Electro-hydraulic actuator state monitoring method based on sound sensing information

Abstract

The application discloses a kind of electro-hydraulic actuator state monitoring methods based on sound sensing information, comprising the following steps: (1) collecting high-frequency sound signal under normal and fault state of electro-hydraulic actuator;(2) the high-frequency sound signal collected is de-noised;(3) the sound signal after de-noising is calculated to obtain the sound equivalent DC value;(4) the equivalent DC value obtained after calculation is extracted in time-frequency domain characteristics;(5) based on the sound signal after feature extraction under normal and fault state, the state of electro-hydraulic actuator is monitored.The present application makes full use of the advantages of high-frequency band response sensitivity, flexible arrangement and no interference to the hydraulic system of sound sensor, significantly improves the reliability and generalization ability of electro-hydraulic actuator state monitoring and fault early warning.

Description

Technical Field

[0001] This invention belongs to the field of electro-hydraulic equipment condition monitoring technology, and specifically relates to a method for monitoring the condition of electro-hydraulic actuators based on sound sensing information. Background Technology

[0002] Electro-hydraulic actuators, as highly integrated electro-hydraulic actuation systems, combine components such as motors, plunger pumps, oil tanks, and actuator cylinders into a single unit. With their significant advantages such as high power density and fast dynamic response, they have been widely used in aerospace, shipbuilding, robotics, and high-end industrial automation. However, as equipment develops towards higher reliability, longer lifespan, and maintenance-free operation, higher demands are placed on real-time monitoring and early fault warning of electro-hydraulic actuator operating status. Under long-term high loads and complex operating conditions, critical components such as pumps, motors, and actuator cylinders are prone to wear, leakage, and cavitation. Failure to detect these faults in a timely manner can lead to system performance degradation or even catastrophic consequences. Therefore, electro-hydraulic actuator condition monitoring technology covering the entire life cycle has become a crucial foundation for ensuring their stable operation.

[0003] In existing technologies, the condition monitoring of electro-hydraulic actuators mainly relies on traditional methods such as pressure sensors, temperature sensors, and current / speed sensors. While these methods can reflect some fault characteristics to a certain extent, they generally have the following shortcomings: First, the sensor installation location is limited and the wiring is complex, which is not conducive to system lightweighting and integration; second, the sensitivity to early and weak faults (such as slight wear, initial cavitation, etc.) is insufficient, making it difficult to achieve early fault diagnosis; third, the multi-sensor information fusion algorithm is complex and computationally intensive, making it difficult to guarantee real-time performance in resource-constrained embedded systems. The acoustic signals generated during actuator operation contain a large amount of dynamic characteristic information and have advantages such as non-contact operation, wide bandwidth, and high sensitivity. However, there is currently no mature method in the existing technology that systematically utilizes acoustic features for multi-fault type identification and condition monitoring of electro-hydraulic actuators, indicating a significant technological gap. Therefore, this invention proposes a condition monitoring method for electro-hydraulic actuators based on sound sensing information. By collecting sound signals from electro-hydraulic actuators under different states, an acoustic database is established for condition monitoring of electro-hydraulic actuators. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the prior art and propose a method for monitoring the status of electro-hydraulic actuators based on sound sensing information, which is used to monitor the electro-hydraulic actuator system in real time through sound signals.

[0005] To achieve the above objectives, this invention proposes a method for monitoring the state of an electro-hydraulic actuator based on sound sensing information, comprising the following steps:

[0006] (1) Based on the preset sound sensor layout, high-frequency sound signals of the electro-hydraulic actuator under normal and fault conditions are collected;

[0007] (2) Noise reduction is performed on the collected high-frequency sound signals, including scene-specific noise suppression and general noise suppression;

[0008] (3) The RMS (Root Mean Square) effective value of the noise-reduced sound signal is calculated in a time window of 0.05s to obtain the equivalent DC value of the sound signal;

[0009] (4) The equivalent DC value obtained after calculation is subjected to time-frequency domain feature extraction, including time-frequency domain features of kurtosis, peak-to-peak value, skewness, spectral centroid and spectral entropy;

[0010] (5) Based on the sound signals extracted from features under normal and fault conditions, the electro-hydraulic actuator is monitored for status.

