Mechanical system fault monitoring method and system based on signal pulse characteristic enhancement

Abstract

The invention provides a mechanical system fault monitoring method and system based on signal pulse characteristic enhancement, and the method comprises the steps: 1, collecting vibration signals of mechanical equipment at different positions, and selecting a proper input signal; step 2, removing stationary signals in the signals based on an AR model so as to extract noise and impact components in the signals, and performing noise reduction processing on the signals; and step 3, based on an ISSA-MOMEDA method, obtaining an optimal filter length L and an optimal period T, and inputting the optimal filter length L and the optimal period T into MOMEDA, so that the MOMEDA performance is optimal, and signal pulse characteristic enhancement is realized. According to the invention, noise reduction processing can be carried out on the vibration signal of the mechanical equipment, and meanwhile, fault features can be enhanced, so that the fault features of the mechanical equipment can be extracted more quickly and accurately, and the health condition of the mechanical equipment can be judged.

Description

Technical Field

[0001] The present invention relates to the field of mechanical equipment fault diagnosis, and in particular to a mechanical system fault monitoring method and system based on signal pulse feature enhancement. Background Art

[0002] With the rapid development of my country's manufacturing industry, various mechanical equipment are constantly being innovated. my country's aerospace vehicles, large generator sets, gas turbine units, ships, deep-diving equipment and other major mechanical equipment are showing a trend of large-scale, complex and integrated components, which is bound to put forward more stringent requirements on their operating indicators. Gears, as a commonly used power transmission component, are widely used in complex electromechanical equipment such as helicopter transmission systems and wind turbines because of their advantages such as large overlap, strong load capacity and high transmission efficiency. Due to long-term operation under harsh conditions of high temperature, high speed and heavy load, gear failures often occur. Once a failure occurs, it will cause the machine to stop and the equipment to be damaged, bringing immeasurable economic losses to the enterprise. Therefore, it is necessary to monitor the faults of mechanical equipment. The existing multi-point optimal minimum entropy deconvolution adjustment method (Multipoint Optimal Minimum Entropy Deconvolution Adjusted, MOMEDA) realizes the extraction of weak fault features in the extraction of weak features of gears. However, MOMEDA needs to directly set the fault cycle and filter length. The settings of these two parameters often rely on manual experience selection, and these two parameters, as important inputs to the MOMEDA algorithm, directly affect the effect of feature extraction. At present, the sparrow optimization algorithm is mainly used to determine the parameters, but in the background of strong noise, MOMEDA is easy to search for false components, which can easily cause misdiagnosis. Therefore, a new monitoring method is needed that can eliminate the influence of noise and enhance the fault characteristics, so as to judge the health status of mechanical equipment earlier and more accurately to avoid greater losses. Summary of the invention

[0003] Purpose of the invention: The technical problem to be solved by the present invention is to provide a mechanical system fault monitoring method based on signal pulse feature enhancement in view of the deficiencies of the prior art, comprising the following steps:

[0004] Step 1, obtaining vibration signals of the mechanical equipment when it is running at multiple positions by using more than two acceleration sensors installed at different positions of the mechanical equipment, and analyzing the collected vibration signals to select the optimal vibration signal;

[0005] Step 2: extract the vibration signal obtained in step 1, use the autoregressive model AR (Autoregressive model, AR) method to pre-whiten the vibration signal, remove the inherent component in the vibration signal, and select the residual signal as the signal to be analyzed; set the order range of the autoregressive model AR iteration, such as 0 to 100 orders, and use the Yule-Walker method to calculate the autoregressive weighted coefficient a corresponding to each order within the given order range. i , i=1,2,…,p,a i Represents the autoregressive weighting coefficient corresponding to the i-th order, p is the order of the autoregressive model AR, based on the autoregressive weighting coefficient, the vibration signal is processed to obtain the residual signal; the residual signal kurtosis maximization criterion is used as the output criterion of the autoregressive model AR optimal filter, and the order of the autoregressive model AR filter when the residual signal kurtosis is maximum is output after the iteration is completed to obtain the optimal autoregressive model AR filter. The optimal autoregressive model AR filter is used to process the signal to obtain the residual signal ε after the optimal order processing k ;

[0006] Step 3, initialize the parameters of the Improved Sparrow Search Algorithm ISSA (Improved Sparrow Search Algorithm, ISSA), and calculate the envelope spectrum peak factor of the signal after each multi-point optimal minimum deconvolution MOMEDA (Multipoint Optimal Minimum Entropy Deconvolution Adjusted, MOMEDA) filtering. When the termination condition is met, the optimal filter length L and period T are obtained, and the optimal filter length L and period T are input into MOMEDA to achieve optimal MOMEDA performance.

