CN-115655706-B - Mechanical fault diagnosis method based on minimum entropy deconvolution and stochastic resonance

CN115655706BCN 115655706 BCN115655706 BCN 115655706BCN-115655706-B

Abstract

The invention relates to a mechanical fault diagnosis system based on minimum entropy deconvolution and stochastic resonance, which comprises the steps of collecting vibration data of a mechanical rotating component as an original signal, inputting the original signal into a bistable stochastic resonance system after frequency scaling, adaptively adjusting system parameters by taking relevant kurtosis as an index function, enabling the system to output a desired signal, obtaining a numerical solution output by a nonlinear system by utilizing a 4-order Dragon-Gerdon algorithm, amplifying impact components in the original signal by utilizing a random resonance phenomenon to obtain a noise-reduced signal, carrying out minimum entropy deconvolution filtering on the noise-reduced signal to obtain a minimum entropy deconvolution filtering signal, carrying out Hilbert envelope spectrum analysis on the signal subjected to the minimum entropy deconvolution filtering, and carrying out fault diagnosis on the mechanical rotating component. The invention can autonomously extract the frequency of the fault signal by designing the corresponding index under the condition of unknown signal fault frequency, thereby realizing accurate fault diagnosis and positioning.

Inventors

WEN DEGANG
SONG XIAOYING
ZHANG QI
Yang Tianlian
YANG QINMIN
Weng Deyu
CHEN XU
CAO WEIWEI

Assignees

山东临工工程机械有限公司
浙江大学

Dates

Publication Date: 20260512
Application Date: 20221014

Claims (4)

1. A method for mechanical fault diagnosis based on minimum entropy deconvolution and stochastic resonance, the method comprising the steps of: S1, acquiring vibration data of a mechanical rotating component by using an acceleration sensor, wherein the vibration data are used as original signals; S2, inputting an original signal into a bistable stochastic resonance system after frequency scaling, adaptively adjusting system parameters by taking relevant kurtosis as an index function to enable the system to output a desired signal, obtaining a numerical solution output by a nonlinear system by utilizing a 4-order Dragon-Gregory tower algorithm, amplifying impact components in the original signal by utilizing a random resonance phenomenon to obtain a noise-reduced signal, wherein the specific implementation steps of inputting the original signal into the stochastic resonance system after frequency scaling are as follows: S21, selecting a frequency scaling multiple to enable an original input signal to meet the characteristic of low frequency; S22, initializing nonlinear system parameters, and giving a parameter selection interval; S23, selecting a relevant kurtosis as an objective function, traversing a parameter initialized interval by using a grid search method, and selecting a parameter combination with the largest relevant kurtosis; S24, setting a stochastic resonance system by using an optimal parameter combination obtained by a grid search method, inputting an original signal into the stochastic resonance system to amplify a low-frequency impact signal, inhibiting a high-frequency part, and restoring the frequency of the original signal by the amplified denoising signal according to the frequency scaling multiple selected in the S21 to obtain a stochastic resonance denoising signal; S3, carrying out minimum entropy deconvolution filtering on the signal subjected to noise reduction by the bistable stochastic resonance system, selecting kurtosis as an index function, and outputting a signal with the maximum kurtosis to obtain a minimum entropy deconvolution filtering signal; S4, performing Hilbert envelope spectrum analysis on the signal subjected to the minimum entropy deconvolution filtering, comparing an envelope spectrum analysis result with the calculated bearing theoretical fault characteristic frequency value and the frequency component with the highest amplitude in the envelope spectrum, and performing fault diagnosis on the mechanical rotating component.
2. The method for diagnosing a mechanical failure based on minimum entropy deconvolution and stochastic resonance according to claim 1, wherein in S2, the method for generating stochastic resonance by the signal through the nonlinear system comprises: the langerhans' path of the Brownian particles in a nonlinear system under weak periodic forces and noise is as follows: (1) Wherein x (t) is the motion track of particles along with time, gamma is the damping coefficient among the particles, U (x) is a system potential function, S (t) is weak periodic external force which changes along with time, and xi (t) is zero-mean Gaussian white noise, and the intensity is D; The potential function U (x) of the bistable system is as follows: (2) Wherein a >0, b >0; the equation for the bistable over-damped nonlinear system is thus the following: (3) Assuming that S (t) is a periodic signal of period Ω, the escape rate of particle runout caused by noise is: (4) Wherein DeltaU is barrier height, is the height difference between the bistable system stable point and the critical stable point, and is influenced by the system parameters a and b, so that it is known that the particle escape rate is influenced by the system parameters a and b and the noise intensity D, when the particle escape rate meets the following conditions, the system generates stochastic resonance, and the noise intensity is used for amplifying the low-frequency signal: (5)。
3. The method for diagnosing mechanical faults based on minimum entropy deconvolution and stochastic resonance according to claim 1, wherein in S3, the method for denoising signals through minimum entropy deconvolution filtering comprises the following steps: And selecting a series of FIR filters, convoluting the signals after stochastic resonance denoising, and only selecting the part of the signals which is not zero to convolve in the convolution process, so as to avoid discontinuity in the convolution process, selecting initial filter coefficients and kurtosis as objective functions, and solving the filters by an iterative algorithm until the objective functions are not reduced any more, thereby obtaining optimal filter coefficients and filtering output signals.
4. The mechanical fault diagnosis method based on minimum entropy deconvolution and stochastic resonance according to claim 1, wherein in S4, the method of hilbert envelope spectrum analysis of the signal is as follows: The signal is subjected to Hilbert transformation to obtain a complex domain part of the original signal, the original signal and the complex domain part are combined to obtain an analysis signal of the signal, the analysis signal is modulo solved to obtain a Hilbert envelope signal, and the amplitude spectrum is calculated to obtain the Hilbert envelope spectrum.

