Hydraulic hoist pump station fault diagnosis method based on sound and vibration signal fusion

Abstract

A hydraulic hoist pump station fault diagnosis method based on sound and vibration signal fusion comprises the following steps that 1, a vibration sensor and a sound sensor are arranged on a preset point position of a hydraulic hoist pump station, and vibration signals and sound signals in the operation process of the pump station are synchronously collected; step 2, carrying out wavelet denoising processing of adaptive threshold optimization on the sound signal, then carrying out variational mode decomposition, and extracting a spectrum entropy feature of a mode component after decomposition; 3, performing adaptive empirical mode decomposition on the vibration signal, and extracting a composite feature formed by a local mean decomposition energy operator and a morphological gradient of an IMF component after decomposition; the method is used for solving the problems that in existing hydraulic hoist pump station fault diagnosis, signal diagnosis information of a single sensor is limited, the capacity of a traditional method for processing non-stable sound vibration signals is insufficient, noise separation is difficult, feature fusion is complex, diagnosis precision is low, and the requirement for on-site real-time monitoring cannot be met.

Description

Technical Field

[0001] This invention belongs to the field of hydraulic equipment fault diagnosis technology, specifically relating to a fault diagnosis method for hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion, which is applied to the operation status monitoring and fault diagnosis of hydraulic gate hoisting pump stations in water conservancy and hydropower engineering and other fields. Background Technology

[0002] Hydraulic gate hoist pump stations are key equipment in water conservancy and hydropower projects, and their operational reliability directly affects the safety and efficiency of the project. Currently, existing diagnostic methods have significant shortcomings: methods based on single sensor signals, such as using only vibration or sound sensors, fail to comprehensively reflect the complex operating conditions of the pump station due to limited information acquisition, resulting in low reliability of diagnostic results; while diagnostic methods based on traditional signal processing face challenges in separating noise from effective signals, high feature fusion dimensionality, and computational complexity when dealing with non-stationary acoustic and vibration signals under complex pump station operating conditions, failing to meet the accuracy and efficiency requirements of real-time on-site monitoring. Therefore, there is an urgent need for a diagnostic method that can fuse acoustic and vibration signals and optimize signal processing and feature fusion algorithms to improve the accuracy and reliability of fault diagnosis for hydraulic gate hoist pump stations. Summary of the Invention

[0003] The purpose of this invention is to provide a fault diagnosis method for hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion. This method addresses the limitations of existing fault diagnosis methods for hydraulic gate hoisting pump stations, which suffer from limited diagnostic information from single sensor signals, insufficient ability to process non-stationary acoustic and vibration signals, difficulty in noise separation, complex feature fusion, and low diagnostic accuracy. These issues fail to meet the requirements for real-time on-site monitoring.

[0004] To solve the above problems, the technical solution of the present invention is as follows:

[0005] A fault diagnosis method for hydraulic gate hoist pump stations based on acoustic and vibration signal fusion includes the following steps:

[0006] Step 1: Install vibration sensors and sound sensors at preset locations in the hydraulic gate hoist pump station to synchronously collect vibration and sound signals during the operation of the pump station.

[0007] Step 2: Perform adaptive threshold optimization wavelet denoising on the audio signal, then perform variational mode decomposition to extract the spectral entropy features of the decomposed modal components;

[0008] Step 3: Perform adaptive empirical mode decomposition on the vibration signal and extract the composite features composed of the local mean decomposition energy operator and morphological gradient of the decomposed IMF components.

[0009] Step 4: Employ an attention-based feature fusion algorithm to weightedly fuse the spectral entropy features of the sound signal and the composite features of the vibration signal to generate sound-vibration fusion features;

[0010] Step 5: Use an extreme gradient boosting tree classifier to learn and train the acoustic-vibration fusion features to obtain fault diagnosis results;

[0011] Step 6: Evaluate and provide feedback on the operating status of the hydraulic gate hoist pump station based on the fault diagnosis results.

[0012] Furthermore, in step 2, the adaptive threshold-optimized wavelet denoising process includes introducing a smoothing factor to optimize the soft threshold function and dynamically adjusting the denoising threshold based on the spectral characteristics of the sound signal.