[0007] Step 2 includes:

[0008] Step 2-1: For a discrete time series x(k), the kth time series point to be predicted in the autoregressive model AR is x k , predict x from the first k-1 time series points of the signal k The performance of the autoregressive model AR is the most critical factor affecting the AR model prediction filter. When the order of the autoregressive model AR is p, x k The predicted value of It is expressed as:

[0009]

[0010] Among them, e k is the residual signal;

[0011] Step 2-2, determine the order range of the autoregressive model AR, set the value range of p to 0-100; each time an order is set, the Yule-walker method is used to calculate the autoregressive weight coefficient a under the corresponding order i ;

[0012] Step 2-3: add the autoregressive weight coefficient a i Substitute it into the autoregressive model AR, use the autoregressive model AR to process the input vibration signal, and obtain the residual signal;

[0013] Step 2-4, according to the characteristics of the mechanical equipment fault signal, the maximum kurtosis criterion is selected to select the optimal order of the autoregressive model AR, and the kurtosis values ​​of the residual signal e(k) under different orders are calculated respectively. The order with the largest kurtosis value is the optimal order of the autoregressive model AR, and finally the residual signal obtained after the optimal order processing is obtained;

[0014] Step 3 includes:

[0015] Step 3-1, assuming that y is an impact signal of the faulty gear, h is the system frequency response function, x is the vibration signal collected by the sensor, and e is random noise. The vibration signal collected by the sensor can be approximately expressed as:

[0016] x=h*y+e

[0017] MOMEDA uses a non-iterative method to find a global filter f to reconstruct y and minimize the impact of background noise and other interference components on the signal. The process of MOMEDA deconvolution is:

[0018]

[0019] Where N is the length of the vibration signal x; L is the length of the filter f, f k represents the kth filter coefficient of filter f, x k+L-1 Indicates the k+L-1th sampling point of the vibration signal x collected by the sensor, k=1,2,…,NL;

[0020] Step 3-2, in order to continuously extract periodic pulses in vibration signals, MOMEDA proposes multi-point D-norm (MDN) as the objective function based on D-norm. The optimal deconvolution effect is obtained by solving the maximum value of the objective function. The solution of multi-point D-norm is expressed as:

[0021]

[0022] in, is the target vector, which consists of constants and determines the position of the impulse signal in the output signal; T represents the matrix transpose; It is a dimensionless indicator used to measure the similarity between the signal y and the target vector t; MOMEDA: max Represents the optimized filter f, maximizing That is, the matching degree between the filtered signal y and the target vector t is maximized; The accuracy of the impact sequence positioning directly affects the deconvolution effect; when When the multi-point D-norm reaches its maximum value when it matches the impulse sequence perfectly, the corresponding filter f is f = [f1, f2, …, f L ] T The optimal filter, where f1,f2,…,f L Represents the weight parameters of L filters, which are used to measure the weight of each coefficient of the filter in the filtering operation;

[0023] Step 3-3, use the ISSA algorithm to search for the optimal parameters of MOMEDA. The fitness function needs to be determined. The sparrow population is evaluated based on the size of the fitness value to determine whether it is the optimal solution. The inverse of the crest factor of envelope spectrum (Ec) is used as the fitness function. Ec is a dimensionless index that can well represent the periodicity and intensity of the signal impulse component. The amplitude sequence of the envelope spectrum of the signal is set to X(z), where z represents the index value of the frequency component in the envelope spectrum, ranging from z=1,2,…,Z. The dimensionless index Ec is expressed as:

[0024]

[0025] The ISSA parameters are set, such as population size N, maximum number of iterations, discoverer ratio PD, alerter ratio SD, alert threshold R, etc., and then the ISSA-MOMEDA method with the above parameters is used to analyze the collected signals.