Description

Mechanical fault diagnosis method based on minimum entropy deconvolution and stochastic resonance Technical Field The invention belongs to the field of intelligent fault diagnosis of bearings, and particularly provides a mechanical fault diagnosis method based on minimum entropy deconvolution and stochastic resonance. Background Mechanical systems have an indispensable role in industrialization, wherein rotating mechanical devices are more of a majority, and a key in rotating mechanical devices is a gear box, which is difficult to maintain due to a severe industrial environment and a closed working environment, so that various faults often occur in the gear box in the rotating mechanical devices, and each fault may cause indispensable money and productivity loss. In the so-called fourth industrial revolution, future factories, the industrial internet of things era, industrial mechanical systems are constantly being intelligent and complicated. Therefore, it is necessary to study and develop data-driven methods and state monitoring techniques that enable rapid, reliable and high quality automated diagnostics. The early failure of the gearbox can be accurately early warned, serious industrial accidents are avoided, and workers can maintain the gearbox in time, so that the early failure warning device has great significance for industrial production. Stochastic resonance (Stochastic resonance, SR) is widely studied in the field of mechanical fault diagnosis as a nonlinear signal processing method capable of extracting weak signal features from vibration signals, with its unique advantage of enhancing weak signals by noise rather than eliminating noise. The minimum entropy deconvolution algorithm (MED) was originally proposed by r.a. wiggins in 1978 and applied to seismic data detection as an iterative algorithm to maximize the kurtosis value of the filtered output signal, which is an indicator that can measure the signal impact component. In 1984, C.A. Cabrelli proposed a kurtosis equivalence index, called D-norm, which solved the problem that the minimum entropy deconvolution algorithm can only be solved iteratively, called Optimal Minimum Entropy Deconvolution (OMED). In order to solve the defect that the kurtosis index can only extract a single pulse, a new optimal minimum entropy deconvolution algorithm is proposed as an iterative algorithm, and the Maximum Correlation Kurtosis Deconvolution (MCKD) can simultaneously extract a plurality of pulses, but the method needs to be used as a priori until the fault characteristic frequency of a signal, and parameters such as the length of a filter, the fault period and the like need to be selected, so that the application of the method is limited. In 2016, geoff l.mcdonald improved the D-norm index, proposed a multi-point optimal minimum entropy deconvolution (MOMED), and proposed an adjusted convolution algorithm that overcomes the discontinuities of the original convolution calculation. The invention combines stochastic resonance and minimum entropy deconvolution algorithm, utilizes the capability of enhancing low-frequency signals by stochastic resonance to enhance low-frequency impact signals, and leads the signals after pulse enhancement to pass through minimum entropy deconvolution and envelope spectrum extraction characteristics, thereby improving the capability of extracting pulses by minimum entropy deconvolution and enhancing fault extraction capability. Disclosure of Invention Based on the background of the problems, the invention provides a mechanical fault diagnosis method based on minimum entropy deconvolution and stochastic resonance, which specifically uses the characteristic that low-frequency components of signals can be amplified by stochastic resonance phenomenon, amplifies low-frequency impact signals of original signals through a stochastic resonance system, suppresses high-frequency parts, and extracts features of the signals after impact enhancement through minimum entropy deconvolution to realize accurate fault diagnosis. The specific technical scheme of the invention is as follows: S1, acquiring gearbox vibration data of an engineering machinery vehicle by using an acceleration sensor as an original signal; S2, performing frequency scaling on an original signal, and then performing a bistable (Bistable) nonlinear system, namely a bistable stochastic resonance system, taking relevant kurtosis (Correlation Kurtosis, CK) as an index function, adaptively adjusting system parameters to enable the system to output a desired signal, obtaining a numerical solution output by the nonlinear system by using a 4-order Longkuh-Tara (range-Kutta) algorithm, and amplifying impact components in the original signal by using a stochastic resonance (Stochastic Resonance, SR) phenomenon to obtain a noise-reduced signal; S3, carrying out minimum entropy deconvolution (Minimum Entropy Deconvolution Ajusted, MEDA) filtering on the signal subjected to noise reduct