[0011] Preferably, the sound sensor is arranged in a position near the motor end of the electro-hydraulic actuator motor housing, plunger pump housing, actuator cylinder, or electro-hydraulic actuator test bench, wherein the sampling frequency is set to 10kHz or higher.

[0012] Preferably, the fault states of the electro-hydraulic actuator include, but are not limited to, inter-turn short circuit of the motor, motor demagnetization, current imbalance, slipper detachment, high-speed air suction, slipper pull-out, displacement interference, displacement offset, displacement drift, wear of the actuator cylinder seal ring, and wear of the actuator cylinder mechanical seal.

[0013] Preferably, step (2) specifically includes:

[0014] The sound signals collected by the sound sensor are fused using two noise reduction methods: scene-specific noise suppression and general noise suppression. Scene-specific noise suppression refers to training an algorithm model for specific noise types to reduce noise from irregular and unpredictable sounds in the test environment, including the operating sound of air-cooled equipment, the working sound of air compressors, and the sound of vent valves opening. General noise suppression refers to reducing noise from noise sources that are unrelated to specific faults, occur under various operating conditions, and are difficult to completely separate through scene modeling.

[0015] Preferably, step (3) specifically involves:

[0016] From the noise-reduced audio signal curve, the RMS effective value is extracted in a time window of 0.05s, converting all AC signals of the audio into DC effective values ​​to ensure that the information loss of the audio signal curve is minimized and all values ​​are greater than 0.

[0017] Preferably, step (4) specifically involves:

[0018] The DC RMS value of the sound signal is subjected to time-domain and frequency-domain feature transformations, including kurtosis, peak-to-peak value, skewness, spectral centroid, and spectral entropy. After the time-frequency domain feature transformations, five new feature values ​​are obtained. These five new feature values ​​are then input into the phase space transformation model and the clustering model to further extract the sound characteristics of the electro-hydraulic actuator under different fault conditions.

[0019] Preferably, step (5) specifically involves:

[0020] After performing transformations 2-4 in the technical solution on the sound signals under normal and fault conditions, the sound signals can be input into the fault diagnosis algorithm model. The fault diagnosis algorithm model outputs whether the current state is normal or faulty. If it is a faulty state, it will display the specific type of fault that has occurred. The fault diagnosis algorithm model is a machine learning model.

[0021] The beneficial effects of this invention are:

[0022] The method of this invention requires no modification to the electro-hydraulic actuator structure, allows for flexible sensor placement, and can be directly used for condition monitoring in practical engineering. Utilizing the wide frequency response range of sound signals, it can accurately capture fault characteristics of early wear in mechanical components. Furthermore, combined with signal processing algorithms, this invention can accurately extract key status information even under strong background noise and varying operating conditions, achieving low-cost, high-sensitivity condition monitoring of electro-hydraulic actuator systems. This invention has broad application value in aerospace, high-end industrial equipment, and intelligent operation and maintenance fields.

[0023] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description

[0024] Figure 1 This is the overall flowchart of the present invention.

[0025] Figure 2 This is a flowchart of scene-specific noise suppression and general noise suppression in the embodiments of the present invention.

[0026] Figure 3 This is a schematic diagram of the original sound data curve in an embodiment of the present invention, with the horizontal axis representing time (unit: seconds) and the vertical axis representing audio amplitude.

[0027] Figure 4 This is a schematic diagram of the sound data curve after noise reduction in an embodiment of the present invention. The horizontal axis represents time (unit: seconds), and the vertical axis represents the audio amplitude.

[0028] Figure 5This is a schematic diagram of the effective value extraction of the sound data curve after noise reduction in an embodiment of the present invention. Its horizontal axis is time (unit: seconds) and its vertical axis is the audio amplitude.

[0029] Figure 6 This is a time-frequency domain phase space diagram of the effective sound value data in this embodiment of the invention. Its horizontal axis represents the effective value with a delay of 5 milliseconds, and its vertical axis represents the effective value.