[0026] The present invention also provides a mechanical system fault monitoring system based on signal pulse feature enhancement, including a data acquisition module and a signal processing module;

[0027] The data acquisition module obtains vibration signals of the mechanical equipment at different positions when the mechanical equipment is running through more than two acceleration sensors installed at different positions of the mechanical equipment, analyzes the collected vibration signals, selects n vibration positions, and uses n groups of vibration signals at each position as input signals;

[0028] The signal processing module is used to pre-process the n groups of vibration signals collected to obtain denoised vibration signals and enhance their features;

[0029] The signal processing module includes an AR denoising module and an ISSA-MOMEDA signal feature enhancement module;

[0030] The denoising module uses the AR method to perform pre-whitening to remove the inherent components in the vibration signal to achieve the denoising effect;

[0031] The signal feature enhancement module is based on the ISSA-MOMEDA algorithm, initializes the ISSA parameters, and calculates the envelope spectrum peak factor of the signal after each MOMEDA filtering. When the termination condition is met, the optimal filter length L and period T are obtained;

[0032] The present invention also provides a computer device, comprising a memory, a processor and an executable computer program stored in the memory, wherein the processor implements the method when executing the computer program.

[0033] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0034] The present invention has the following beneficial effects: firstly, the present invention screens vibration signals at different positions of mechanical equipment based on the effective value of the signal, selects appropriate input signals, and improves the accuracy of fault identification; secondly, the present invention effectively removes noise in the signal by constructing an autoregressive model (AR) for noise reduction; finally, the present invention proposes a method based on ISSA-MOMEDA, which can obtain a clearer spectrum diagram and improve the accuracy of fault feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0036] Figure 1 The figure is an overall flow chart of the method of the present invention.

[0037] Figure 2 This is a time domain diagram of the vibration acquisition signal.

[0038] Figure 3 This is a frequency domain diagram of the vibration acquisition signal.

[0039] Figure 4 This is the schematic diagram of the AR model.

[0040] Figure 5 It is the time domain diagram of the signal after AR processing.

[0041] Figure 6 It is the frequency domain diagram of the signal after AR processing.

[0042] Figure 7 This is a graph showing the relationship between the AR model order and the kurtosis of the signal after AR processing.

[0043] Figure 8 This is the time domain diagram of the signal after ISSA-MOMEDA processing.

[0044] Fig. 9 This is the frequency domain diagram of the signal after ISSA-MOMEDA processing. DETAILED DESCRIPTION

[0045] The embodiment of the present invention discloses a mechanical system fault monitoring method based on signal pulse feature enhancement, the specific process is as follows: Figure 1 As shown, the following steps are included:

[0046] Step 1: Use multiple acceleration sensors placed at different positions of the mechanical equipment to obtain vibration signals when the mechanical equipment is running at multiple positions. Analyze the collected vibration signals and select the vibration position. Select the vibration signal at the optimal position as the input signal of the vibration signal feature extraction algorithm. The time domain and frequency domain diagrams of the vibration signal are shown in Figure 1. Figure 2 and Figure 3 As shown;

[0047] Step 2: Extract the vibration signal of the mechanical equipment position obtained in step 1, use the AR method to pre-whiten the collected vibration signal, remove the inherent component in the vibration signal, and select the residual signal as the signal to be analyzed. Set the order range of AR iteration, such as 0 to 100 orders, and use the Yule-Walker method to calculate the autoregressive weighting coefficient a corresponding to each order within the given order range. i (i=1,2,…,p), based on the calculated regression weight coefficient, process the input vibration signal to obtain the residual signal. The residual signal kurtosis maximization criterion is used as the AR optimal filter output criterion. After the iteration is completed, the order of the AR filter when the residual signal kurtosis is maximum is output to obtain the optimal AR filter. The optimal AR filter is used to process the signal to obtain the residual signal ε after the optimal order processing. k ;

