Research Of Classification Method For Fault Data Of Rotor And Development Of Test Information System For Test Bench

With the rapid development of information technology, controlling theory andartificial intelligence, rotating machinery fault diagnosis technology appears lowinformatization level of automation, real-time and intelligentialization, which cannot satisfy the reality application requirements in advanced industrial production.Therefore, traditional mechanical equipment tries to extend to the integrate device ofmechanical equipment and information, which has the characteristics of controllingautomation, real-time monitoring and diagnosing intelligentialize, will become thefrontier of mechanical fault intelligent diagnosis technology. The science problemresearch of fault data classification has becoming the core content which canimprove fault pattern recognition technology.This study took the rotor system which is the core component of rotatingmachinery as the research object, made use of common statistical features and datamining methods comprehensively, focusing on the realization of classification forfault feature datas. And with the help of virtual instruments, a vibration experimenttesting and feedback controlling system for double cross rotor test bench isdeveloped. Concrete researches carried out and main results obtained are as follows:1) On the basis of analysis of common statistical features, features which cancharacterize different running state of the rotor were selected and extracted fromsignals. Fault feature data set established by features appears high dimension andbad separability. On the basis, existing problems of interfence by nonlinear anddefect of dimension reduction criterion for classical dimension reduction method areconcluded.2) The essence and internal relationship between Principal ComponentAnalysis and Fisher Discriminatory Analysis was discussed and summarized byformula derivation. On the premise of theory analysis above, a dimension reductionmethod of data set by Biased Fisher Discriminatory Analysis was introduced, anddemonstration by examples was also provided, which shows that based on achievingthe performance of FDA, BFDA is simpler.3) Based on dimension reduction method of data set by the proposed method ofKernel Principal Component Analysis combined with Fisher Discriminatory,equivalence relation in fitness function of PSO was deduced. And a dimensionreduction method of data sets by Kernel Principal Component Analysis combinedwith Biased Fisher Discriminatory was proposed further. Scheme of PSO based onBiased Fisher Criterion was also proposed. Two dimension reduction algorithm proposed was applied to reduce dimension of fault feature data set, and classificationverification was implemented by inputting result of dimension reduction into theclassifier designed. Classification effects are both notable.4) Taking advantage of virtual instrument technology, a software platform ofcondition monitoring and feedback controlling for a double span rotor system wasdeveloped. Software and hardware capability of existing double span rotor systemwas expounded comprehensively. And the platform was characterized by intuitiveman-machine interaction, convenient development and maintenance, extendingfunction easily. On the basis of realization of conventional controlling andmonitoring capability, several difficulties existing in condition monitoringcapability was studied and developed.Researches showed that feature data set contains a large number of informationwhich can reflect running state of the rotor, so how to obtain new breakthrough inthe research of data mining algorithm, and how to embed the research results into theautomatic testing and controlling system reasonably, will be an important directionof fault diagnosis research.