Fault diagnosis method for electric actuating mechanism

Abstract

The invention relates to the technical field of industrial control, and provides a fault diagnosis method for an electric actuating mechanism, which comprises the following steps of: 1, acquiring operation data of the electric actuating mechanism in real time at a sampling frequency of 1kHz to 10kHz through a multi-sensor fusion technology, comprising a combination of a Hall sensor, an acceleration sensor and an infrared temperature sensor and comprises motor current I (t), voltage V (t), vibration signals A (t) and temperature data T (t), operation data are collected in a segmented mode through the sliding window technology, and the window length is dynamically adjusted according to equipment loads. According to the method, current, voltage, vibration and temperature signals are fused, the weight distribution is 0.3: 0.2: 0.4: 0.1, the information utilization rate is improved, the wavelet energy coefficient (db4 wavelet basis, 3-5-layer decomposition) and the FFT frequency spectrum peak value (the frequency resolution delta f is smaller than or equal to 9.77 Hz) are combined, the early fault recognition sensitivity is improved by 30%, the characteristic stability under different working conditions is guaranteed, and the anti-interference capacity is improved by 40% or above.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial control, and particularly to a fault diagnosis method for an electric actuator. Background Art

[0002] In large and complex production processes, actuators are characterized by wide distribution, large quantity, and often relatively special installation positions. Electric actuators have developed the fastest in recent years and their application scope has been continuously expanding due to the advantages of using electricity as power, which are incomparable to several other types of media. As the generating units put into operation by power enterprises in China are increasingly developing towards large-scale, precision, and high automation, and the requirements for the stability and economy of the units are continuously increasing, ensuring the high reliability of equipment, diagnosing and eliminating faults in a timely manner have become important contents of equipment management and maintenance. For large and complex production processes, the demand for fault detection and diagnosis of the control system is increasing day by day. It is very difficult to detect and repair actuator faults in a timely, accurate, and effective manner only by relying on human power.

[0003] Existing fault diagnosis technologies for electric actuators mostly rely on a single sensor (such as a current or vibration sensor), are easily interfered by environmental noise, and have insufficient information dimensions, resulting in a fault recognition accuracy rate of less than 85% under complex working conditions. At the same time, the fixed threshold mechanism cannot adapt to the dynamic changes of the load, and the false alarm rate is as high as 15%, making it difficult to meet the industrial real-time requirements; traditional methods only use single analysis in the time domain or frequency domain, and have weak ability to distinguish complex fault modes such as bearing faults and winding short circuits. For example, the frequency domain analysis based on FFT has insufficient spectral resolution (Δf≥10Hz) under low-frequency noise interference, resulting in difficulty in capturing early fault characteristics; the generalization ability of traditional machine learning models (such as decision trees, KNN) under multiple working conditions is limited, and the classification accuracy rate is generally less than 90%. In addition, sensor calibration relies on manual intervention, and the signal-to-noise ratio (SNR) decreases due to device aging after long-term operation of the system, affecting the diagnostic stability. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides a fault diagnosis method for an electric actuator, which solves the problems of relying on a single sensor (such as a current or vibration sensor), being easily interfered by environmental noise, insufficient information coverage under multiple working conditions, resulting in a complex fault recognition accuracy rate of less than 85%, being unable to adapt to the dynamic changes of the load and the characteristics of environmental noise, with a false alarm rate as high as 15%, being difficult to meet the industrial real-time requirements, and lacking a self-calibration mechanism when the signal-to-noise ratio (SNR) decreases due to sensor aging.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A fault diagnosis method for an electric actuator includes the following steps:

[0006] Step 1: Real-time collect the operation data of the electric actuator at a sampling frequency of 1 kHz to 10 kHz. The operation data is collected through multi-sensor fusion technology, including a combination of Hall sensors, acceleration sensors, and infrared temperature sensors, including motor current I(t), voltage V(t), vibration signal A(t), and temperature data T(t). And the sliding window technology is used to segment and collect the operation data, and the window length is dynamically adjusted according to the device load;

[0007] Step 2: Preprocess the collected data, including noise filtering, signal normalization, and feature extraction. Feature extraction includes time-domain analysis, frequency-domain analysis, and time-frequency joint analysis, and extracts at least one of the mean value, variance, spectral peak value, and wavelet energy coefficient. The adaptive filtering algorithm is used to adjust the filter parameters in real time according to the environmental noise spectrum, and the Butterworth low-pass filter with a cut-off frequency of f c = 2f 额定 is used to filter the noise, where f rated is the rated frequency of the motor, and the signal is normalized:

[0008]

[0009] where μ X is the mean value, and σ X is the standard deviation;

[0010] Step 3: Based on the preset fault feature library, classify the extracted features through a support vector machine (SVM) or a deep neural network, and the fault type recognition accuracy rate ≥ 95%;

[0011] Step 4: Generate a diagnostic report and output it through the human-machine interface, and the response time ≤ 200 ms.

