Application Of Dynamic Bayesian Networks In Reliability Computation Of Aero Engine Turbine Blade Disk Systems

Bayesian network, which are expressed by the network structure and probabilitydistribution table, is a inference network. Dynamic Bayesian networks is combine bystatic Bayesian networks and time factors, mainly working for the system which ischanging the parameters over time.The objective of this paper is to use dynamic Bayesian networks, throughimproved network modeling and network inference, to large-scale complex mechanicalsystems.In the second chapter, this paper puts forward the Dynamic Bayesian networksmodeling methods of improvement. The traditional typical connection Bayesiannetworks modeling and fault tree Bayesian networks modeling can’t solve theconnection complex, multi-state node system. Minimal path sets Bayesian networksmodeling transform complexly and is a heavy workload. Through the adjacency matrixof the system, set up and judge a tree structure. The systems can be divided into singleconnectivity systems and multiply connected systems. Through the logical relationshipof nodes and connecting lines, simply connected system is establish of the Bayesiannetwork directly. Multiply connected system sets the Bayesian networks by the way ofminimal path sets.In the third chapter, the paper introduces the dynamic Bayesian networks inference.In the inference process of dynamic Bayesian networks, the form of the probabilitydistribution of the system nodes is not uniform. A number of points are selected in thetime coordinate parameters and extracting all the nods of the probability of discretevalues. By the way of discrete of the continuous data, the dynamic Bayesian networksare transformed to a plurality of static Bayesian networks. Bucket elimination methodis widely used in Bayesian network inference, can be applied to simply connected andmultiply connected Bayesian networks. Simplify the calculation process by changingthe variable elimination order.In the fourth chapter, proposed the improvement method of Bayesian networksinference. In the inference process of simply connected Bayesian networks, the improved bucket elimination method calculates directly using the probabilitydistribution table. In the inference process of the multiply connected Bayesian networks,working by the introduction of Boolean truth table based of bucket elimination. Thepresentation is more concise than the bucket elimination method. In the calculationprocess, the bucket elimination method can only solve a joint probability. The improvedbucket elimination method can solve multiple joint probabilities and improve thecomputational efficiency.In the fourth chapter, the aero-engine turbine blade disk system is a large andcomplex mechanical system. Failure modes distribute in the structure of the blades, theturbine disk and shaft parts. By the methods of stress analysis, thermal analysis,vibration analysis and summary of experience, achieving the failure modes of thesystem and establishing the dynamic Bayesian networks. In the process of turbine bladedisk system by the method of improved bucket elimination, solving the failure rate ofthe Turbine Blade overall system and individual member. Through comparing theresults of exact method, verifying the accuracy and superiority of this methodThe complex system block diagram based Bayesian networks method proposed inthis paper, makes up for the traditional method expression ability is limited andmodeling process of complex faults. The dynamic Bayesian networks are accurately, asmall amount of calculation and concise expression. It becomes an important research inthe direction of the large-scale complex mechanical systems.