A Rotating Machinery Fault Prediction Method Based on Scattering Transform

Abstract

The invention relates to a method for predicting faults of rotating machinery based on scattering transformation. The method mainly includes the following steps: 1. signal acquisition; 2. feature transformation and fault feature extraction, performing scattering transformation on the original signal to obtain the scattering transformation coefficient, and then Scattering transform coefficients of each sub-band to calculate its energy value as the characteristic value of signal distinction; 3. Fault diagnosis, using the least squares projection double support vector machine as a classifier for fault prediction. A method for predicting faults of rotating machinery based on scattering transformation provided by the present invention improves its translation invariance and elastic deformation stability by using non-subsampled dual-tree complex wavelet scattering transformation, and increases information redundancy , the features that are beneficial to extraction have better representation ability, and better solve the fault prediction problem of rotating machinery signals.

Description

technical field [0001] The invention relates to a mechanical fault prediction method, in particular to a rotating mechanical fault prediction method based on scattering transformation. Background technique [0002] Rotating mechanical systems have been widely used in aviation, ship, machine tool and vehicle engineering and are playing an increasingly important role. When the rotating machinery is damaged and fails, it will not only seriously affect the reliability and safety of the entire engineering system, but also bring huge economic losses. Therefore, relevant research and engineering implementation have been carried out at home and abroad. [0003] The fault prediction system of the rotating machinery system is mainly composed of three parts: one is the data collection of the transmission system. According to the characteristics of each component device, sensors are arranged at different positions to collect data on the operation status of the equipment in different sta...

AI Technical Summary

Problems solved by technology

The choice of complex wavelet in scattering transformation is very important. DT-CWT is commonly used, but it uses down-sampling technology, which has the characteristics of approximate time-shift invariance and low redundancy, which is not conducive to extracting the essential characteristics of mechanical rotation signals.

[0006] In the fault prediction machine learning algorithm, the commonly used algorithm Gaussian mixture model (GMM) has optimization parameters that are sensitive to the initial method, and it is difficult to determine the optimal number of components; ANN has no standard method to determine the learning structure of the network, and there is an over-learning problem; The rough set method requires discretization, is not suitable for continuous numerical changes, and the threshold for decision-making is not reliable

Images