简介：Wehavedevelopedafuzzyneuralexpertsystemthathastheprecisionandlearningabilityofaneuralnetwork.Knowledgeisacquiredfromdomainexpertsasfuzzyrulesandmembershipfunctions.Then,theyareconvertedintoaneuralnetworkwhichimplementsfuzzyinferencewithoutrulematching.Theneuralnetworkisappliedtoproblem-solvingandlearnsfromthedataobtainedduringoperationtoenhancetheaccuracy.Thelearningabilityoftheneuralnetworkmakesiteasytomodifythemembershipfunctionsdefinedbydomainexperts.Also,bymodifyingtheweightsofneuralnetworksadaptively,theproblemofbeliefpropagationinconventionalexpertsystemscanbesolvedeasily.Convertingtheneuralnetworkbackintofuzzyrulesandmembershipfunctionshelpsexplaintheinnerrepresentationandoperationoftheneuralnetwork.

简介：ThroughthepracticeofdevelopinganexpertsystemFBCDES(FeedBoxConceptualDe-signExpertSystem)fortheplanomillingmachineinCIMSERC(CIMSEngineeringResearchCenter)inChina,thepaperintendstointroduceacommonproblem-solvingstrategyforthemechanicalproductconceptualdesignexpertsystem.Meanwhilesomeproblemssuchassystemarchitecture,knowledgerepresentationsandinferenceenginesetsarealsodescribed.Ayetmoredetaileddiscussionisgiventothetechnicalproblemsforrealizingtwokeymodules,thatisconceptdesignmoduleandlayoutmodule.