[0012] Preferably, the frequency-domain analysis in Step 2 uses the fast Fourier transform (FFT), and the frequency resolution is where f8 is the sampling frequency (1 kHz ≤ f8 ≤ 10 kHz), and N is the number of sampling points (1024 ≤ N ≤ 4096), and the spectral peak value P peak = max(|F(k)|), where F(k) is the frequency-domain signal.

[0013] Preferably, the support vector machine (SVM) uses the Gaussian kernel function where the value range of the kernel width σ is 0.1 ≤ σ ≤ 1.0, and the classification error rate ≤ 3%.

[0014] Preferably, the dynamic threshold adjustment mechanism updates the fault determination threshold through the sliding average formula:

[0015] T(t) = aT(t - 1) + (1 - a)X(t) (0.8 ≤ a ≤ 0.95)

[0016] When |X(t)-T(t)|≥3σ X An early warning is triggered.

[0017] Preferably, the accuracy of the Hall sensor is ±0.5%, the sampling rate of the acceleration sensor is 5 kHz, the resolution of the temperature sensor is 0.1 °C, and the multi-sensor data fusion weights satisfy:

[0018] W I :W V :W A :W T =0.3:0.2:0.4:0.1

[0019] where W I 、W V 、W A and W T are the weight coefficients of current, voltage, vibration, and temperature respectively.

[0020] Preferably, the deep neural network is a 4- to 6-layer convolutional neural network (CNN), the activation function is ReLU, the loss function is cross-entropy, and the learning rate η satisfies:

[0021] η = 0.001×e -0.01t (t≥0)

[0022] where t is the number of training iterations.

[0023] Preferably, the length L of the sliding window is dynamically adjusted according to the load:

[0024]

[0025] Preferably, the calculation of the wavelet energy coefficient uses the Daubechies wavelet basis (db4), the decomposition level is 3 to 5 layers, and the energy coefficient of the j-th layer is:

[0026]

[0027] where D j (k) is the detail coefficient of the j-th layer.

[0028] Preferably, the health assessment of the self-check module is determined by the signal-to-noise ratio (SNR). Specific embodiments

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0030] Example:

[0031] An embodiment of the present invention provides a method for fault diagnosis of an electric actuator, including the following steps:

[0032] Step 1: Real-time collect the operation data of the electric actuator at a sampling frequency of 1 kHz to 10 kHz. The operation data is collected through multi-sensor fusion technology, including a combination of a Hall sensor, an acceleration sensor, and an infrared temperature sensor. The accuracy of the Hall sensor is ±0.5%, the sampling rate of the acceleration sensor is 5 kHz, the resolution of the temperature sensor is 0.1 °C, and the multi-sensor data fusion weights satisfy: W I :W V :W A :W T = 0.3:0.2:0.4:0.1, where W I 、W V 、W A and W T are the weight coefficients of current, voltage, vibration, and temperature respectively, and the sliding window technology is used to segment and collect the operation data. The window length is dynamically adjusted according to the device load. The sliding window length L is dynamically adjusted according to the load:

[0033]

[0034] Step 2: Preprocess the collected data, including noise filtering, signal normalization, and feature extraction. Feature extraction includes time-domain analysis, frequency-domain analysis, and time-frequency joint analysis, and at least one of mean value, variance, spectral peak value, and wavelet energy coefficient is extracted;

[0035] Frequency-domain analysis uses fast Fourier transform (FFT), and the frequency resolution is where f8 is the sampling frequency (1 kHz ≤ f8 ≤ 10 kHz), and N is the number of sampling points (1024 ≤ N ≤ 4096),

[0036] Extract the spectral peak value P peak = max(|F(k)|), where F(k) is the frequency-domain signal,

[0037] Use a Butterworth low-pass filter with a cut-off frequency of f c = 2f 额定 to filter out noise, where f rated is the rated frequency of the motor, and normalize the signal:

[0038]

[0039] where μ X is the mean value, and σ X is the standard deviation;

[0040] The wavelet energy coefficient is calculated using the Daubechies wavelet basis (db4), and the decomposition level is from 3 to 5 layers. The energy coefficient of the j-th layer is:

[0041]

[0042] where D j (k) is the detail coefficient of the j-th layer;

[0043] The adaptive filtering algorithm is adopted to adjust the filter parameters in real time according to the environmental noise spectrum. The parameter update formula of the adaptive filter is:

[0044] where the orders n and m ≤ 4, and the cut-off frequency dynamically matches ±10% of the main frequency of the environmental noise.