[0030] Figure 7 This is a time-frequency domain clustering diagram of the effective sound value data in an embodiment of the present invention. Its horizontal and vertical axes are dimension 1 and dimension 2 after dimensionality reduction, respectively. In the diagram, solid circles represent normal state samples, and crosses represent fault state samples. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the present invention includes the following steps:

[0033] 1) Based on a preset sound sensor layout, high-frequency sound signals are collected from the electro-hydraulic actuator under normal and fault conditions. The sound sensor layout involves placing the sound sensors near the motor housing, plunger pump housing, actuator cylinder, or electro-hydraulic actuator test bench, with a sampling frequency set to 10 kHz or higher. Fault conditions of the electro-hydraulic actuator include, but are not limited to, inter-turn short circuits in the motor, motor demagnetization, current imbalance, slipper detachment, high-speed air intake, slipper pull-out, displacement interference, displacement offset, displacement drift, wear of the actuator cylinder seal ring, and wear of the actuator cylinder mechanical seal.

[0034] In this embodiment, the sound sensor is attached to the housing of the electro-hydraulic actuator test bench, close to the motor end. The straight-line distance between the sound sensor and the electro-hydraulic actuator is less than 30 cm. Considering that the maximum speed of the permanent magnet synchronous motor is 12,000 rpm and the number of plungers in the axial plunger pump is 9, the sound sensor is mainly used to monitor the fundamental frequency and harmonics of the electro-hydraulic actuator. Therefore, the sampling frequency of the sound sensor is set to 20 kHz. Considering the superposition of environmental noise and equipment fault noise, the upper limit of the dynamic range of the sound sensor is 164 dB, and the typical operating current excitation is 4 mA.

[0035] 2) Noise reduction of the acquired high-frequency sound signals, including scene-specific noise suppression and general noise suppression. The sound signals acquired by the sound sensor are fused using two noise reduction methods: scene-specific noise suppression and general noise suppression. Scene-specific noise suppression refers to training an algorithm model for specific noise types to reduce noise from irregularly occurring, unpredictable, and categorized sounds in the experimental environment, including the operating sounds of air-cooled equipment, air compressors, and vent valves. General noise suppression refers to reducing noise from sources that are unrelated to specific faults, occur under various operating conditions, and are difficult to completely isolate through scene modeling.

[0036] In this embodiment, under the scenario of electro-hydraulic actuator status monitoring, a cascaded noise suppression framework is constructed to preprocess the collected high-frequency sound signals, thereby achieving synergistic suppression of general noise and scenario-specific noise. The process of constructing the cascaded noise suppression framework is as follows: Figure 2 As shown in the diagram. First, a high-frequency sound sensor collects the raw sound signal, which is then transmitted to the system. The raw sound signal is as follows: Figure 3 As shown, the signals are then sent to the scene-specific noise suppression branch and the general noise suppression branch, respectively, according to the system architecture.

[0037] For the scene-specific noise suppression branch, specific noise signals are collected from the operating sound of the air-cooled equipment, the working sound of the air compressor, and the opening sound of the vent valve. :

[0038] ;

[0039] in For continuous time variables, For the target sound source, Scene-specific noise;

[0040] The collected scene-specific noise is input into the U-Net (a U-shaped convolutional neural network) spectrogram denoising model for training. The U-Net model first converts the signal into an amplitude spectrum. :

[0041] ;

[0042] in (Short-Time Fourier Transform) is the short-time Fourier transform operator; Indicates frequency index; Indicates the time frame index;

[0043] The U-Net model outputs a mask. :

[0044] ;

[0045] in Indicates parameters The U-Net network mapping function is represented;

[0046] Reconstruct the clean spectrogram based on the mask and amplitude spectrum. :

[0047] ;

[0048] Finally, by inverse transforming the reconstructed clean spectrogram to the time domain, the time-domain enhanced speech signal output by the scene-specific noise suppression branch can be obtained. :

[0049] ;

[0050] in (Inverse Short-Time Fourier Transform) represents the inverse short-time Fourier transform operator; The imaginary unit;

[0051] For the general noise suppression branch, acquire audio signals containing background noise. :

[0052] ;

[0053] in For the target sound source, This is general noise.

