uses acoustic sensors to acquire multi-channel acoustic signature signals for fault identification; CN120160706A discloses a system and method for monitoring and identifying discharge noise in the flow channel of a hydro-turbine, which uses an interferometric hydrophone array to collect flow channel noise and provide fault early warning. These solutions achieve effective utilization of acoustic signals, but relying on only a single physical quantity makes it difficult to conduct a more complete assessment of the structural effects of hydraulic excitation. Specifically, these solutions may not fully consider the physical homogeneity of acoustic and vibration signals in the sensor deployment, resulting in insufficient in-depth analysis of the correlation between signals, which in turn affects the accuracy and comprehensiveness of fault identification.

[0004] Several technical solutions have been attempted in the fusion of multiple physical quantities. For example, CN119878421A discloses a method for evaluating the stability of mixed-flow turbines by fusing the impact characteristic values ​​of acoustic and vibration signals. This method simultaneously collects vibration and noise signals, extracts the energy entropy ratio, and performs a sum-of-squares operation to obtain the fused characteristic values. CN119989112A discloses a method for turbine anomaly detection and fault diagnosis based on multi-sensor data fusion. This method collects stress, audio, vibration, and operating condition data, and performs anomaly detection after dimensionality reduction through principal component analysis. These solutions have initially achieved the fusion of multi-source information, but there are still aspects that need optimization. 1. Sensor Deployment: While some solutions simultaneously collect acoustic and vibration signals, the measurement points do not adhere to the principle of pairing similar sources, failing to achieve synchronous acquisition of acoustic and vibration signals at the same measurement point. This makes it difficult to establish a physical correlation model between the two signals. Consequently, in subsequent data analysis, it becomes difficult to accurately uncover the intrinsic relationship between the acoustic and vibration signals, thus affecting the accuracy of fault identification.

[0005] 2. Data Acquisition Strategy: Existing solutions mostly employ fixed sampling frequencies and storage modes, failing to consider the signal differences arising from the operating characteristics of pumps and turbines. Steady-state hydraulic excitation is primarily composed of periodic components, while transient hydraulic excitation during transitions contains abundant high-frequency non-stationary components. Using fixed parameters for acquisition makes it difficult to balance data integrity and storage efficiency. For example, during transitions, high-frequency non-stationary components may be missed due to insufficient sampling rates, resulting in inaccurate capture of fault characteristics.

[0006] 3. Fusion and Representation Level: Existing solutions primarily rely on feature splicing or numerical combination for fusion, failing to achieve coupled feature mining based on physical homology. They also lack a feature system capable of representing the transmission relationship between hydraulic excitation and structural response, resulting in insufficient sensitivity and accuracy in identifying abnormal unit states and hydraulic instability. This simplistic fusion approach cannot fully uncover the deep connections between acoustic and vibration signals, thus limiting the improvement of fault identification capabilities.

[0007] In summary, existing technologies still have many shortcomings in the condition monitoring and fault diagnosis of pump-turbines, making it difficult to meet the monitoring needs under complex and ever-changing operating conditions. Specifically, existing technologies have significant limitations in reflecting the coupling relationship between hydraulic excitation and structural response, pairing acoustic and vibration measurement points from the same source, acquiring signal differences, and representing multi-source data fusion, resulting in the inability to perform comprehensive and accurate condition monitoring and fault diagnosis of pump-turbines. Therefore, this invention proposes a pump-turbine acoustic and vibration collaborative monitoring and signal fusion system and method. It aims to achieve accurate classification of the unit's health status and quantitative assessment of hydraulic stability through the layout of acoustic and vibration measurement points from the same source, an adaptive acquisition strategy for all operating conditions, and an acoustic and vibration fusion feature system based on physical homology, thereby providing strong support for the safe and stable operation of pump-turbines. This addresses the key problems in existing technologies and improves the accuracy and comprehensiveness of pump-turbine condition monitoring and fault diagnosis. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a system and method for coordinated monitoring and signal fusion of acoustic and vibration of water pumps and turbines, which overcomes the defects of existing water pump and turbine condition monitoring technologies, such as different sources of acoustic and vibration measurement points, acquisition strategies that are not adapted to operating conditions, lack of physical basis for fusion methods, and insufficient accuracy of condition assessment, so as to realize accurate perception of unit operating status, health status classification and quantitative assessment of hydraulic stability.

[0009] To achieve the above technical objectives, the present invention adopts the following technical solution: This invention provides a water pump and turbine acoustic and vibration collaborative monitoring and signal fusion system, comprising an acoustic and vibration signal sensing and adaptive acquisition layer, a signal preprocessing and multi-dimensional feature extraction layer, an acoustic and vibration homogeneous coupling and feature-level fusion layer, a state assessment and decision application layer, and a data storage and management unit that is communicatively connected to the above four layers.

[0010] 1. Acoustic and vibration signal sensing and adaptive acquisition layer In key areas such as the top cover of the pump turbine, the volute, the tailrace pipe, and the shaft bearings, spatially corresponding paired measuring points are arranged to synchronously collect acoustic signals and vibration response signals, ensuring the physical origin of the signals. By communicating with the power plant's SCADA system, steady-state and transient operating conditions are identified in real time, and differentiated sampling rates, storage modes, and triggering strategies are automatically matched to balance transient signal integrity and steady-state storage efficiency.

[0011] 2. Signal preprocessing and multidimensional feature extraction layer Differential filtering and denoising are implemented to address the differences in physical properties of acoustic and vibration signals, and initial features in multiple dimensions, including time domain, frequency domain, and time-frequency domain, are extracted. Two-level feature selection is completed through correlation analysis and variance contribution analysis to eliminate weakly correlated and redundant features, resulting in a highly recognizable and effective feature set.

[0012] 3. Acoustic-Vibration Homogeneous Coupling and Feature-Level Fusion Layer Based on the physical homology of acoustic vibration, three types of homologous fusion features are constructed: time-domain correlation, frequency-domain coupling, and nonlinear correlation. After standardization, the features are spliced ​​together, and the core principal components are extracted using a dimensionality reduction algorithm to output a low-dimensional, high-representational-capability fusion feature set, thus achieving feature-level fusion based on physical essence.