[0013] Furthermore, in step 2, a dynamic penalty factor is introduced during variational mode decomposition, and the number of decomposed modes is adaptively determined based on the local features of the sound signal.

[0014] Furthermore, in step 3, the adaptive empirical mode decomposition dynamically adjusts the decomposition process using an adaptive stopping criterion designed based on the standard deviation of the vibration signal.

[0015] Furthermore, in step 4, an attention weight calculation module is constructed based on the feature fusion algorithm of the attention mechanism, and the weights are dynamically allocated according to the contribution of sound and vibration features to different fault types.

[0016] Furthermore, in step 5, the optimized extreme gradient boosting tree classifier introduces a regularization parameter adaptive optimization strategy and enhances its learning ability for fused features with the help of a Bayesian optimization algorithm.

[0017] Furthermore, in step 1, the vibration sensor is an accelerometer, and the sound sensor is an anti-interference microphone.

[0018] Furthermore, in step 3, the local mean decomposition energy operator is used to highlight the energy changes of the vibration signal, and the morphological gradient is used to characterize the edge features of the vibration signal.

[0019] Furthermore, in step 5, the fault diagnosis results include normal operation, oil pump failure, valve failure, pipeline leakage, and motor failure.

[0020] Furthermore, in step 6, the hydraulic gate hoist pump station is maintained and managed according to the fault diagnosis results, and the diagnosis results are fed back to the system to update the model threshold library.

[0021] The beneficial effects of this invention are as follows:

[0022] 1. By fusing acoustic and vibration signals, this invention fully utilizes the complementary advantages of wide bandwidth of acoustic signals and strong anti-interference capability of vibration signals, breaking through the limitations of traditional single-mode diagnosis. It can more comprehensively reflect the operating status of hydraulic gate hoisting pump stations, improving the accuracy and reliability of fault diagnosis.

[0023] 2. The use of adaptive threshold-optimized wavelet denoising and variational mode decomposition to process sound signals, and adaptive empirical mode decomposition to process vibration signals, can effectively process non-stationary signals, separate noise from effective signals, extract more representative features, and improve the accuracy of feature extraction.

[0024] 3. The feature fusion algorithm based on the attention mechanism can dynamically allocate weights according to the contribution of features to the fault type, and the generated fused features are more discriminative, which helps to improve the accuracy of fault diagnosis.

[0025] 4. The optimized extreme gradient boosting tree classifier, combined with regularization parameter adaptive optimization and Bayesian optimization algorithm, improves the classifier's learning and generalization capabilities, enabling rapid and accurate fault diagnosis and meeting the needs of real-time on-site monitoring.

[0026] 5. This diagnostic method has good practicality and scalability. It is not only applicable to fault diagnosis of hydraulic gate hoist pump stations, but can also be extended to similar equipment such as fixed winch gate hoists, gantry cranes, and bridge cranes, and has broad engineering application value. Attached Figure Description

[0027] The invention will be further described below with reference to the accompanying drawings:

[0028] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] A fault diagnosis method for hydraulic gate hoist pump stations based on acoustic and vibration signal fusion includes the following steps:

[0031] Step 1: Signal Acquisition;

[0032] At pre-defined locations such as oil pumps, valves, and pipelines in the hydraulic gate hoist pump station, wide-frequency response acceleration and vibration sensors and interference-resistant, high-sensitivity microphones are installed to synchronously collect real-time acoustic and vibration signals during pump station operation (normal and fault thresholds are pre-stored in the system and do not need to be collected repeatedly). The specific deployment is as follows:

[0033] Oil pump monitoring points: Install IEPE acceleration vibration sensors (such as PCB352C65 type, sampling frequency 16kHz) in the axial and radial directions of the oil pump bearing housing to capture vibration signals caused by faults such as bearing wear and rotor imbalance; place an electret microphone (such as Knowles SPM0404LR5H-MB type, sampling frequency 32kHz) 10cm away from the pump body to collect sound signals generated by cavitation and friction of parts inside the pump.

[0034] Valve monitoring points: A uniaxial vibration sensor is installed on the top of the valve stem to monitor the vibration characteristics of the valve core during operation; sound sensors are placed 5cm from the valve body inlet and outlet to capture the sound characteristics of fluid impact and seal wear.

