Research On Intelligent Diagnosis Of Gearbox Fault Based On Spectral Patterns

Gearboxes are mechanical devices commonly used in power transmission parts. Due to the effects of impact load and the working environment and fatigue, aging effects, failure of hinges, bearings and gears are inevitable.In this paper, gearbox status analysis is based on the quantitative characteristics extracting vibration signals, including statistical characteristics of time domain waveform, spectrum distribution of the statistic characteristics of wavelet packet frequency band energy distribution. Gearbox vibration signal is highly nonlinear, complex and irregular. Fractal analysis can be carried out through a nonlinear signal analysis, looking complicated phenomenon which can reveal its irregularities inherent in law by means of global and local self-similarity through local image as a whole, or by recognizing an overall deepening the understanding of the local. Trend of vibration signal based on the analysis of multi-fractal method implementation on the basis of multi-parameter distribution characteristics, tectonic features of vibration signal based on multi-fractal parameters indicators allows the nonlinear characteristics of the signal to be more fully and effectively utilized.From multiple dimensions of indicator parameters reflect the vibration signal contains information. Among characteristic indexes of the same feature set appears inevitable for vibration signal contains significant differences in sensitivity of information, and there is a correlation between the different indicators. Time domain and frequency domain statistical characteristics, for comprehensive utilization of energy and multi-fractal characteristics, this paper uses of principal component analysis and principal component analysis of three kinds of character sets for integration, extraction integrated three characteristic features of vibration signal of the main units, compared with unilateral characteristics principal components, contain more information, will help improve the gear box running accuracy.BP neural networks is used to learn the relations for data distribution model and gearbox running conditions, comparatively study of single identity data and data integration features trained neural network performance difference. BP neural network convergence is slow and easy into a local optimum, a comparative study of various improved methods to enhance the performance of a single neural network, multiple neural network integration methods enhance the stability of neural networks.