[0013] 4. Status Assessment and Decision Application Layer The fused features are input into the pre-trained evaluation model, which outputs four levels of health status: normal, attention, abnormal, and fault. A hydraulic stability quantitative scoring system is established based on acoustic-vibration correlation features, and the evaluation results are given in percentage form. When an abnormality occurs, an early warning is automatically triggered, the fault measurement point is located, and the fault type is indicated to support operation and maintenance decisions.

[0014] 5. Data Storage and Management Unit It unifies the storage of raw acoustic and vibration signals, feature data, fusion results, and evaluation reports, and supports historical data backtracking, querying, and exporting, providing data support for model iteration, fault analysis, and condition-based maintenance.

[0015] This invention also provides a method for coordinated monitoring and signal fusion of acoustic and vibration signals of a water pump and turbine, based on the aforementioned coordinated monitoring and signal fusion system for acoustic and vibration signals of a water pump and turbine, comprising the following steps: Step 1, Deployment of Same-Source Measurement Points: Complete the same-source pairing and installation of acoustic and vibration sensors in the areas of top cover, volute, tailwater pipe, and water guide bearing, and perform system debugging and synchronous calibration; Step 2, Adaptive Acquisition Control: Real-time identification of unit operating conditions, automatic loading of steady-state / transient process acquisition templates, and synchronous acquisition and digitization of acoustic and vibration signals; Step 3, Differentiated preprocessing and feature extraction: Denoise and perform time-frequency transformation on the sound and vibration signals respectively, extract multi-dimensional features and complete two-level screening to obtain an effective feature set; Step 3, Coupling of Common Origins and Feature Fusion: Construct three types of common origin association features in the time domain, frequency domain, and nonlinear domain. After standardization, splicing, and dimensionality reduction, a fused feature set is obtained. Step 4, Status Assessment and Application: Input the assessment model and output the health status and hydraulic stability score to achieve visualization, anomaly warning and decision support; Step 5: Model Iteration and Optimization: Incremental learning based on operation and maintenance data to continuously improve the accuracy of assessment.

[0016] The system and method for coordinated monitoring and signal fusion of acoustic vibration of water pumps and turbines provided by this invention have the following beneficial effects: 1. This invention adopts a pairing arrangement of acoustic and vibration measurement points with the same source, and realizes a one-to-one spatial correspondence between acoustic measurement points and vibration measurement points in core areas such as the top cover, volute, tailpipe, and shaft bearings. This ensures the strong homogeneity of acoustic and vibration signals from a physical source, significantly improves the signal correlation and analysis reliability, and solves the pain points of spatial mismatch and poor correlation of acoustic and vibration signals in the prior art.

[0017] 2. This invention designs a multi-channel synchronous acquisition system with an acquisition synchronization error of ≤10μs, providing a high-precision time reference for acoustic-vibration coupling analysis and enhancing the accuracy of signal synchronous acquisition.

[0018] 3. This invention introduces an adaptive acquisition strategy for operating conditions, which can identify the steady-state and transient operating conditions of the unit in real time and automatically match the different sampling rates, accuracy and storage modes. This not only ensures the complete capture of transient signals during the transient process, but also greatly improves the utilization rate of storage resources under steady-state operating conditions, and solves the problem that fixed sampling frequencies are difficult to adapt to different operating conditions.

[0019] 4. This invention employs differentiated preprocessing technology to address the differences between acoustic and vibration signals, effectively suppressing ambient noise and improving the signal-to-noise ratio and feature fidelity. Simultaneously, through differentiated preprocessing and a simplified feature extraction mechanism, high-fidelity features are efficiently extracted. Redundant components are eliminated through feature screening, providing a higher-quality signal foundation and high-quality basic features required for subsequent processing and fusion processing.

[0020] 5. This invention eliminates weakly correlated and redundant features through a two-level feature screening mechanism, reducing the amount of computation while improving the feature's ability to identify the unit's status, thus improving the efficiency and accuracy of feature extraction.

[0021] 6. Based on the homogeneity characteristics, this invention constructs three types of homogeneity correlation feature systems: time-domain correlation, frequency-domain coupling, and nonlinear fusion. Starting from the inherent transmission law of hydraulic excitation-structural response, it realizes the physical essence fusion of acoustic and vibration signals, which greatly improves the ability to characterize the unit's operating status and hydraulic stability.

[0022] 7. This invention significantly improves the sensitivity of identifying hydraulic instability phenomena by integrating the intrinsic transmission law of hydraulic excitation-structural response through feature depth characterization, providing more accurate fault warnings for operation and maintenance personnel.

[0023] 8. This invention employs standardization and PCA dimensionality reduction techniques, eliminating dimensional differences while retaining core information, improving model training efficiency and evaluation accuracy, and providing more reliable data support for unit health status assessment.

[0024] 9. This invention introduces a random forest model to achieve a four-level classification of health status, which can intuitively give the classification conclusions of normal, attention, abnormal and fault, providing clear status assessment results for operation and maintenance personnel.

[0025] 10. This invention establishes a quantitative scoring system for hydraulic stability, which uses a percentage system to intuitively evaluate the operational stability of the unit, making it easier for operation and maintenance personnel to quickly judge the hydraulic stability status of the unit and improving decision-making efficiency.

[0026] 11. This invention has the ability to automatically warn of anomalies and locate faults, and can quickly locate abnormal measurement points and make a preliminary judgment on the fault type and component, providing timely fault handling support for operation and maintenance personnel and reducing operation and maintenance costs.

[0027] 12. This invention supports full-process data storage and backtracking functions, providing complete data support for fault review, model iteration, and life assessment, which helps to continuously optimize the performance of the monitoring system.

[0028] 13. This invention can be seamlessly integrated with the power plant SCADA system without changing the existing operation and maintenance process. It is easy to deploy in projects and promote on-site, reducing the difficulty and cost of system integration.

[0029] 14. This invention realizes closed-loop management of the entire process from data collection, processing, fusion to evaluation and early warning, providing precise support for condition-based maintenance and safe and stable operation, and improving the overall operational safety and reliability of the unit.

[0030] 15. By optimizing sensor deployment and acquisition strategies, this invention reduces data storage costs and improves data processing efficiency, making large-scale data monitoring possible and providing strong support for the intelligent operation and maintenance of pumped storage power stations.