[0035] Pipeline monitoring points: Vibration sensors are installed along the pipeline axis at points of fluid disturbance such as pipe bends and tees; sound sensors are placed near flange connections to monitor high-frequency noise generated by leaks.

[0036] Motor monitoring points: An acceleration sensor is installed radially on the motor bearing housing, and an audio sensor is placed 10cm away from the motor housing to collect fault signals such as rotor eccentricity and bearing wear.

[0037] All sensors are connected to the data acquisition card (such as NI 9234) via a 485 bus. A hardware clock synchronization mechanism is used to ensure that the sampling time deviation of the acoustic and vibration signals is ≤1ms. The acquisition time covers two complete operating cycles of the equipment (e.g., if the opening and closing time of the hoist is 10s, then 20s of data will be collected) to ensure signal integrity.

[0038] Step 2: Sound signal processing;

[0039] Step 2.1: Adaptive Threshold Optimized Wavelet Denoising. Since the acquired sound signal is inevitably affected by environmental interference such as water flow noise and motor electromagnetic noise, an adaptive threshold optimized wavelet denoising algorithm is used to preserve fault characteristic sounds. Specifically, a smoothing factor is introduced to optimize the traditional soft threshold function, enabling it to better preserve the detailed features of the signal during processing. Simultaneously, the denoising threshold is dynamically adjusted based on the spectral characteristics of the sound signal to effectively suppress noise, thereby obtaining a cleaner sound signal.

[0040] Specifically, the audio signal is decomposed into 5 levels using the db4 wavelet, the spectrum is obtained through Fast Fourier Transform, the standard deviation σ of the noise frequency band (e.g., 0-3kHz) is calculated, and the dynamic threshold is determined. (Where σ is the standard deviation of the noise frequency band, and n is the signal length). A smoothing factor α = 0.7 is introduced to optimize the soft threshold function:

[0041]

[0042] In the formula: w j,k represents the k-th wavelet coefficient of the j-th layer after wavelet decomposition, sign(·) is the sign function, which returns the sign of the parameter (1, -1 or 0), and α is the smoothing factor (value 0.7), used to optimize the smoothness of the soft threshold function; thereby suppressing water flow noise, motor electromagnetic noise, and retaining fault characteristic sounds such as leakage impact sound (10-15kHz).

[0043] Step 2.2: Variational Mode Decomposition (VMD) is performed on the denoised audio signal. During the decomposition process, a dynamic penalty factor is introduced, which adaptively adjusts based on the local features of the audio signal to determine the optimal number of decomposed modes. Through variational mode decomposition, the audio signal is decomposed into multiple more physically meaningful modal components.

[0044] Specifically, a dynamic penalty factor α = 1200 is introduced to construct the VMD objective function:

[0045]

[0046] In the formula: u k (t) represents the time-domain signal of the k-th modal component; w k The center frequency of the k-th modal component; δ(t) is the Dirac function used for frequency domain localization; j is the imaginary unit (j 2 =-1); * is the convolution operator; α is the dynamic penalty factor (initial value 1200), used to control the decomposition accuracy; For u k The frequency domain representation of (t) (Fourier transform result), ‖·‖2 is L 2 Norm, used to measure the energy of a signal in the frequency domain; f k (ω) is the preset center frequency of the kth mode (used to constrain the decomposition process), and k is the number of decomposed modes (determined through adaptive calculation, such as k=6). The optimal number of decomposed modes K=6 is determined by iterative solution using the Alternating Direction Multiplier Method (ADMM), and the signal is decomposed into 6 modal components.

[0047] Step 2.3: Spectral entropy feature extraction. Perform FFT transformation on each modal component and calculate the spectral probability distribution.

[0048] In the formula: S(ω) i ) is the spectral value of the i-th frequency point (FFT transformation result), and pi is the spectral energy probability distribution of the i-th frequency point. Extract the spectral entropy H = -∑pi·lnpi, which is used to characterize the disorder of the signal frequency distribution and form a 6-dimensional sound feature vector Fs = [H1,H2,…,H6].

[0049] Spectral entropy can effectively characterize the nonlinear features of sound signals. Different fault types will cause the spectral entropy of sound signals to exhibit different characteristics. Therefore, spectral entropy features serve as a basis for judging fault types.

