Study On The Incipient Fault Detection And Diagnosis Of Gear

As the key part in a gearbox, a gear is the highest rate of failure part in a gearbox,due to bad working conditions and other influences factors. The study on the incipientfault diagnosis of gear has a very important significance to ensure security and efficientoperation of equipment and avoid the accident. It has drawn wide attention by people.In this thesis, the research object is a gear in the stage of early fault, and thecombination of theoretical analysis and experiment study is adopted to study the incipientfault diagnosis of gears. In the theoretical analysis, characteristics of an incipient fault ofgear are analyzed by constructing dynamics model of gear transmission system. Due tothe problem of weak fault feature and much complex, changeable noise in the incipientfault of gear, the best wavelet threshold de-noising method based on comprehensiveevaluation parameter is proposed in this thesis. The de-noising method can find the bestcondition from many kinds of wavelet threshold de-noising conditions, and achieve thebest de-noising effect for a vibration signal. In this thesis, an improved Empirical modedecomposition method is proposed. The improved method is based on the combination ofpiecewise Hermit interpolation of order three with extreme truncation and correlationcoefficient screening method. The proposed method can improve some main problems inthe original method, such as the end effect, envelopes overshooting and undershooting,difficult selection of intrinsic mode functions, and so on, and then achieve the purpose ofextracting the incipient fault feature of gear effectively. According to the problem ofbeing difficult to judge fault types from the extracted features above accurately, a BPneural network is established. This neural network treats the spectral energy distributionas characteristic parameters, and achieves the accurate pattern recognition for theincipient fault of gear. In the experiment study, the de-noising method, the extractingfeature method and the pattern recognition method proposed in this thesis are verified intwo ways strictly which includes simulation experiment analysis and measured signalsexperiment analysis.The result of practice shows that the methods of signal processing and patternrecognition proposed in this thesis can recognize the incipient type accurately, and have avery important practical significance and the practical application value for protecting the equipment keeping reliable and steady condition.