[0031] 16. By constructing a fusion feature system based on physical homology, this invention not only improves the accuracy and comprehensiveness of the evaluation results, but also provides new ideas and methods for the research on the failure mechanism of water pumps and turbines and the development of condition monitoring technology. Attached Figure Description

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural framework diagram of the system of the present invention; Figure 2 This is a flowchart of the adaptive acquisition strategy of the present invention; Figure 3 This is a schematic diagram of the layout of the water pump turbine acoustic and vibration signal monitoring system of the present invention; In the diagram: volute 1, tailpipe 2, top cover 3, bearing 4. Detailed Implementation

[0033] The technical solutions of the present invention will be further described below with reference to the embodiments and accompanying drawings: Example 1 This embodiment provides a water pump and turbine acoustic and vibration collaborative monitoring and signal fusion system, the overall architecture of which is as follows: Figure 1 As shown, a four-layer progressive processing architecture is adopted, which includes a layer for sensing and adaptive acquisition of acoustic and vibration signals, a layer for signal preprocessing and multi-dimensional feature extraction, a layer for acoustic and vibration co-location and feature-level fusion, and a layer for state assessment and decision application, which are connected in sequence. At the same time, a data storage and management unit is set up to establish communication connection with the above four layers for the storage, traceability and management of data throughout the process.

[0034] I. Acoustic and Vibration Signal Sensing and Adaptive Acquisition Layer This layer is the hardware core of the system, used to synchronously acquire acoustic and vibration response signals during the operation of the unit based on the same measuring point, and automatically match the acquisition strategy according to the unit's operating conditions. Specifically, it includes three parts: an acoustic and vibration sensor array and a unit for arranging paired measuring points, an adaptive trigger control unit, and a synchronous signal modulation and acquisition unit.

[0035] (I) Acoustic and vibration sensor array and its arrangement In this embodiment, the acoustic sensor array uses industrial-grade waterproof microphones (either the GRAS46BL model or the domestic Hangzhou Aiwa AWA14423 model), with a frequency response range covering 20Hz~20kHz, a dynamic range of no less than 120dB, and a protection rating of no less than IP67 (Ingress Protection 67). It is suitable for humid industrial environments with strong electromagnetic interference and can effectively collect hydraulic noise and structural acoustic signals transmitted through the flow channel. As an alternative implementation, the acoustic sensor array can be replaced with fiber optic FP (Fabry-Perot) acoustic sensors or an interferometric fiber optic hydrophone array.

[0036] The vibration sensor array uses an industrial-grade IEPE (Integrated Electronics Piezo-Electric) accelerometer (either the PCB352C33 model or the LC0159 model tested by Lancet), with a measurement range of ±10g, a frequency response range of 0.5Hz~5kHz (resolution ≤0.001g), a built-in signal modulation circuit, and an IP68 (Ingress Protection 68) protection rating, meeting the installation environment requirements of the unit under strong vibration and high humidity. As an alternative implementation, the vibration sensor array can be replaced with a piezoelectric accelerometer.

[0037] The acoustic sensor array and the ICP (Integrated Circuit Piezoelectric) accelerometer array are spatially paired and arranged at corresponding points in key areas of the water pump and turbine. The overall layout of the measuring points is as follows: Figure 3 As shown, the specific layout area and implementation details are as follows: Top Cover Area: On the upper plane of the top cover 3 of the water pump turbine, corresponding to the circumferential position of the upper crown of the runner, four sets of paired measuring points of the same source are evenly arranged circumferentially. In each set of measuring points, the acoustic sensor is attached to the flat mounting base surface of the outer wall of the top cover with a coupling agent to collect the structural acoustic signal transmitted by the hydraulic excitation of the runner area through the top cover shell; the vibration accelerometer is fixed to the same mounting base surface with a coupling agent to collect the vibration response signal of the top cover structure. The spatial straight-line distance between the two does not exceed 1cm to ensure strong homology of the signals.

[0038] Volute area: A set of paired measuring points of the same source are arranged on the outer wall of the main inlet of volute 1. The acoustic sensor is attached to the outer wall of the inlet to collect the structural acoustic signal transmitted through the shell by the water pressure pulsation inside the volute; the vibration accelerometer is fixed to the same mounting base surface by coupling agent; Tailwater pipe area: A set of paired measuring points of the same source are arranged on the outer wall of the inlet gate of the straight conical section of tailwater pipe 2. The acoustic sensor is attached to the outer wall of the inlet gate of the straight conical section to collect the structural acoustic signals transmitted through the shell by the vortex belt excitation at the runner outlet and the water flow turning and separating; the vibration accelerometer is attached to the same mounting base surface of the corresponding inlet gate.

[0039] Shaft bearing area: On the rigid mounting plane outside the bearing housing of water-guided bearing 4, two sets of paired measuring points are arranged at 90° intervals around the circumference to monitor the acoustic and vibration signals in the horizontal radial (X-direction) and vertical radial (Y-direction) directions, respectively. In each set of measuring points, the acoustic sensor is tightly attached to the outer wall of the bearing housing through a coupling agent to collect the acoustic signals of bearing operation, oil film eddy, and structural transmission; the vibration accelerometer is rigidly fixed to the same mounting base to collect the radial vibration response signal of the bearing housing.

[0040] Before installation, all the above measuring points have their base surfaces ground, rust removed, and leveled. The sensors are rigidly fixed with high-strength epoxy structural adhesive or bolts and are waterproof and corrosion-resistant encapsulated to ensure long-term operational reliability and guarantee signal transmission efficiency.

[0041] (ii) Adaptive signal acquisition and control unit In this embodiment, the adaptive trigger control unit communicates bidirectionally with the power plant computer monitoring system (SCADA, Supervisory Control and Data Acquisition) to acquire operating parameters such as unit speed, guide vane opening, active power, and operating condition commands in real time. It accurately identifies steady-state and transient operating conditions and automatically loads the corresponding acquisition parameter templates. It also has an emergency trigger acquisition function. The adaptive acquisition strategy process is as follows: Figure 2 As shown.

[0042] This unit includes a parameter configuration module, an adaptive execution module, and an emergency triggering module. The specific implementation methods of each module are as follows: Acquisition parameter configuration module: It has preset acquisition parameter templates that correspond one-to-one with the working condition type. The template content includes sampling rate, recording duration, triggering method, storage interval, and filtering parameters.