[0048] Step 201: The residual signal in the AR model can be regarded as a combination of the shock component and the random noise. According to this characteristic of the AR model, an AR prediction model filter is constructed as follows: Figure 4 As shown in the figure, x k represents the original signal input to the system, Represents the input signal x through the AR filter k The estimated value of k is the residual signal, representing the actual signal x kand prediction signals For a discrete time series x(k), the kth time series point xk to be predicted in the AR model can be predicted by the performance of xk based on the first k-1 time series points of the signal. When the order of AR is set to p, the predicted value of xk is It can be expressed as:

[0049]

[0050] Among them, a i is the autoregressive weighted coefficient of the p-order AR model, e k is the residual signal.

[0051] Step 202: Determine the order range of the AR model, where p is in the range of 0 to 100; then use the Yule-walker method to calculate the autoregressive weighting coefficient a at the corresponding order. i (i=1,2,…,p). The Yule-walker method is based on the Yule-walker linear equations, which connect the autocorrelation sequence of the signal and the equations of the AR model parameters. The p+1 order model parameters can be calculated by calculating the Yule-Walker equation with the p+1 order autocorrelation coefficient. The autocorrelation function is

[0052]

[0053] Where R x (m) is the autocorrelation function of the signal, m is the lag time, a k is the autoregressive coefficient, σ 2 is the variance of white noise. Using the symmetry of the autocorrelation function, it can be written in matrix form as

[0054]

[0055] When the order p and the autocorrelation function are known, the parameter estimation problem of the AR model is the solution of the Yule-Walker equation.

[0056] Step 203: Based on the calculated regression weight coefficient, the input vibration signal is processed to obtain its residual signal.

[0057] Step 204: According to the characteristics of the mechanical equipment fault signal, the maximum kurtosis criterion is selected to select the optimal order of the AR model, and the kurtosis values ​​of the residual signal e(k) under different orders are calculated respectively. The order corresponding to the largest kurtosis value is the optimal order of the AR model, and finally the residual signal obtained after the optimal order processing is obtained. The time domain diagram and frequency domain diagram of the signal after AR processing are respectively as follows: Figure 5 and Figure 6 shown.

[0058] The relationship between the AR model order and the kurtosis of the residual signal after AR processing is as follows: Figure 7 As shown;

[0059] Step 3: Initialize ISSA parameters and calculate the envelope spectrum peak factor of the signal after each MOMEDA filtering. When the termination condition is met, the optimal filter length L and period T are obtained and input into MOMEDA to achieve MOMEDA performance optimization.

[0060] Step 301: Assume that y is an impact signal of a faulty gear, h is the system frequency response function, x is the vibration signal collected by the sensor, and e is random noise. Then the signal collected by the sensor can be approximated as:

[0061] x=h*y+e

[0062] The core idea of ​​MOMEDA is to find a global filter f in a non-iterative way to reconstruct y and minimize the impact of background noise and other interference components on the signal. The deconvolution process is:

[0063]

[0064] Where N represents the length of the signal x; L represents the length of the filter f.

[0065] Step 302: In order to continuously extract periodic pulses in the vibration signal, MOMEDA proposes a multi-point D-norm as the objective function based on the D-norm, and optimizes the deconvolution effect by finding the maximum value of the multi-point D-norm. That is:

[0066]

[0067] in, is the vector of input signals, is a vector The norm of . is the target vector, which determines the location and weight of the deconvolved target pulse. The accuracy of the impact sequence positioning directly affects the deconvolution effect. When the multi-point D-norm reaches its maximum value when it matches the impulse sequence perfectly, the corresponding filter f is f = [f1, f2, ..., f L ] T Optimal filter. The solution of MOMEDA can be equivalent to solving the following equation, and the optimal filter is iteratively solved through gradient update:

[0068]

[0069] Among them, k = 1, 2,…, NL.