[0050] Step 3: Vibration signal processing;

[0051] Step 3.1: Adaptive Empirical Mode Decomposition (AEMD). To address the non-stationary characteristics of vibration signals, an adaptive empirical mode decomposition algorithm is employed. By using an adaptive stopping criterion designed based on the standard deviation of the vibration signal, the decomposition process is dynamically adjusted to effectively avoid mode aliasing. This decomposition process can decompose the vibration signal into a series of pure intrinsic mode function components, thereby separating fault characteristic components such as low-frequency vibrations caused by oil pump wear.

[0052] Specifically: The upper and lower envelopes of the vibration signal are generated through cubic spline interpolation. The envelope mean m(t) is calculated to obtain the initial component c(t) = y(t) - m(t). The design stopping criterion is based on standard deviation: when the standard deviation of adjacent decomposed components SD = ∑t|ci-1(t) - ci(t)|2 / ci-1 2 The decomposition terminates when (t) < 0.2, resulting in 8 intrinsic mode function (IMF) components. Each IMF must satisfy the following conditions: the difference between the number of extreme points and the number of zero crossing points is ≤ 1 and the mean of the envelope is zero.

[0053] Step 3.2: Composite Feature Extraction. From the IMF components obtained by adaptive empirical mode decomposition, the local mean decomposition energy operator and morphological gradient are extracted to form composite features. Among them, the local mean decomposition energy operator can highlight the energy changes of the vibration signal, while the morphological gradient can characterize the edge features of the vibration signal. The two complement each other and can more comprehensively describe the fault characteristics of the vibration signal.

[0054] The local mean decomposition energy operator includes: calculating the energy value E of the IMF component.

[0055] The 5-dimensional energy eigenvector E = [E1, E2, ..., E5] is obtained.

[0056] Morphological gradient: Dilation (⊕) and erosion (Θ) operations are performed on the IMF components using the structuring element B = [1,1].

[0057] In the formula, x is the input vibration signal (IMF component); The dilation operation is used to enhance signal edges; the erosion operation is used to suppress signal details; B is the structuring element (here it is [1,1], i.e., a linear window); G(x) is the morphological gradient result, used to characterize signal edge features; these constitute a 5-dimensional morphological feature vector G = [G1, G2, ..., G5], which is then merged to obtain a 10-dimensional vibration composite feature Fv = [E1, ..., E5, G1, ..., G5].

[0058] Step 4: Feature Fusion. An attention-based fusion algorithm based on a multilayer perceptron (MLP) is used to fuse the spectral entropy features of the sound signal and the composite features of the vibration signal. An attention weight calculation module is constructed, which dynamically assigns weights to each feature based on the contribution of sound and vibration features to different fault types. Through weighted fusion, a more discriminative sound-vibration fusion feature is generated. This fusion feature combines the advantages of sound and vibration signals and can more accurately reflect the fault state of the equipment.

[0059] The attention mechanism fusion algorithm based on multilayer perceptron (MLP) includes: inputting sound features Fs (6-dimensional) and vibration features Fv (10-dimensional), and calculating weights through two hidden layers (16 and 8 neurons, ReLU activation function):

[0060]

[0061] In the formula: F s F is the spectral entropy feature vector (6-dimensional) of the sound signal; v is the composite feature vector (10-dimensional) of the vibration signal; MLP(·) is the multilayer perceptron mapping function used to calculate feature importance; w s Attention weights for sound features; w v Attention weights for vibration characteristics;

[0062] This generates a 16-dimensional fusion feature F = ws·Fs + wv·Fv, and dynamically assigns feature weights to different fault types (e.g., ws = 0.65 for pipeline leakage, ws = 0.35 for oil pump wear).

[0063] Step 5: Fault Diagnosis; Fault diagnosis is performed using an optimized Extreme Gradient Boosting Tree (XGBoost) classifier. An adaptive optimization strategy for regularization parameters is introduced into the classifier, and a Bayesian algorithm is used to enhance the classifier's learning ability for acoustic-vibration fusion features. The generated acoustic-vibration fusion features are input into the optimized XGBoost classifier for training. The classifier can accurately classify the fault types of the hydraulic gate hoist pump station, outputting diagnostic results including normal operation, oil pump failure, valve failure, pipeline leakage, and motor failure. During the classification process, the real-time collected acoustic-vibration fusion features are compared with the normal and fault thresholds stored in the computer to determine the fault type.