[0043] The steady-state data acquisition template is set as follows: the sampling rate of the sensor channel is no less than 25.6kHz, and the sampling rate of the vibration sensor channel is no less than 12.8kHz. A cyclic buffer timed storage mode is adopted, storing data every 15 minutes, with each storage lasting 20 seconds. The stored data includes 10 seconds before and after the trigger moment.

[0044] The transient process condition acquisition template is set as follows: the sampling rate of the acoustic sensor channel is no less than 51.2kHz, and the sampling rate of the vibration sensor channel is no less than 25.6kHz. A continuous acquisition and storage mode is adopted, recording continuously from the trigger moment until 60 seconds after the condition stabilizes.

[0045] The adaptive execution module receives operating condition identification results in real time, automatically loads the acquisition parameter template matching the current operating condition, and sends acquisition commands to the synchronization signal modulation and acquisition unit to achieve fully automatic switching of acquisition strategies. It also adds an operating condition transition prediction mechanism, loading the transition process acquisition template based on the SCADA system's scheduling commands to ensure complete signal acquisition during operating condition switching. For abnormal fluctuations under steady-state operating conditions, a short-term high-speed acquisition trigger mechanism is added. When the signal amplitude exceeds a threshold under steady-state operating conditions, a 10-second high-speed acquisition is automatically triggered to capture transient abnormal characteristics, balancing data integrity and storage efficiency.

[0046] The preset threshold of the emergency trigger module follows the statistical 3σ (3-Sigma, 3 times the standard deviation) criterion, the requirements of the industry standards GB / T6075.5-2022 and DL / T556-2021, and is also compatible with the common practical experience of pumped storage power station projects at home and abroad. The calibration method is as follows: collect 20 sets of continuous normal acoustic and vibration signals when the unit is running stably under rated operating conditions, calculate the arithmetic mean μ and standard deviation σ of the amplitude of each set of signals, and use μ+3σ as the trigger threshold of the corresponding channel. The emergency threshold is set to twice the trigger threshold.

[0047] (III) Synchronization Signal Modulation and Acquisition Unit The synchronous signal modulation and acquisition unit is built with a multi-channel synchronous data acquisition chassis based on the PXIe (PCI eXtensions for Instrumentation Express) architecture, and has a built-in synchronous clock module, multi-channel signal modulation module and synchronous AD (Analog-to-Digital) acquisition module.

[0048] The synchronization clock module provides a highly stable synchronization clock reference, provides synchronization trigger signals for all acoustic and vibration signal acquisition channels, and controls the multi-channel synchronous acquisition error to ≤10μs, meeting the requirements of source correlation analysis. The signal modulation module provides corresponding excitation, amplification, and filtering for acoustic and vibration sensors, with an analog-to-digital conversion accuracy of no less than 16 bits. The synchronous AD acquisition module completes analog-to-digital conversion according to the issued acquisition parameters, and transmits the digitized acoustic and vibration synchronization data to the signal preprocessing and multi-dimensional feature extraction layer via Ethernet.

[0049] II. Signal Preprocessing and Multidimensional Feature Extraction Layer This layer connects to the sensing and acquisition layer and is used to perform differentiated preprocessing on the digital acoustic and vibration signals. It extracts multi-dimensional initial features of the acoustic and vibration signals respectively, and filters the initial features to obtain effective feature sets of acoustic and vibration signals. Specifically, it includes a filtering and denoising module, a time-frequency transformation module, and a feature extraction and filtering module.

[0050] The filtering and denoising module employs differentiated filtering and denoising strategies to address the differences in the physical properties of acoustic and vibration signals. It also incorporates an adaptive noise matching function, automatically adjusting the denoising algorithm parameters based on the unit's current operating conditions and environment to ensure a high signal-to-noise ratio signal is obtained under various conditions. The specific implementation method is as follows: For acoustic signals: a 50Hz power frequency notch filter is used to eliminate power frequency electromagnetic interference at the power station site; then wavelet threshold denoising or adaptive filtering methods are used to retain signal components in the effective frequency band of 20Hz~20kHz, which is suitable for the wide frequency and low amplitude physical characteristics of acoustic signals. For vibration signals: an 8th-order Butterworth bandpass filter is used to remove low-frequency drift and high-frequency noise, retaining 0.5Hz~5kHz; if necessary, empirical mode decomposition (EMD) is used to remove trend terms and eliminate signal baseline drift caused by changes in unit operating conditions.

[0051] Time-frequency transformation module: For stationary signals under steady-state conditions, Fast Fourier Transform (FFT) is used to obtain the frequency domain amplitude and energy distribution of the signal, with a spectral resolution of 1Hz; for non-stationary signals during transient processes, Short-Time Fourier Transform (STFT) is used for time-frequency analysis to provide basic time-frequency domain data for subsequent multi-dimensional feature extraction.

[0052] The feature extraction and filtering module includes an acoustic signal feature extraction unit, a vibration signal feature extraction unit, and a feature filtering unit. It also adds an adaptive feature weight allocation function, which prioritizes increasing the weight of steady-state frequency domain features for steady-state conditions and prioritizes increasing the weight of transient time-frequency domain features for transient processes, further enhancing the feature representation capability of the unit's operating status. The specific implementation method is as follows: Three core features—time domain, frequency domain, and time-frequency domain—are extracted from the preprocessed single-channel acoustic signal.

[0053] The acoustic signal feature extraction unit, wherein; Temporal characteristics include kurtosis, root mean square (RMS), and peak value. Frequency domain characteristics: including centroid frequency, spectral entropy, and energy proportion in the 125Hz~500Hz frequency band.

[0054] Vibration signal feature extraction unit, wherein: Temporal characteristics include kurtosis, root mean square (RMS), and peak value. Frequency domain characteristics: including frequency conversion amplitude, second harmonic amplitude, and the proportion of energy in the high-frequency band above 1kHz; Feature filtering unit: Employs a two-level filtering mechanism to eliminate weakly correlated and redundant features, reducing subsequent computational load while improving the feature's ability to identify unit status. Specific rules are as follows: The first stage uses Pearson correlation analysis to eliminate weak correlation characteristics with operating parameters such as active power and guide vane opening, where the absolute value of the correlation coefficient is <0.2. The second stage processes the remaining features and performs variance contribution analysis to remove redundant features with a variance contribution of less than 1%. Finally, effective feature sets of acoustic signals and vibration signals that are sensitive to the unit's operating status are obtained, reducing the amount of computation.

