Bearing early fault diagnosis method based on weighted segmented random sink pool network

Abstract

The invention provides a bearing early-stage fault diagnosis method based on a weighted segmented random sink pool network, and belongs to the field of fault diagnosis. According to the method, a minimum mean square adaptive algorithm based on Wiener filtering is introduced, and a weighted segmented bistable random sink pool network model is constructed. In the first optimization strategy, a sparse measurement Gini index is selected as an evaluation index, the output of the model is solved based on a fourth-order Runge-Kutta algorithm, and an optimal parameter pair of a random sink pool network model unit subsystem is searched by using a particle swarm optimization algorithm. And in the second optimization strategy, inputting the optimal parameter pair and the original signal into a weighted segmented bistable random sink pool network model, adding independent and identically distributed Gaussian white noise into each unit of the network, and enhancing weak signal characteristics by using the noise to realize detection of an unknown bearing fault. According to the method, the noise utilization capability and the anti-noise robustness are high, and bearing fault characterizing features can be accurately found out and enhanced without bearing priori knowledge.

Description

Technical Field

[0001] The present invention belongs to the field of fault diagnosis, and relates to a bearing early fault diagnosis method based on a weighted segmented stochastic pooling network. Background Art

[0002] The robustness and reliability of fault diagnosis depend on the quality of measurement data. Vibration signals directly originate from key components such as bearings and contain rich fault feature information, which is helpful for diagnosis. Vibration analysis method is the most commonly used and effective method for fault feature enhancement and extraction. When early faults occur in mechanical components, it is difficult to extract fault features. The main reasons are as follows: First, when early weak faults occur, the useful components of the fault signals are very weak themselves. Second, due to the existence of equipment operation noise, environmental background noise, sensor and acquisition system noise, etc., the signal-to-noise ratio of the monitoring signals is very low. Therefore, how to extract early weak fault features from a strong noise background is the core of early weak fault diagnosis. Stochastic Resonance (SR) is a phenomenon originating from the field of nonlinear dynamics. Its origin can be traced back to the 1980s and was initially proposed by Italian scientists Benzi et al. in 1981 to explain this periodic phenomenon of the Earth's climate change. After that, stochastic resonance has been applied in the field of mechanical fault diagnosis, and the weak signal detection models based on SR theory have been continuously improved. However, most of these models require prior knowledge, such as the characteristic frequency information in the signal-to-noise ratio and amplitude gain, thus lacking research on faults with unknown frequency information.

[0003] In view of the above analysis, the present invention proposes a bearing early fault diagnosis method based on a weighted segmented stochastic pooling network, selects the Gini index to quantify the concentration and distribution of energy in the signal, and uses it as a fault feature index to optimize the parameters of each detection algorithm, so as to realize the application of SR theory in engineering. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that most of the unknown fault diagnosis methods based on stochastic resonance use adaptive algorithms to search for fault features to achieve fault diagnosis. The most important knowledge for detecting unknown faults is to find out the indicators that can characterize the bearing fault features and enhance them. Some time-domain features that do not contain frequency-domain features have received attention, such as root mean square, kurtosis, and impulse factor, etc. However, the dimensioned root mean square is affected by load and speed; the well-concerned kurtosis is also easily affected by noise and outliers. The purpose of the present invention is to provide a bearing early fault diagnosis method based on a weighted segmented stochastic pooling network, which can accurately find out and enhance the indicators that can characterize the bearing fault features without prior knowledge of the bearing, which is of great significance for timely discovering bearing early faults in engineering applications and reducing equipment production losses.

[0005] The technical solution adopted by the present invention to solve its technical problems is: a bearing early fault diagnosis method based on a weighted segmented stochastic reservoir network. First, a segmented bistable stochastic reservoir network model with the number of subsystems being num is constructed, and the application of the SR model in the field of weak signal detection is further improved. Then, the particle swarm optimization algorithm is used to optimize the subsystem parameters, and the algorithm converges stably. Finally, the optimized system parameters and independent and identically distributed white noise are input into the stochastic reservoir network model. The bearing fault experiment results show that the proposed method can identify and significantly enhance the bearing fault characteristics, and has engineering value for the early fault diagnosis of bearings. The specific steps are as follows:

[0006] Step 1: Use an acceleration sensor to collect vibration signals that can characterize the working state of mechanical equipment;

[0007] Step 2: Use frequency shift transformation to convert the collected vibration signals into small-parameter signals to meet the requirements of the stochastic resonance system;

[0008] Step 3: Based on the structure of array stochastic resonance, construct a weighted segmented bistable stochastic reservoir network model;

[0009] Step 4: Analyze the Gini index to characterize the bearing fault characteristics and quantify the stochastic resonance effect;

[0010] Step 5: Determine the optimization range of the particle swarm optimization system parameters and the algorithm initialization, and let