[0070] Let X0 = [M1, M2, ..., M k ], the above formula is simplified to:

[0071]

[0072] Will Substituting it into the equation, we have:

[0073]

[0074] Assume (X0X0 T ) -1 Exists, and the optimal filter is solved as:

[0075]

[0076] Step 303: The MOMEDA algorithm input includes the length L of the filter and the length of the fault period T. The MOMEDA algorithm input includes the length L of the filter and the length T of the fault period. The length L of the filter directly determines the effect of pulse feature extraction. In order to ensure that the extracted impulse signal can cover the entire frequency band, the length of the filter should meet the following requirements:

[0077]

[0078] Among them, f s is the sampling frequency of the signal, f c is the fault characteristic frequency.

[0079] For the fault cycle T, since the speed of the input motor does not remain stable during actual operation, the speed of the faulty gear shaft is always fluctuating. The fluctuation range of the faulty gear shaft is 5%, so T can be obtained. s =1 / (1-0.05)*f i , T f =1 / (1+0.05)*f i , the value of the fault cycle is [T s ,T f ] is selected. When using the ISSA algorithm to search for the optimal parameters of MOMEDA, it is necessary to determine the fitness function. The sparrow population is evaluated based on the size of the fitness value to determine whether it is the optimal solution. The inverse of the crest factor of envelope spectrum (Ec) is used as the fitness function. Ec is a dimensionless indicator that can well represent the periodicity and intensity of the signal's impact component. Assuming that the amplitude sequence of the signal's envelope spectrum is X(z) (z=1,2,…,Z), Ec can be expressed as:

[0080]

[0081] Step 304: Set the number of ISSA iterations, the dimension of the parameters to be optimized (fault cycle and filter length), the number of ISSA populations, set the value range of the fault cycle and the value range of the filter length, and use the ISSA-MOMEDA method to analyze the collected signal. The results of the analysis are as follows: Figure 8 and Fig. 9 shown. Figure 8 The figure shows the change of the vibration signal in the time domain. After filtering, the signal originally mixed with random noise becomes smoother and clearer, and the noise is effectively suppressed. In particular, the local magnified part of the figure clearly shows the periodic characteristics of the pulse signal, and the impact component of each period T is significantly enhanced, which shows that after being processed by the method of the present invention, the periodic impact signal is well extracted and highlighted. Fig. 9 The frequency domain performance of the processed signal is further demonstrated. The characteristic frequency f can be clearly seen in the spectrum. r The vibration signal feature enhancement method can effectively remove the noise in the vibration signal, enhance the pulse component in the signal, and clearly see the characteristic frequency and its multiples.

[0082] The present invention provides a mechanical system fault monitoring method and system based on signal pulse feature enhancement. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A mechanical system fault monitoring method based on signal pulse feature enhancement, characterized in that: The following steps are involved: Step 1, obtaining vibration signals of the mechanical equipment when it is running at multiple positions by using more than two acceleration sensors installed at different positions of the mechanical equipment, and analyzing the collected vibration signals to select the optimal vibration signal; Step 2, extracting the vibration signal obtained in step 1, pre-whitening the vibration signal using an autoregressive model AR method, removing the inherent component in the vibration signal, and selecting the residual signal as the signal to be analyzed; Set the order range of the autoregressive model AR iteration. Within the given order range, use the Yule-Walker method to calculate the autoregressive weighted coefficient a corresponding to each order. i , i=1,2,…,p,a i Represents the autoregressive weighting coefficient corresponding to the i-th order, p is the order of the autoregressive model AR, based on the autoregressive weighting coefficient, the vibration signal is processed to obtain the residual signal; the residual signal kurtosis maximization criterion is used as the output criterion of the autoregressive model AR optimal filter, and after the iteration is completed, the order of the autoregressive model AR filter when the residual signal kurtosis is maximum is output to obtain the optimal autoregressive model AR filter; the optimal autoregressive model AR filter is used to process the signal to obtain the residual signal ε after the optimal order processing k ; Step 3, initialize the parameters of the improved sparrow search algorithm ISSA, and calculate the envelope spectrum peak factor of the signal after each multi-point optimal minimum deconvolution MOMEDA filtering. When the termination condition is met, the optimal filter length L and period T are obtained, and the optimal filter length L and period T are input into MOMEDA to achieve the optimal MOMEDA performance.