[0064] For example, the parameters are optimized using a Bayesian optimization algorithm (the search space includes the penalty factor C∈[0.1,100], the regularization parameter λ∈[0.01,1], etc.), and the optimal parameters are finally determined (e.g., C=10, λ=0.05). The fused features are then input into the classifier and compared with a pre-stored threshold library.

[0065] Oil pump wear: spectral entropy > 0.85 and energy operator > 2.3;

[0066] Valve jamming: morphological gradient > 1.5 and spectral entropy > 0.7;

[0067] Pipeline leakage: spectral entropy > 0.65 and energy operator > 1.8;

[0068] Output the fault type and confidence level (e.g., "oil pump wear, confidence level 92%)".

[0069] Step 6: Result Application and Feedback; Generate maintenance recommendations based on the diagnostic results (e.g., immediate shutdown and repair for severe faults), and upload new fault data to the cloud. Update the threshold library using a sliding window method (window size 100 data sets): expand feature dimensions for newly added fault types, update the mean and standard deviation of thresholds for existing fault types, and optimize the adaptability of the diagnostic model.

[0070] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A fault diagnosis method for hydraulic gate hoist pump stations based on acoustic and vibration signal fusion, characterized in that, Includes the following steps: Step 1: Vibration sensors and sound sensors are installed at each preset point in the hydraulic gate hoist pump station to synchronously collect vibration and sound signals during the operation of the pump station. Step 2: Perform adaptive threshold optimization wavelet denoising on the audio signal, then perform variational mode decomposition to extract the spectral entropy features of the decomposed modal components; Step 3: Perform adaptive empirical mode decomposition on the vibration signal and extract the composite features composed of the local mean decomposition energy operator and morphological gradient of the decomposed IMF components. Step 4: Employ an attention-based feature fusion algorithm to weightedly fuse the spectral entropy features of the sound signal and the composite features of the vibration signal to generate sound-vibration fusion features; Step 5: Use an extreme gradient boosting tree classifier to learn and train the acoustic-vibration fusion features to obtain fault diagnosis results; Step 6: Evaluate and provide feedback on the operating status of the hydraulic gate hoist pump station based on the fault diagnosis results.

2. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 2, the adaptive threshold-optimized wavelet denoising process includes introducing a smoothing factor to optimize the soft threshold function and dynamically adjusting the denoising threshold based on the spectral characteristics of the sound signal.

3. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 2, a dynamic penalty factor is introduced during variational mode decomposition, and the number of decomposed modes is adaptively determined based on the local features of the sound signal.

4. The method for fault diagnosis of hydraulic gate hoist pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 3, the adaptive empirical mode decomposition dynamically adjusts the decomposition process using an adaptive stopping criterion designed based on the standard deviation of the vibration signal.

5. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 4, an attention weight calculation module is constructed based on the feature fusion algorithm of the attention mechanism, and the weights are dynamically allocated according to the contribution of sound and vibration features to different fault types.

6. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 5, the optimized extreme gradient boosting tree classifier introduces a regularization parameter adaptive optimization strategy and enhances its learning ability for fused features with the help of a Bayesian optimization algorithm.

7. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 1, the vibration sensor is an accelerometer, and the sound sensor is an anti-interference microphone.

8. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 3, the local mean decomposition energy operator is used to highlight the energy changes of the vibration signal, and the morphological gradient is used to characterize the edge features of the vibration signal.

9. The method for fault diagnosis of hydraulic gate hoisting pump stations based on acoustic and vibration signal fusion according to claim 1, characterized in that, In step 5, the fault diagnosis results include normal operation, oil pump failure, valve failure, pipeline leakage, and motor failure.

10. A method for fault diagnosis of hydraulic gate hoist pump stations based on acoustic-vibration signal fusion according to claim 1, characterized in that, In step 6, the hydraulic gate hoist pump station is maintained and managed according to the fault diagnosis results, and the diagnosis results are fed back to the system to update the model threshold library.

Images