[0011] Step 6: Use the vibration signal processed in Step 2 as the input of the weighted segmented bistable stochastic reservoir network model unit; select the Gini index as the fitness function, iterate and update the optimal solution, and stably converge to obtain the best parameter pair p, u, k corresponding to the maximum Gini index value;

[0012] Step 7: Fix the parameters p i = p, u i = u, k i = k of other units of the weighted segmented bistable stochastic reservoir network model, and add independent and identically distributed Gaussian white noise to each model unit;

[0013] Step 8: Use the vibration signal processed in Step 2 as the input of the weighted segmented bistable stochastic reservoir network model;

[0014] Step 9: Select the output signals of the weighted segmented bistable stochastic reservoir network model with num = 40 for spectrum analysis, and obtain the fault frequency corresponding to the peak value of the vibration signal according to the spectrogram to realize the identification of unknown faults.

[0015] The beneficial effects of the present invention are as follows: A bearing early fault diagnosis method based on a weighted segmented stochastic pooling network has the following specific effects: The present invention can enhance and diagnose unknown weak fault features without prior knowledge. First, a weighted segmented stochastic pooling network model is constructed based on the array stochastic resonance structure. Then, in the first optimization strategy, an adaptive search method for bearing fault frequencies based on the sparse measure Gini index is given; in the second optimization strategy, based on the above fault frequency search method, the parameters of each unit subsystem are determined, and the weak fault features are enhanced by adding noise to the segmented stochastic pooling network. It is found through analysis that the signal-to-noise ratio of the model output tends to increase as the number of network unit subsystems increases. Finally, the present invention successfully identifies unknown bearing fault features, can further enhance the weak bearing fault features, is more helpful for the detection of unknown faults, and has important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the drawings and embodiments.

[0017] Figure 1 is the workflow diagram of the present invention;

[0018] Figure 2 is the comparison chart of the changing trends of the Gini index and the signal-to-noise ratio of the bearing inner ring fault simulation signal with the noise intensity;

[0019] Figure 3 is the curve graph of the Gini index and the signal-to-noise ratio varying with the noise under different system parameters p;

[0020] Figure 4 is the curve graph of the Gini index and the signal-to-noise ratio varying with the noise under different system parameters u;

[0021] Figure 5 is the time-domain waveform diagram and frequency-domain waveform diagram of the original vibration signal;

[0022] Figure 6 is the iterative process diagram of the algorithm;

[0023] Figure 7 is the time-domain diagram and frequency spectrum diagram of the system output under the first optimization strategy;

[0024] Figure 8 is the time-domain diagram and frequency spectrum diagram of the system output under the second optimization strategy. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The present invention will now be further described in detail in conjunction with the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0026] As Figures 1 to 8As shown, it is the optimal embodiment of the present invention, a bearing early fault diagnosis method based on a weighted segmented stochastic pooling network, and the specific steps are as follows:

[0027] Step 1: Use an acceleration sensor to collect vibration signals that can characterize the working state of mechanical equipment. The time-domain waveform diagram and frequency-domain waveform diagram of the original vibration signals are as Figure 5 shown;

[0028] Step 2: Use frequency shift transformation to convert the collected vibration signals into small-parameter signals to meet the requirements of the stochastic resonance system;

[0029] Step 3: Based on the structure of the array stochastic resonance, construct a weighted segmented bistable stochastic pooling network model, as Figure 1 shown; The array stochastic resonance system is composed of num subsystems combined in an array manner. In each subsystem U(y i ), the sine signal S(t) is respectively added with i (i = 1, 2,..., num) independent and identically distributed Gaussian white noises n i (t); The outputs y i (t) of each system are summed and averaged as the output x(t) of the entire array segmented system, which can be expressed by the Langevin equation as:

[0030]

[0031] Among them, the expression of the potential function of each subsystem is:

[0032]

[0033] Based on the structure of the system, introduce the least mean square algorithm based on Wiener filtering to construct a weighted segmented bistable stochastic pooling network model. Assume that there are num dipoles in total. The i-th (i = 1, 2,..., num) dipole can be regarded as different sensors or quantizers. The input signal is a discrete random signal. The output signals y i (t) of each subsystem are multiplied by the weighting coefficient w i , and then pooled and summed to obtain the output of the network. The structure of the segmented pooling network is as Figure 1 shown. The vector W = [w 1 , w 2 ,..., w num , (i = 1, 2,..., num) represents the weighting coefficients of each dipole, num represents the total number of dipoles, w 0 represents the bias coefficient, and Y = [y 1 , y 2 ,..., y num T represents the outputs of each dipole. The ideal output of the network can be expressed as:​

[0034] x = WY + w 0 (3)

[0035] Step 4: Analyze the Gini index (GI) to characterize the bearing fault features and quantify the stochastic resonance effect. For a signal x with N elements, GI can be calculated as follows:

[0036]

[0037] where g n = [2(N - n) + 1] / N 2 , Use a periodic impulse sequence to simulate the repetitive transient characteristics caused by the inner ring fault of a rolling bearing. As the added noise intensity increases, the trend of the GI curve is consistent with the SNR trend, as Figure 2 shown, indicating that GI can characterize the repetitive transient response caused by mechanical damage. Figure 3 and Figure 4 In, both GI and SNR increase first and then decrease with the increase of noise, showing peaks, which indicates that GI can quantify the stochastic resonance effect.