[0055] III. Acoustic-Vibration Coupling and Feature-Level Fusion Layer This layer connects to the feature extraction layer and is used to standardize the effective acoustic and vibration feature sets extracted by the preceding modules. Based on the physical homology of acoustic and vibration signals, it constructs an acoustic-vibration correlation feature system, completing the deep fusion of multi-source acoustic and vibration features and outputting a low-dimensional, highly recognizable fused feature set. The specific implementation consists of three core steps: Construction of a correlation feature system based on the physical origin of acoustic and vibration signals. Acoustic sensors and accelerometers at corresponding spatial locations constitute the effective feature sets of acoustic and vibration signals for the same paired measurement point. Three types of correlation and fusion features with clear physical meaning are constructed to achieve a quantitative characterization of the intrinsic correlation between hydraulic excitation and structural response, specifically including: Temporal correlation fusion features: Calculate the maximum cross-correlation coefficient and feature time delay difference of the acoustic and vibration signals from the same paired measurement point. The calculation formula is as follows: (1); In the formula, For delay down signal With vibration signal The cross-correlation coefficient is dimensionless; It is a time-domain sequence of the acoustic signal; Vibration signal time-domain sequence; The mean of the acoustic signal samples; The mean of the vibration signal samples; The length of the signal sample, in points; Time delay, unit: seconds; For time series indexing.

[0056] The maximum cross-correlation coefficient is taken as the final temporal correlation feature, and the corresponding time delay is the feature time delay difference, which quantifies the temporal characteristics of hydraulic excitation transmitted from the flow channel to the structure.

[0057] Frequency domain coupling and fusion characteristics: Calculate the transfer function and coherence coefficient of acoustic signal energy and vibration signal energy within the same frequency band. The formula for calculating the coherence coefficient is as follows: (2); In the formula, For frequency The coherence coefficients of the acoustic and vibration signals range from 0 to 1. The cross power spectral density of acoustic and vibration signals; The power spectral density of the acoustic signal; The power spectral density of the vibration signal; Signal frequency, unit: Hertz.

[0058] The closer the coherence coefficient is to 1, the higher the degree of coupling between the two at that frequency; this feature is used to quantify the coupling strength between hydraulic excitation and structural response at different frequencies.

[0059] Nonlinear fusion characteristics: The mutual information entropy (MI) between the acoustic and vibration signals at the same paired measurement point is calculated using the built-in mutualinfo function in Matlab, which is used to characterize the degree of nonlinear correlation between the acoustic and vibration signals.

[0060] Meanwhile, a global acoustic-vibration correlation feature matrix is ​​constructed for multiple paired measurement points across the entire machine, enabling a comprehensive characterization of the hydraulic excitation distribution and structural response characteristics of the entire machine.

[0061] Feature standardization processing applies Z-score (standard score) standardization to all features in the effective feature sets of acoustic signals, vibration signals, and the acoustic-vibration co-origin feature system. This eliminates the influence of differences in the dimensions and orders of magnitude of different features on the fusion results. The calculation formula is as follows: (3); In the formula, These are the standardized eigenvalues; These are the original eigenvalues; This is the sample mean of this feature; This is the sample standard deviation of this feature.

[0062] The effective features of the standardized acoustic signal, effective features of the vibration signal, and features related to the same source as the acoustic signal are concatenated to form a high-dimensional feature set. Principal Component Analysis (PCA) is then used to perform linear dimensionality reduction on the high-dimensional feature set, extracting principal component features with a cumulative variance contribution ≥90%, resulting in a low-dimensional, highly discriminative fused feature set. As an alternative implementation, the dimensionality reduction algorithm can be replaced with Independent Component Analysis (ICA) or Kernel Principal Component Analysis (KPCA).

[0063] Through the above steps, information complementarity of the original acoustic and vibration features is achieved, and the inherent physical coupling relationship between the two is explored through the source correlation features. Redundant information is eliminated by dimensionality reduction, thereby improving the ability of the monitored signal features to represent the unit's operating status.

[0064] IV. Status Assessment and Decision Application Layer This layer is connected to the feature-level deep fusion unit, which is used to input the fused feature set into the pre-trained acoustic and vibration feature fusion evaluation model, output the health status classification results and hydraulic stability evaluation results under the real-time operating conditions of the unit, and realize early warning and decision support through a visual interface.

[0065] This layer receives the low-dimensional fusion feature set output by the fusion layer. The core information is retained through dimensionality reduction. The fusion features are then input into a pre-trained Random Forest (RF) classification model. This model has been trained using normal operation data, typical fault simulation data, and historical abnormal event data within the unit. The core hyperparameters of the model are: 100 decision trees, Gini coefficient as the splitting criterion, maximum depth of 15, minimum number of split samples of 5, and minimum number of leaf node samples of 2.

[0066] The model outputs the unit's health status, divided into four levels: normal, attention, abnormal, and fault. Normal: This indicates that the unit is operating smoothly, all characteristics are within the normal threshold range, and there are no abnormal signs. Note: If the feature deviates from the normal range but does not exceed the preset threshold, it indicates a slight abnormality and requires closer monitoring. Abnormal: The feature exceeds the preset threshold and there are obvious fault characteristics. It is recommended to arrange a shutdown for inspection. Fault: The characteristics significantly exceed the safety threshold, indicating a serious risk that requires immediate attention, facilitating a rapid response from operations and maintenance personnel.

[0067] In terms of hydraulic stability assessment, this layer constructs a quantitative scoring system for hydraulic stability based on the acoustic-vibration co-origin correlation feature in the fusion characteristics. This score comprehensively reflects the hydraulic excitation intensity, structural response transmission efficiency, and tailrace pressure pulsation level, and is output as a percentage with a maximum score of 100. A higher score indicates better hydraulic stability.

