Research On Fault Data Classifcation Based On Hybrid Invasive Weed Optimization-Neural Network

In recent years, with the modern machinery equipment becoming larger and morecomplicated, automatic and continuous, fault diagnosis of machinery and equipment to carryout technology research has important practical significance. At present, domestic and foreignscholars have done a lot of work in this area, making the relevant theory and application toachieve rapid development. Mechanical fault diagnosis by studying the relationship betweenfaults and symptoms to determine equipment failure, and between failure and showed signs ofa very complex non-linear relationship, it is difficult to be precise description of themathematical model, to mechanical failure diagnosis of a great deal of inconvenience.Artificial neural network is an important artificial intelligence act, is a non-linear computingsystem can be achieved between the fault and symptom complex non-linear mapping, so inthe field of machinery fault diagnosis has been a great application potential.This article builds hybrid invasive weed optimization algorithm to optimize BP neuralnetwork prediction model, rotor test rig to simulate a large number of failure data to support,information entropy approach to quantitative data on the fault feature extraction, hybridinvasive weed optimization algorithm to optimize neural network structure. Main tasks andtheoretical research as follows:(1) Four types of typical fault were simulated on the rotor experiment table, whosemechanism was analyzed. We can get the singular spectrum entropy value, power spectrumentropy value, wavelet energy spectrum entropy value and wavelet space feature entropyvalue after computed the entropy value band of fault signal in the time domain, frequencydomain and time-frequency domain.(2) Four spectral entropy as the original data, the data is normalized, and theestablishment of training samples and test samples of library databases.(3) Authors take advantage of the genetic algorithms and particle swarm optimization,presents a new hybrid invasive weed optimization. A hybrid invasive weed optimizationalgorithm (HIWO) is designed. The hybrid algorithm introduces genetic algorithm crossoveroperator and particle swarm algorithm of the vector tracking thought.(4) Establish HIWO optimize BP neural network model, by HIWO algorithm got theinitial training of BP network optimal weights and threshold value, and then in the BPnetwork to test sample in the forecast. The hybrid invasive weed optimization algorithmoptimization neural network, with genetic algorithm, particle swarm algorithm and theoriginal algorithm to optimize the neural network were compared.(5) Process based on HIWO algorithm developed a MATLAB GUI rotor fault diagnosissystem. One subsystem can analysis the vibration signal, including spectral analysis, axis path analysis; the second subsystems to achieve the normalization of entropy data; the thirdsubsystem achieve four algorithms to optimize the initial weights and thresholds of the neuralnetwork; the forth subsystem can optimize the parameter of neural network and distinct theunknown data sample. The experimental results verify the effectiveness of the algorithm andsystem.