Study Of Fault Diagnosis Based On Data Mining

With the rapid development of science and technology and modernization of large equipments tending to structure complex and multi-functional, fierce competition have a higher request in equipment accuracy. Once a failure occurs, a chain-reaction will force the entire production at a standstill. If the fault doesn't be discovered timely, it would affect the daily work, which causes severely casualties and major economic losses, and even catastrophic consequences. So prediction and monitoring of the fault has caused extensive concern in various fields. At the end of the 20th century, data mining as an intelligent data analysis techniques have been widely used in many fields. By data mining, useful or potentially useful knowledge can be found, which contains lots of information, and can generate rules or laws.Firstly, this paper researches abnormal vibration of motor failure in the work of steel fabricating machine. Unusual motor failure of rotor imbalance, rotor misalignment and rotor rubbing are analyzed. And with the analysis of frequency can we get training sample set by characteristic signal.Second, the paper especially studied classification algorithm of data mining in the application of fault diagnosis. The diagnosis method based on data mining, actually, is a method of have the fault signal classified and induction. According to the spectrum signature of fault signals to extract the characteristic frequency, and categorize the fault signals, obtain different fault type. In this paper, I select decision tree. As a kind of classification method, decision tree can generative easy to understand rule by learning a small training sample. Decision tree algorithm of short time is suitable to practical working system. There is more redundancy in a training sample. Therefore, attribute was reduced by discernible matrix algorithm of rough set theory. Method of discernible matrix algorithm is: based on constructed discernible matrix, a discernible function of conjunctive normal form is elicited. Then converts conjunctive normal form to alternative normal form, and each conjunction expression of alternative normal form is the knowledge reduction. With decision tree classify the reducted failure sample, this can reduce computational complexity. And generate a brief and compact rule.Finally, diagnosis platform is illustrated. There are three layers C/S structure. Application function is divided into three parts:the presentation layer, the functional layer, and the data layer. After partition hard, the three parts keep its logically-independent. And flexibility of the functional extension is guaranteed. Based on store failure data on Server and new failure date modifying the rule, classification accuracy of failure signal can come ture. With real-time monitoring effectively of vibration, abnormal information can be detected in time so as to make decision ahead of time, which can stop the great incident.The innovation of this topic is that combined decision tree classification algorithm with discernible matrix algorithm used in generating classifying rule of failure message. It is showed from the results of definition of potential knowledge which is improved by the generating decision tree after attribute reduction. Based on reduce the condition attribute, information content from database is shrinking gradually. Bother from large amount of miscellaneous, which laid a good foundation for the efficient knowledge discovery.