[0068] The formula for calculating the hydraulic stability score is: (4); In the formula, Hydraulic stability score, unit: points; , , The weighting coefficients for hydraulic excitation intensity, structural response transmission efficiency, and tailrace vortex instability are 0.4, 0.35, and 0.25, respectively. The normalized value of hydraulic excitation intensity is calculated based on the root mean square and peak values ​​of the acoustic signal. The normalized value of the structural response transfer efficiency is calculated based on the acoustic-vibration frequency domain coherence coefficient and transfer function. The value is the normalized value of the vortex instability in the tailrace pipe, calculated based on the nonlinear correlation characteristics of acoustic vibration at the measuring points in the tailrace pipe.

[0069] The scoring results are divided into four levels: Excellent (90-100 points), Good (80-89 points), Average (60-79 points), and Poor (below 60 points). When the score is below 60 points, the system automatically matches the corresponding hydraulic instability type based on the feature contribution, providing data for operators to adjust the load and optimize the operating range.

[0070] When the health status assessment indicates an abnormality or malfunction, the system automatically triggers an audible and visual warning. Based on feature contribution analysis, it reverse-locates the main source measurement points of the abnormal features and, combined with acoustic-vibration coupling characteristics, preliminarily determines the possible faulty components and types, such as turbine runner cavitation, guide vane wear, and bearing rubbing. The warning information and diagnostic results are simultaneously pushed to the power plant control room, maintenance duty room, and mobile terminals.

[0071] The visual interface displays real-time unit operating conditions, acoustic and vibration waveforms and spectra at key measuring points, health status levels, hydraulic stability scores, and historical trend curves. When an alarm is triggered, the interface automatically pops up an alarm window displaying fault prompts and handling suggestions. The system has a built-in historical data backtracking function, allowing users to query status assessment results and characteristic change trends for any time period, and automatically generate health status analysis reports, providing data support for unit condition-based maintenance. This layer achieves a closed loop from data acquisition and feature fusion to status diagnosis and decision support, with a simple and clear overall process that can seamlessly integrate with the power plant's existing monitoring system.

[0072] V. Data Storage and Management Unit This unit establishes communication connections with the signal preprocessing and multidimensional feature extraction layer, the acoustic-vibration homogeneous coupling and feature-level fusion layer, and the state assessment and decision application layer. It is used to store the acquired acoustic-vibration raw signals, extracted feature data, fusion processing results, and state assessment reports. It also supports historical data retrieval, backtracking, and export, providing data support for subsequent fault analysis, model optimization, and life assessment.

[0073] Example 2 In another preferred embodiment, based on Embodiment 1, this embodiment provides a method for coordinated monitoring and signal fusion of acoustic and vibration characteristics of a water pump and turbine, implemented based on the coordinated monitoring and signal fusion system for acoustic and vibration characteristics of a water pump and turbine described in Embodiment 1. The specific steps and workflow are as follows: Step 1: Following the principle of locating points with the same source, arrange and install sensors at the pump turbine top cover 3, the inlet of the volute 1, the inlet of the tailrace pipe 2, and the water guide bearing 4 to complete the pairing of acoustic and vibration measurement points. Complete the system hardware connection, multi-channel time delay calibration, and functional debugging. The measurement point arrangement is as follows: Figure 3 As shown; Step 2: The adaptive trigger control unit acquires unit operating parameters in real time, identifies the current operating condition of the unit, and automatically loads the matching acquisition parameter template. The adaptive acquisition strategy process is as follows: Figure 2 As shown, a data acquisition command is sent to the synchronization signal modulation and acquisition unit; Step 3: The synchronous signal modulation and acquisition unit drives the acoustic sensor array and accelerometer array to complete the synchronous acquisition, modulation and analog-to-digital conversion of acoustic and vibration signals, and outputs time-synchronized digital acoustic and vibration data. Step 4: Signal preprocessing and multidimensional feature extraction layer: The digital acoustic and vibration signals are subjected to targeted denoising and time-frequency transformation. Multidimensional initial features of the acoustic and vibration signals are extracted respectively, and two-level screening is completed to obtain the effective feature set of the acoustic signal and the effective feature set of the vibration signal. Step 5: Acoustic-Vibration Homogeneous Coupling and Feature-Level Fusion Layer. Based on the physical homogeneity of acoustic and vibration, an acoustic-vibration correlation feature system is constructed, and feature standardization, splicing and dimensionality reduction are completed to output a low-dimensional, highly recognizable fusion feature set. Step 6: The state assessment and decision-making application layer will integrate the feature set input into the trained acoustic and vibration feature fusion assessment model, outputting the unit health status and hydraulic stability assessment results, completing visualization and anomaly early warning. The overall system architecture is as follows: Figure 1 As shown; Step 7: Based on the actual operation and maintenance results on site, update the model parameters through incremental learning to continuously improve the accuracy of condition assessment.

[0074] In the preferred embodiment, the acoustic and vibration signal sensing and adaptive acquisition layer includes an acoustic and vibration sensor array and a paired measurement point arrangement unit, an adaptive trigger control unit, and a synchronous signal modulation and acquisition unit. Within the paired measurement points, acoustic sensors and vibration sensors are spatially arranged in a one-to-one correspondence, with a distance of no more than 1 cm between them. The adaptive trigger control unit communicates with the power plant's SCADA system, identifies steady-state and transient operating conditions, and loads corresponding acquisition parameter templates. The synchronous signal modulation and acquisition unit adopts a PXIe architecture, with a multi-channel synchronous acquisition error ≤10μs. These settings ensure the spatiotemporal synchronization and high precision of the acoustic and vibration signal acquisition, effectively avoiding analytical deviations caused by signal asynchrony or acquisition errors. This provides reliable data support for subsequent accurate analysis of the relationship between hydraulic excitation and structural response, and enhances the overall monitoring system's sensitivity to the unit's operating status.

[0075] In the preferred embodiment, the adaptive trigger control unit presets two sets of acquisition templates: steady-state and transient processes. For steady-state operation, the acoustic channel sampling rate is no less than 25.6kHz, and the vibration channel sampling rate is no less than 12.8kHz, with periodic cyclic buffering. For transient operation, the acoustic channel sampling rate is no less than 51.2kHz, and the vibration channel sampling rate is no less than 25.6kHz, with continuous acquisition and storage. These settings match differentiated acquisition strategies to the characteristics of different operating conditions. Storage resources are rationally allocated under steady-state conditions, while transient signals are fully captured under transient conditions. This comprehensively preserves key information during unit operation, facilitating in-depth analysis of the unit's operating characteristics under different conditions and providing abundant data for accurately assessing the unit's health status.

