Research On Intelligent Diagnosis For Equipment Fault Based On Knowledge Management

Equipment maintenance and fault diagnosis are the important guarantees foradapting the globalization process in modern manufacturing enterprises. Research andapplication of advanced plant maintenance and fault diagnosis model will ensure thesafe and orderly production, meanwhile improving the reliability and effectiveness ofproduction equipment or devices. As a diagnostic maintenance mode combiningartificial intelligence technology and traditional fault identification method,intelligent fault diagnosis (IFD) can integrate the inference and decision-makingcapability of diagnostic and maintenance knowledge to achieve the optimization ofdiagnostic reasoning results and maintenance decision via efficient management fordiagnostic knowledge and dynamic allocation for maintenance process.Diagnostic and maintenance knowledge is one of the core resources in intelligentfault diagnosis and critical support element of maintenance process. By effectivelyobtaining, transmitting, processing and sharing of the diagnostic information,intelligent fault diagnosis takes advantage of intelligent diagnostic inference andflexible diagnostic strategy to make the right judgments and decisions for operationalstatus and failure of supervisory object, thereby improving the quality and efficiencyof diagnosis and maintenance work and providing support to the efficient managementfor diagnosis and maintenance knowledge resources. In order to upgrade the level ofdiagnostic and maintenance knowledge resources management in the intelligent faultdiagnosis effectively, sponsored by the National High Technology Research andDevelopment Program of China (863Program)(No.2009AA04Z414), the presentdissertation did a profound and systematic research on the key technologies ofknowledge based intelligent fault diagnosis.The main contributions of this dissertation are summarized as follows(1) Considering the demands of the diagnostic and maintenance knowledgeresources management, a knowledge-orient intelligent diagnosis (KOID) model wasintroduced in the present dissertation. In the KOID model, diagnostic andmaintenance ontologies was adopted, Bayesian networks, uncertain knowledgemanagement and sensor networks technology were employed in integration andreasoning of diagnostic and maintenance knowledge, and then intelligent maintenancemode was formed by diagnostic-maintenance-process-centric method. A formal definition of model was introduced to identify the components of KOID model. Theconnotation, characteristic and application mode of model were discussed. Thearchitecture and support system of model were designed from the perspective ofengineering practice.(2) To solve the problem of unsatisfactory integration, bad adaptability and lowutilization of condition data in the existing condition monitoring applications, anintelligent condition monitoring system based on wireless sensor networks (WSNs)was investigated and established for industrial facility, and the core of the system wasthe design of intelligent sensor networks nodes based on MSP430microcontroller. Byutilizing the signal analysis ability of embedded processor to realize localizedprocessing and data fusion of condition data, a distributed condition monitoring modewas achieved to combine data acquisition with signal processing functions, and sensornetworks for condition based maintenance, which possessed preliminaryself-diagnosis ability, were formed.(3) A semantic representation for diagnostic maintenance knowledge wasintroduced. By establishing the models of equipment structure information, empiricalmaintenance knowledge and diagnostic process, an ontology driven inference modelof fault diagnosis was established. An ontology mapping algorithm was proposed forthe mapping between the devices’ operating status and fault symptoms, and adiagnostic instance matching algorithm was proposed to map the symptom space intothe fault case space. As a result, the static maintenance knowledge and the dynamicdiagnostic process were consolidated, furthermore, the automation andintellectualization of fault diagnosis and maintenance decisions were achieved.(4) A hybrid fault probabilistic reasoning model, which combined ontologysemantic representation and Bayesian networks, was proposed. An ontology-baseddiagnostic Bayesian networks (OntoDBN) was established. By extractingmulti-source heterogeneous diagnostic information and non-structured experts’knowledge with OntoDBN, a diagnostic semantic knowledge model was constructedwith probabilistic extension. Abnormal working conditions were identified usingBayesian classifier, and a fault probabilistic inference algorithm based MPE (MostProbable Explanation) was given, consequently diagnostic explanations could beobtained by the algorithm in accordance with the operating conditions, faultsymptoms and evidences.(5) In the process of fault diagnosis and maintenance decision-making tasks,which involve high levels of uncertainty, a maintenance group decision making method in view of the uncertain nature was proposed. On the basis of integration andmodeling for multi-source heterogeneous manufacturing process knowledge, Bayesiannetworks and fuzzy AHP (Analytic Hierarchy Process) reasoning were used to processdiagnostic reasoning and fault causes analysis, and the optimal solution ofmaintenance decision-making was given by combining Bayesian networks and fuzzyAHP with the empirical knowledge of diagnostic experts.(6) In order to ensure the KOID model has a practical significance, consideringthe demands of equipment maintenance and fault diagnosis in a petrochemicalenterprise, a corresponding prototype system had been developed, validated, andimproved. Together with the implementation of intelligent fault diagnosis in theenterprise, the implementation method of KOID model had been designed. Theapplication of the prototype system demonstrated the effectiveness and feasibility ofthe KOID model in practice.The preliminary study of this dissertation showed that integrating intelligentmethods, such as WSNs, ontologies, Baysian networks and FAHP et al., into thedomain of equipment maintenance and fault diagnosis is feasible and effective. Theresults of this study can provide new ideas and practical experience for furtherresearch and development of intelligent fault diagnosis.