[0045] Step 3: Based on the preset fault feature library, classify the extracted features through a support vector machine (SVM) or a deep neural network, and the accuracy rate of fault type recognition ≥ 95%;

[0046] The support vector machine (SVM) adopts a Gaussian kernel function where the value range of the kernel width σ is 0.1 ≤ σ ≤ 1.0, and the classification error rate ≤ 3%;

[0047] The dynamic threshold adjustment mechanism updates the fault determination threshold through a moving average formula: T(t) = aT(t - 1) + (1 - a)X(t) (0.8 ≤ a ≤ 0.95). When |X(t) - T(t)| ≥ 3σ X a warning is triggered;

[0048] The deep neural network is a 4 - to - 6 - layer convolutional neural network (CNN), the activation function is ReLU, the loss function is cross - entropy, and the learning rate η satisfies: η = 0.001×e -0.01t (t ≥ 0), where t is the number of training iterations.

[0049] Step 4: Generate a diagnostic report and output it through a human - machine interface. The response time ≤ 200ms, and it also includes a self - test module that regularly evaluates the health status of the sensor and the data processing module to ensure the reliability of the diagnostic system. The health assessment of the self - test module is determined by the signal - to - noise ratio (SNR):

[0050]

[0051] If the SNR < 20dB, a sensor calibration instruction is triggered.

[0052] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing faults of an electric actuator, characterized in that: The following steps are involved: Step 1: The operation data of the electric actuator is collected in real time at a sampling frequency of 1kHz to 10kHz. The operation data is collected through multi-sensor fusion technology, including a combination of Hall sensor, acceleration sensor and infrared temperature sensor, including motor current I(t), voltage V(t), vibration signal A(t) and temperature data T(t). The sliding window technology is used to collect the operation data in segments, and the window length is dynamically adjusted according to the equipment load. Step 2: preprocess the collected data, including noise filtering, signal normalization and feature extraction. Feature extraction includes time domain analysis, frequency domain analysis and time-frequency joint analysis. At least one of the mean, variance, spectrum peak and wavelet energy coefficient is extracted. Adaptive filtering algorithm is used to adjust the filter parameters in real time according to the ambient noise spectrum. The cutoff frequency is f. c =2f 额定 The Butterworth low-pass filter removes the noise, where frated is the rated frequency of the motor, and the signal is normalized: where μ X is the mean, σ X is the standard deviation; Step 3: Based on the preset fault feature library, the extracted features are classified by support vector machine (SVM) or deep neural network, and the fault type recognition accuracy is ≥ 95%; Step 4: Generate a diagnostic report and output it through the human-machine interface. The response time is ≤200ms.

2. The electric actuator fault diagnosis method according to claim 1, characterized in that: The frequency domain analysis in step 2 uses fast Fourier transform (FFT) with a frequency resolution of Where f8 is the sampling frequency (1kHz≤f8≤10kHz), N is the number of sampling points (1024≤N≤4096), and the spectrum peak value P is extracted. peak =max(|F(k)|), where F(k) is the frequency domain signal.

3. The electric actuator fault diagnosis method according to claim 1, characterized in that: The support vector machine (SVM) uses a Gaussian kernel function The value range of kernel width σ is 0.1≤σ≤1.0, and the classification error rate is ≤3%.

4. The electric actuator fault diagnosis method according to claim 1, characterized in that: The dynamic threshold adjustment mechanism updates the fault judgment threshold through a sliding average formula: T(t)=aT(t-1)+(1-a)X(t)(0.8≤a≤0.95) When |X(t)-T(t|≥3σ X The warning is triggered.

5. The electric actuator fault diagnosis method according to claim 1, characterized in that: The accuracy of the Hall sensor is ±0.5%, the sampling rate of the acceleration sensor is 5kHz, the resolution of the temperature sensor is 0.1°C, and the multi-sensor data fusion weight satisfies: IN I :IN V :IN A :IN T =0.3:0.2:0.4:0.1 Where W I , W V , W A and W T are the weight coefficients of current, voltage, vibration and temperature respectively.

6. The electric actuator fault diagnosis method according to claim 1, characterized in that: The deep neural network is a 4-6 layer convolutional neural network (CNN), the activation function is ReLU, the loss function is cross entropy, and the learning rate η satisfies: η=0.001×e -0.01t (t≥0) Where t is the number of training iterations.

7. The electric actuator fault diagnosis method according to claim 1, characterized in that: The sliding window length L is dynamically adjusted according to the load:

8. The electric actuator fault diagnosis method according to claim 1, characterized in that: The wavelet energy coefficient calculation adopts Daubechies wavelet basis (db4), the decomposition level is 3 to 5, and the energy coefficient of the jth level is: Where D j (k) is the detail coefficient of the jth layer.

9. The electric actuator fault diagnosis method according to claim 1, characterized in that: The health assessment of the self-check module is determined by the signal-to-noise ratio (SNR): If SNR < 20dB, the sensor calibration command is triggered.

10. The electric actuator fault diagnosis method according to claim 1, characterized in that: The parameter update formula of the adaptive filter is: The orders n and m are ≤ 4, and the cut-off frequency dynamically matches ±10% of the main frequency of the ambient noise.

Images