[0076] In the preferred embodiment, the signal preprocessing and multidimensional feature extraction layer uses power frequency notch filtering and wavelet thresholding to denoise the acoustic signal, performs Butterworth bandpass filtering and empirical mode decomposition on the vibration signal, and completes two-level feature screening through Pearson correlation analysis and variance contribution analysis. The above settings effectively remove noise interference from the signal, extract feature information that better reflects the essence of unit operation, improve the quality and effectiveness of features, and enable subsequent analysis to be based on purer and more representative features, thereby more accurately revealing the relationship between unit operating status and features.

[0077] In the preferred embodiment, the acoustic-vibration co-source coupling and feature-level fusion layer constructs three types of fusion features: time-domain correlation, frequency-domain coupling, and nonlinear correlation. After Z-score standardization, feature splicing and PCA dimensionality reduction are performed to obtain a fusion feature set. The above settings deeply explore the intrinsic relationship between acoustic and vibration signals from multiple dimensions, eliminate the influence of dimensions through standardization, and then obtain more representative and distinguishable fusion features through splicing and dimensionality reduction. This can more comprehensively and accurately reflect the unit's operating status and provide stronger feature support for subsequent status assessment.

[0078] In the preferred embodiment, the state assessment and decision-making application layer uses a pre-trained random forest model to output four levels of health status: normal, attention, abnormal, and fault. Based on the acoustic-vibration homology feature, it outputs a hydraulic stability quantitative score of 0-100 points. In abnormal states, it automatically triggers early warnings and locates the source of the fault. The above settings utilize the powerful classification capabilities of the random forest model to achieve accurate hierarchical assessment of the unit's health status. The quantitative score intuitively presents the hydraulic stability status. The automatic early warning and fault location functions can promptly detect unit anomalies and determine the fault location, providing maintenance personnel with timely and effective decision-making basis and ensuring the safe and stable operation of the unit.

[0079] In the preferred embodiment, the effective feature screening in step 3 includes: removing weakly correlated features with an absolute value of correlation coefficient < 0.2 with respect to unit operating parameters, and then removing redundant features with a variance contribution of < 1%, to obtain an effective feature set of acoustic and vibration signals. The above settings remove features that have little impact on the analysis of unit operating status, reduce data dimensions and computational load, and improve analysis efficiency, while retaining key features closely related to unit operation, ensuring that subsequent analysis can focus on feature information that is of great significance to the assessment of unit status.

[0080] In the preferred embodiment, the acoustic-vibration co-source correlation feature system described in step 4 includes: the maximum cross-correlation coefficient and time delay difference characteristics of acoustic-vibration signals at the same measuring point, the frequency domain coherence coefficient and transfer function characteristics, and the nonlinear correlation characteristics of mutual information entropy. The above settings characterize the correlation between acoustic-vibration signals from multiple perspectives such as time domain, frequency domain, and nonlinearity, comprehensively reflecting the complex transmission process between hydraulic excitation and structural response. This provides rich feature information for a deeper understanding of the unit's operating mechanism and helps to more accurately assess the unit's health status and hydraulic stability.

[0081] In the preferred embodiment, the hydraulic stability quantification score in step 5 is obtained by weighted calculation of hydraulic excitation intensity, structural response transmission efficiency, and tailrace vortex instability index, and the stability level is divided according to the score. The above settings comprehensively consider multiple key factors affecting hydraulic stability. The quantification score obtained through weighted calculation can objectively and comprehensively reflect the hydraulic stability status. The division of stability levels provides operation and maintenance personnel with an intuitive reference standard, which facilitates timely implementation of corresponding measures to ensure the stable operation of the unit.

[0082] In the preferred embodiment, when the health status assessment indicates an abnormality or malfunction, the pump-turbine acoustic-vibration collaborative monitoring and signal fusion system automatically triggers an audible and visual early warning, locates the abnormal measuring point, determines the fault type, and generates a health status analysis report. These features enable rapid response to abnormal unit conditions. The audible and visual early warning promptly alerts maintenance personnel, the location of abnormal measuring points and the determination of fault types facilitate rapid maintenance, and the health status analysis report provides detailed data support for subsequent fault handling and unit optimization, thereby improving the efficiency and accuracy of unit operation and maintenance.

[0083] In summary, the pump-turbine acoustic-vibration coordinated monitoring and signal fusion system and method proposed in this invention provides a comprehensive and effective solution to the technical challenges in monitoring the operating status of pump-turbines in pumped storage power stations. In existing technologies, the operating conditions of pump-turbines are complex and variable, with significant differences in hydraulic excitation characteristics within the flow channel. Traditional monitoring methods often rely on a single physical quantity (such as vibration or acoustic signals), making it difficult to comprehensively assess the impact of hydraulic excitation on the unit structure, resulting in inaccurate and incomplete assessments of the unit's health status and hydraulic stability.

[0084] This invention employs a paired measurement point layout concept based on the homogeneity of acoustic and vibration signals. In key areas such as the top cover 3, volute 1, tailrace pipe 2, and shaft bearing 4, acoustic and vibration measurement points are spatially paired and arranged in a one-to-one correspondence. This design ensures strong homogeneity of acoustic and vibration signals, effectively solving the problems of spatial mismatch and poor correlation between these signals, laying a solid foundation for subsequent signal fusion and analysis. Simultaneously, a full-condition adaptive acquisition strategy is designed, automatically matching differentiated sampling rates, accuracy, and storage modes according to the real-time operating conditions of the pump and turbine. This strategy not only ensures complete capture of transient signals during transition processes but also significantly improves the utilization rate of storage resources under steady-state conditions, achieving dual optimization of acquisition efficiency and signal quality.

