Research On Data Mining Technology Of Computer Testing System For Mechanical Engineering

With the development of science, technology and economy, objects of testing, such as weapon system, aerial vehicle, flow-line manufacturing system, are increasingly larger, complicate and modernize. So much more and much higher requirements are put forward toward testing system now. These requirements urge testing system develops toward the direction of intelligence, automation and net connected. Now, technologies of computer, automation and communications are widely used in testing systems. Though a mass of process data are collected and stored, these data which contain information about running state does not be used effectively, and causes the so called phenomena "abundant data, absent information". How to transform these data into useful information and excavate hidden information so as to improve supervisory ability, is a hotspot of testing research, and also the focal point of this paper.The background of this paper is the project of "Research on Testing Technology of Large Scale Linear Accelerometer" , which supported by the project of Foundation Projects of Defence Technology of China Academic of Engineering Physics. This paper takes the technology of data mining as the main line, improves traditional data mining methods aimed at different testing objects according to their respective characteristics, and brings forward some new methods of testing data analysis processing and fault diagnosis forecasting. The work mainly includes two parts, one is static data mining towards data produced by net testing system of linear accelerometer, and the other is dynamic data mining towards data prodeced by monitored control systems such as aerial vehicle and flow-line production systems. The research findings and core ideas are summed up as follows.(1) Introduces the project's background and main targets of the testing data base and data mining system of linear accelerometer, and put forwards general design proposal. Discusses key technologies, including the process of data base setting, the application of data mining, the interface design of Matlab and external program, and so on. Introduces each part of the system by running different examples.(2) Investigates Service Oriented Architecture (SOA), and applies this method in architecture analysing and designing for net testing system which oriented to data mining service, and makes up the prototype system. The system adopts data mining guide based on instance reasoning to help users fulfils the whole process of knowledge discovery from data cleaning-off to knowledge presentation. The system provides effective methods for sharing data mining service of web distributing testing system, and makes for resource sharing and service intefration.(3) As a kind of non-supervisory learning method, clustering analysis plays important role in unknown information discovery towards mass data, and also be investigated widely. Based on analysis of the algorithms ground on density and gridding, this paper advances a kind of equivalent regulation for denseness cell identify and density reachable searching, thereout put forwards a clustering algorithm CLGRID based on gridding and density. By clustering within different phases and expanding classes by selecting seed objects, this algorithm fulfils the aim of reducing extent query times, cutting I/O spending and realizes speediness clustering. Aims at the problems of DBSCAN of hard to choose parameters and to discover clusters with large difference on density, this paper brings forward an improved multi-density threshold DBSCAN algorithm, which adopts gridding density matrix to protract density distributing chart, automatically ensure density administrative levels partition, and obtain much more elaborate clustering results on multi-density levels through multi-density level clustering process, so settles the problem of clustering quality depravation for the use of global value of e.(4) Time sequence data mining is an important content of data mining, and time sequence mode mining is to get potential useful knowledge or information through digging sequence mode within time sequence data. Among the fields of aviation arms supervise and complex flow industry inspect, aims at the characteristics of multi-variable, parameters time-varying, and serious associate coupling among variable, adopts the method of sequence mode mining, which has obvious preponderant at the aspect of relating discovering, to find out associate coupling relationship in the process of fault discovering, so as to provide bases for fault diagnose. This paper analyses the dynamic characters of complex dynamic system inspect data, such as higher dimensional, time-variation, non-synchronization, put forwards the method of labeling abnormal nod, fault time window restrict, and time sequence simplify based on character factors, transforms multi-dimensional time sequence data into symbol sequence aggregates which are fit for data mining. The algorithm of PrefixSpan incarnates fault characteristic information's sequence mode, and instances are taken on emulation platform of chemical engineering production process simulator TE. (5) On the bases of research productions of researchers home and overseas, aims at the demands of industry supervise and testing systems' online fault recognition, put forwards the increment DFT algorithm, mapping the supervised data from time series to frequency-domain through increment DFT transform, and takes some coefficient in time series frequency-domain which contribute the most for fault classifying as the characteristic coefficient to fulfil comparability query based on Euclidean distance, so as to achieve online fault forecast.