[0054] The collected background noise is estimated using spectral subtraction. Spectral subtraction noise estimation first involves performing a short-time Fourier transform on the signal.

[0055] ;

[0056] Then, the noise power spectrum is estimated using the silent segment. :

[0057] ;

[0058] in, For mathematical expectation operators; This is the set of time frame indices corresponding to the silent segment;

[0059] After obtaining the estimated noise power spectrum of the silent segment, a spectral subtraction operation is performed to deduce the amplitude of the spectral subtraction. ;

[0060] ;

[0061] in and These are the over-subtraction factor and the noise floor factor, respectively. For finding the maximum value operator;

[0062] Finally, the time-domain signal is reconstructed to obtain the time-domain speech output of the general noise suppression branch. :

[0063] ;

[0064] in The phase spectrum corresponding to the noisy speech;

[0065] To achieve better overall noise reduction, this invention performs weighted fusion of the outputs of the general noise suppression branch and the scene-specific noise suppression branch. The fused signal is the final enhanced audio signal. :

[0066] ;

[0067] in and These are the weights for general noise suppression and scene-specific noise suppression, respectively.

[0068] The sound curve after noise reduction is as follows Figure 4 As shown;

[0069] 3) Calculate the RMS effective value of the noise-reduced audio signal using a time window of 0.05s to obtain the equivalent DC value of the audio signal. Convert all AC signals of the audio signal into DC effective values ​​to ensure minimal information loss in the audio signal curve and that all effective values ​​of the audio signal are greater than 0.

[0070] In this embodiment, the RMS effective value of the noise-reduced audio signal is calculated using a sampling frequency of 20 kHz and a time window of 0.05 s. The formula for calculating the RMS effective value is as follows:

[0071] ;

[0072] in Let be the signal function, and T be the period. The sound curve after RMS transformation is as follows: Figure 5 As shown.

[0073] 4) The calculated equivalent DC value is used for time-frequency domain feature extraction, including kurtosis, peak-to-peak value, skewness, spectral centroid, and spectral entropy. The effective DC value of the sound signal is subjected to time-domain and frequency-domain feature transformations, including kurtosis, peak-to-peak value, skewness, spectral centroid, and spectral entropy. After time-frequency domain feature transformation, five new feature values ​​are obtained. These five new feature values ​​are input into the phase space transformation model and clustering model to further extract the sound characteristics of the electro-hydraulic actuator under different fault conditions.

[0074] In this embodiment, the equivalent DC value processed with a time window of 0.05s is subjected to time-frequency domain transformation, and the kurtosis calculation formula is as follows:

[0075] ;

[0076] The formula for calculating peak-to-peak value is:

[0077] ;

[0078] The formula for calculating skewness is:

[0079] ;

[0080] The formula for calculating the spectral centroid is:

[0081] ;

[0082] The formula for calculating spectral entropy is:

[0083] ;

[0084] in, The total number of samples; For signal samples; The mean of the sample; The standard deviation is the sample standard deviation. For discrete-time indexing; Frequency index; This is the actual frequency value; Power spectrum; This is the spectral entropy value; The total number of frequencies to be divided into in the spectrum;

[0085] Combining the feature vectors obtained from each time window yields:

[0086] ;

[0087] In this embodiment, Feature vector combinations are combined to perform phase space transformation and t-SNE (t-distributed stochastic neighbor embedding) clustering to further obtain the state monitoring features of the electro-hydraulic actuator, such as... Figure 6 , Figure 7 As shown. Phase space transformation and phase space reconstruction adopt the delayed coordinate method. Let the sample index be... If the embedding dimension is 𝑚 and the delay time is 𝜏, then the phase space vector is:

[0088] ;

[0089] t-SNE transforms the similarity of high-dimensional features into a probability distribution, and then approximates this distribution in a low-dimensional space. Therefore, the optimization objective of t-SNE is:

[0090] ;

[0091] in, It represents the similarity of samples in a high-dimensional space; It represents the similarity of samples in a low-dimensional space.