[0038] Step 5: Determine the optimization range of the particle swarm optimization system parameters and initialize the algorithm. The maximum number of genetic generations is set to 50, and the population size is 10. To ensure the rationality of the system potential function structure, let the system parameters p ∈ [0, 1], u ∈ [2, 5], k ∈ [1, 2].

[0039] Step 6: Use the vibration signal processed in Step 2 as the input of the weighted piecewise bistable stochastic reservoir computing network model unit. Select the Gini index as the fitness function. The algorithm iteration update process is as Figure 6 shown. After convergence and stability, the optimal parameter pair corresponding to the maximum Gini index value is obtained as p = 0.0229, u = 2.000, k = 1.0192;

[0040] Step 7: Fix the parameters of the other units of the weighted piecewise bistable stochastic reservoir computing network model as p = p i = 0.0229, u = u i = 2.000, k = k i = 1.0192, and add independent and identically distributed Gaussian white noise with a noise intensity of 3 to each model unit;

[0041] Step 8: Use the vibration signal processed in Step 2 as the input of the weighted piecewise bistable stochastic reservoir computing network model;

[0042] Step 9: Select the output signal of the weighted piecewise bistable stochastic reservoir computing network model with num = 40 as the final output signal, and obtain asFigure 8 as shown Figure 7 Figure 7 gives the output signal obtained by the primary optimization unit system. Compared with the output of the primary optimization strategy, after adding strong noise, the amplitude of the method of the present invention increases from 1.59 to 2.844 at the frequency spectrum of 128.9 Hz, the peak value increases by 1.2546, and the signal-to-noise ratio also increases by 1.3611. This shows the unique advantage of the present invention in noise utilization, which is more conducive to the detection of weak signals and also highlights the important value of the present invention in early bearing fault diagnosis under the background of strong noise.

[0043] As mentioned above, only the relatively optimal specific implementation manner of the present invention is described, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made by those skilled in the art within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A bearing early fault diagnosis method based on weighted piecewise random sink network, characterized in that: Here are the steps: Step 1: Use an acceleration sensor to collect vibration signals that can characterize the working state of the mechanical equipment; Step 2: Use frequency shift transformation to convert the collected vibration signal into a small parameter signal to meet the requirements of the stochastic resonance system; Step 3: Based on the structure of array stochastic resonance, a weighted piecewise bistable random sink network model is constructed; Step 4: Analyzing the Gini index can characterize the bearing fault characteristics and quantify the stochastic resonance effect; Step 5: Determine the optimal range of the particle swarm optimization system parameters and the algorithm initialization. Step 6: Use the vibration signal processed in step 2 as the input of the weighted piecewise bistable random sink network model unit; select the Gini index as the fitness function, iteratively update the optimal solution, and converge to obtain the optimal parameter pair p, u, k corresponding to the maximum Gini index value; Step 7: Fix the parameters p of other units in the weighted piecewise bistable random pooling network model i =p,u i =u,k i = k, and add independent and identically distributed Gaussian white noise to each model unit; Step 8: Use the vibration signal processed in step 2 as the input of the weighted piecewise bistable random sink network model; Step 9: Select the output signal of the weighted piecewise bistable random sink network model with num=40 for spectrum analysis, obtain the fault frequency corresponding to the peak value of the vibration signal according to the spectrum diagram, and realize the identification of unknown faults.

2. According to the method for early bearing fault diagnosis based on weighted piecewise random sink network described in claim 1, it is characterized in that: Step 3: In the weighted piecewise bistable random sink network model, there are num arrays in total. The i-th (i=1, 2, …, num) array can be regarded as a different sensor or quantizer. The output signal y of each subsystem is i (t) and weighting coefficient w i Multiply, then perform pooling and sum to get the network output x: x=WY+w0 Among them, the vector W=[w1,w2,...,w num ], (i=1,2,...,num) represents the weighted coefficient of each array, num represents the total number of arrays, w0 represents the bias coefficient, Y=[y1,y2,...,y num ] T Represents the output of each array, and the output of each array is expressed as: Where y(t) is the output of the array segmented bistable stochastic resonance; Among them, U(y i ) is the potential function of the piecewise bistable stochastic resonance.

3. According to the method for early bearing fault diagnosis based on weighted piecewise random sink network described in claim 1, it is characterized in that: Step 4 The value of the Gini index can be obtained by the following formula: in,

Images