2. The method according to claim 1, characterized in that Step 2 includes: Step 2-1: For a discrete time series x(k), the kth time series point to be predicted in the autoregressive model AR is x k , predict x from the first k-1 time series points of the signal k When the order of the autoregressive model AR is p, the predicted value of xk is It is expressed as: Among them, e k is the residual signal; Step 2-2, determine the order range of the autoregressive model AR, and then use the Yule-walker method to calculate the autoregressive weighting coefficient a under the corresponding order for each order. i ; Step 2-3: add the autoregressive weight coefficient a i Substitute it into the autoregressive model AR, use the autoregressive model AR to process the input vibration signal, and obtain the residual signal; Step 2-4, according to the characteristics of the mechanical equipment fault signal, the maximum kurtosis criterion is selected to select the optimal order of the autoregressive model AR, and the kurtosis values ​​of the residual signal e(k) at different orders are calculated respectively. The corresponding order with the largest kurtosis value is the optimal order of the autoregressive model AR, and finally the residual signal obtained after the optimal order processing is obtained.

3. The method according to claim 2, characterized in that Step 3 includes: Step 3-1, assuming that y is an impact signal of the faulty gear, h is the system frequency response function, x is the vibration signal collected by the sensor, and e is random noise. The vibration signal collected by the sensor is expressed as: x=h*y+e MOMEDA uses a non-iterative method to find a global filter f to reconstruct y and minimize the impact of background noise and other interference components on the signal. The process of MOMEDA deconvolution is: Where N is the length of the vibration signal x; L is the length of the filter f, f k represents the kth filter coefficient of filter f, x k+L-1 Indicates the k+L-1th sampling point of the vibration signal x collected by the sensor, k=1,2,…,NL; Step 3-2, in order to continuously extract periodic pulses in vibration signals, MOMEDA proposes multi-point D-norm as the objective function based on D-norm, and obtains the optimal deconvolution effect by solving the maximum value of the objective function. The solution of multi-point D-norm is expressed as: in, is the target vector; T represents matrix transpose; It is a dimensionless indicator used to measure the similarity between the signal y and the target vector t; Represents the optimized filter f, maximizing That is, the matching degree between the filtered signal y and the target vector t is maximized; When the multi-point D-norm reaches its maximum value when it matches the impulse sequence perfectly, the corresponding filter f is f = [f1, f2, …, f L ] T The optimal filter, where f1,f2,…,f L represents the weight parameters of L filters; Step 3-3, use the ISSA algorithm to search for the optimal parameters of MOMEDA. The sparrow population is evaluated based on the size of the fitness value to determine whether it is the optimal solution. The inverse of the envelope spectrum peak factor is used as the fitness function. Ec is a dimensionless index. The amplitude sequence of the envelope spectrum of the signal is set to X(z). z represents the index value of the frequency component in the envelope spectrum, ranging from z = 1, 2, ..., Z. The dimensionless index Ec is expressed as: The ISSA parameters are set, and then the ISSA-MOMEDA method is used to analyze the collected signals.

4. A mechanical system fault monitoring system based on signal pulse feature enhancement implemented by the method according to claim 1, characterized in that: Including data acquisition module and signal processing module; The data acquisition module obtains vibration signals of the mechanical equipment at different positions when the mechanical equipment is running through more than two acceleration sensors installed at different positions of the mechanical equipment, analyzes the collected vibration signals, selects n vibration positions, and uses n groups of vibration signals at each position as input signals; The signal processing module is used to preprocess the n groups of collected vibration signals to obtain denoised vibration signals and enhance their features.

5. The system according to claim 4, characterized in that The signal processing module includes an AR denoising module and an ISSA-MOMEDA signal feature enhancement module; The denoising module uses the AR method to perform pre-whitening to remove the inherent components in the vibration signal to achieve the denoising effect; The signal feature enhancement module is based on the ISSA-MOMEDA algorithm, initializes ISSA parameters, and calculates the envelope spectrum peak factor of the signal after each MOMEDA filtering; when the termination condition is met, the optimal filter length L and period T are obtained.

6. A computer device comprising a memory, a processor and an executable computer program stored in the memory, wherein the processor implements the method according to any one of claims 1 to 3 when executing the computer program.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images