[0085] Furthermore, this invention constructs three types of fusion feature systems based on physical homology: time-domain correlation, frequency-domain coupling, and nonlinear correlation. Starting from the inherent transmission law of hydraulic excitation-structural response, it achieves deep fusion of acoustic and vibration signals, breaking through the simple fusion methods of traditional feature splicing or numerical combination. Through the physical homology guarantee mechanism of acoustic and vibration signals, and utilizing the deployment of paired measurement points and synchronous acquisition technology, synchronous monitoring of acoustic and vibration signals excited by the same hydraulic phenomenon is achieved, providing a reliable data foundation for exploring the coupling relationship between hydraulic excitation and structural response. Simultaneously, the proposed adaptive acquisition and control method based on operating condition identification automatically matches differentiated acquisition strategies for steady-state and transient processes by real-time monitoring of parameters such as unit speed and guide vane opening, resolving the technical contradiction between ensuring signal integrity and improving storage efficiency under complex operating conditions.

[0086] Finally, the acoustic-vibration fusion feature construction method proposed in this invention, based on physical homology, achieves a quantitative description of the hydraulic excitation transmission process through the synergistic representation of time-domain correlation features, frequency-domain coupling features, and nonlinear correlation features. This method significantly improves the accuracy of abnormal state identification and hydraulic stability assessment of the unit, providing a strong guarantee for the safe and stable operation of the pump turbine.

Claims

1. A system for monitoring and signal fusion of acoustic and vibration of a pump-turbine, characterized in that: The system includes a sequentially connected acoustic and vibration signal sensing and adaptive acquisition layer, a signal preprocessing and multidimensional feature extraction layer, an acoustic and vibration homogeneous coupling and feature-level fusion layer, a state assessment and decision application layer, and a data storage and management unit that is communicatively connected to the above four layers. The acoustic and vibration signal sensing and adaptive acquisition layer synchronously acquires acoustic signals and vibration response signals at the same source paired measurement points in the key areas of the pump turbine's top cover (3), volute (1), tailrace pipe (2), and shaft bearing (4), and adaptively matches the acquisition strategy according to the unit's operating conditions. The acquired acoustic and vibration signals are sent to the signal preprocessing and multidimensional feature extraction layer for differentiated preprocessing, extraction, and screening of effective features. The screened effective features are input into the acoustic and vibration homogeneous coupling and feature-level fusion layer to construct an acoustic and vibration homogeneous associated feature system, and complete feature fusion and dimensionality reduction. Finally, the fused and dimensionality-reduced features are sent to the state assessment and decision application layer to output the health status classification and hydraulic stability quantitative scoring results.

2. The system according to claim 1, characterized in that: The acoustic and vibration signal sensing and adaptive acquisition layer includes an acoustic and vibration sensor array and a homogeneous paired measurement point arrangement unit, an adaptive trigger control unit, and a synchronous signal modulation and acquisition unit; the acoustic sensors and vibration sensors are spatially arranged in a one-to-one correspondence within the homogeneous paired measurement points, with a distance between them not exceeding 1cm. The adaptive trigger control unit communicates with the power plant's SCADA system to identify steady-state and transient operating conditions and load the corresponding acquisition parameter templates; the synchronous signal modulation and acquisition unit adopts the PXIe architecture, and the multi-channel synchronous acquisition error is ≤10μs.

3. The system according to claim 2, characterized in that: The adaptive trigger control unit presets two sets of acquisition templates: steady state and transient process. Steady state condition: the sound channel sampling rate is not less than 25.6kHz, the vibration channel is not less than 12.8kHz, and the data is stored in a cyclic buffer at regular intervals. Transient process condition: the sound channel sampling rate is not less than 51.2kHz, the vibration channel is not less than 25.6kHz, and the data is stored continuously.

4. The system according to claim 1, characterized in that; The signal preprocessing and multidimensional feature extraction layer uses power frequency notch filtering and wavelet thresholding to denoise the acoustic signal, performs Butterworth bandpass filtering and empirical mode decomposition on the vibration signal, and completes two-level feature screening through Pearson correlation analysis and variance contribution analysis.

5. The system according to claim 1, characterized in that: The acoustic-vibration homogeneous coupling and feature-level fusion layer constructs three types of fusion features: time-domain correlation, frequency-domain coupling, and nonlinear correlation. After Z-score standardization, feature splicing and PCA dimensionality reduction are performed to obtain the fusion feature set.

6. The system of claim 1, wherein: The state assessment and decision application layer uses a pre-trained random forest model to output four levels of health status: normal, attention, abnormal, and fault. Based on the acoustic-vibration homology correlation features, it outputs a hydraulic stability quantitative score of 0-100 points. In abnormal states, it automatically triggers early warnings and locates the source of the fault.

7. The method of the water pump turbine acoustic vibration cooperative monitoring and signal fusion system based on any one of claims 1-6, wherein, Includes the following steps: Step 1: Complete the arrangement of sound and vibration matching measurement points and sensor installation on the top cover (3), volute (1), tailrace pipe (2), and shaft bearing (4) of the water pump turbine, and perform system debugging; Step 2: The acoustic and vibration signal sensing and adaptive acquisition layer identifies the unit's operating conditions, adaptively matches the acquisition strategy, and synchronously acquires and digitizes acoustic and vibration signals; Step 3: Send the digitized acoustic and vibration signals into the signal preprocessing and multidimensional feature extraction layer to complete differential preprocessing, multidimensional feature extraction, and effective feature selection; Step 4: Input the selected effective features into the acoustic-vibration homogeneous coupling and feature-level fusion layer to construct the acoustic-vibration homogeneous correlation feature system, and perform feature standardization, splicing and dimensionality reduction processing; Step 5: Input the fused and dimensionality-reduced features into the state assessment and decision application layer, and output the unit health status classification results and hydraulic stability quantitative score results.

8. The method of claim 7, wherein, The effective feature screening in step 3 includes: removing weakly correlated features with an absolute value of correlation coefficient < 0.2 with the unit operating parameters, and then removing redundant features with a variance contribution of < 1%, to obtain an effective feature set of acoustic and vibration signals.

9. The method of claim 7, wherein the method further comprises: The acoustic-vibration co-source correlation feature system described in step 4 includes: the maximum cross-correlation coefficient and time delay difference characteristics of acoustic-vibration signals at the same measurement point, the frequency domain coherence coefficient and transfer function characteristics, and the nonlinear correlation characteristics of mutual information entropy.

10. The method of claim 7, wherein the method further comprises: The hydraulic stability quantification score mentioned in step 5 is calculated by weighting the hydraulic excitation intensity, structural response transmission efficiency, and tailrace vortex instability index, and the stability level is divided according to the score.

Images