Research Of Fault Pattern Recognition And Trend Prediction Methods Based On SVM And GA

This paper proposed fault pattern recognition and trend predictionmethods based on Support Vector Machine (SVM) and the Geneticalgorithm (GA), using SVM for pattern recognition to bearing typicalfault. Meanwhile using prediction model to detect the trend of gear state,and made an optimization analysis for SVM classification process andtrend prediction process using GA respectively. Main works were givenas follows:(1) Based on the advantage of the new learning method supportvector machine which can solve the small sample learning problems, thispaper proposed to use SVM training and diagnosing four state of rollingbearing: normal state, outer-race flaw, inner-race flaw and rolling elementflaw. In order to improve the classification recognition rate，based on theadvantage of the genetic algorithm (GA) space search rule: Survival of the fittest, this article put forward to use GA on SVM to optimize twoimportant parameters: gamma and sigma2. From the experimental resultswe knew, SVM classification recognition rate can be increased after theparameters gamma and sigma2being optimized, making four state ofrolling bearing well separated.(2) To solve the fault diagnosis problem of rolling bearing under lowrotating speed during working, based on the performance advantage ofwavelet transform technology with high and low frequency separation,local refining and the time frequency domain feature extraction and so on,this paper suggested making diagnosis and identification for low speedrolling bearing vibration signals of outer-race flaw, inner-race flaw withwavelet transform technology, and using SVM training and diagnosingfour state signals. From the experimental results we knew, using themethod of wavelet transform with SVM technique for diagnosing thefault type of low speed rolling bearing is an effective method foranalyzing and processing.(3) In order to predict the trend of gear fault, the establishment ofthree order function equation prediction model for gear trend wassimulated, based on the GA good spatial search ability, this paperproposed GA to optimize the prediction model function, do comparativeanalysis research between the new prediction model and two order, threeorder function by linear fitting function. The results show: three order function equation prediction model can realize the gear fault trendforecast simulation after optimizing by GA.