[0092] 5) Based on the feature-extracted sound signals under normal and fault conditions, the electro-hydraulic actuator is status monitored. After performing the transformations described in steps 2)-4) above on the sound signals under normal and fault conditions, the sound signals can be input into the fault diagnosis algorithm model. The fault diagnosis algorithm model outputs whether the current state is normal or faulty. If it is a faulty state, it will display the specific type of fault that has occurred. The fault diagnosis algorithm model is a machine learning model.

[0093] In this embodiment, the algorithm model for predicting fault state and fault type is the TabPFN (Tabular Prior-Data Fitted Networks) classification model. The input of the TabPFN classification model is the RMS effective value, kurtosis, peak-to-peak value, skewness, spectral centroid, spectral entropy, phase space transformation, and t-SNE clustering features at 0.05s. The output is the current state category. When the output is 0, it is a normal state. When the output is other integers, it corresponds to different electro-hydraulic actuator fault types.

[0094] The above embodiments are illustrative of the present invention and are not intended to limit the present invention. Any simple modifications to the present invention are within the scope of protection of the present invention.

Claims

1. A method of electro-hydraulic actuator condition monitoring based on acoustic sensing information, characterized by, Includes the following steps: (1) High-frequency sound signals under normal and various preset fault states are collected by sound sensors arranged on key components of the electro-hydraulic actuator; (2) The high-frequency sound signal is subjected to noise reduction processing, which includes cascaded scene-specific noise suppression and general noise suppression, and the two suppressed signals are weighted and fused. (3) Calculate the effective value of the fused and denoised sound signal using a preset time window, and convert the AC signal into an equivalent DC value; (4) Extract time-frequency domain features from the equivalent DC value to obtain a time-frequency domain feature vector containing kurtosis, peak-to-peak value, skewness, spectral centroid and spectral entropy; (5) Input the time-frequency domain feature vector into a trained machine learning classification model to identify the current operating state or specific fault type of the electro-hydraulic actuator.

2. The method of claim 1, wherein, In step (1), the key components include the motor housing, the plunger pump housing, the actuator cylinder, or the test bench near the motor end; the sampling frequency of the sound sensor is not less than 10kHz.

3. The method of claim 1, wherein, In step (2), the scene-specific noise suppression adopts the spectrogram denoising method based on the U-Net model. The suppressed noise includes the operating sound of the air-cooled machine, the working sound of the air compressor, and the opening sound of the vent valve. The general noise suppression method employs spectral subtraction based on noise estimation in silent segments.

4. The method of claim 1, wherein, In step (3), the preset time window is 0.05 seconds, and the RMS effective value calculation converts the AC signal of the sound into an equivalent DC value with minimal loss and all values ​​greater than 0.

5. The method of claim 1, wherein, In step (4), after obtaining the time-frequency domain feature vector, it is further input into the phase space transformation model and the t-SNE clustering model to separate feature clusters under different states in a low-dimensional space.

6. The method of claim 1, wherein, In step (5), the machine learning classification model is the TabPFN model.

7. The method according to claim 1, characterized in that, The preset fault states include at least one of the following: motor inter-turn short circuit, motor demagnetization, current imbalance, slipper detachment, high-speed air suction, slipper pull-out, displacement interference, displacement offset, displacement drift, actuator cylinder seal wear, and actuator cylinder mechanical seal wear.

8. A state monitoring system for an electro-hydraulic actuator based on sound sensing information, characterized in that, include: The sound data acquisition unit is used to acquire sound characterization data of different preset fault states during the testing of the electro-hydraulic actuator. The sound noise reduction unit integrates scene-specific noise suppression and general noise suppression methods to suppress background noise from different sources based on different environmental background noise. The effective value calculation unit is used to calculate the effective value of the noise-reduced sound signal and convert the AC signal into a DC effective value signal. The time-frequency domain feature extraction unit is used to extract kurtosis, peak-to-peak value, skewness, spectral centroid and spectral entropy features, and input the new feature values ​​into the phase space transformation model and clustering model to observe the acoustic characteristics of the electro-hydraulic actuator under different faults; The fault diagnosis result generation unit is used to predict the current operating status of the electro-hydraulic actuator by taking the transformed sound characteristics as input.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7; or, a computer program product comprising the computer program / instructions.

Images