Hybrid diagnosis of fault tree and Bayesian network in BIW automatic welding production line

Fault diagnostics are increasingly important for ensuring production line system safety and reliability. In body in white (BIW) automatic welding production line, parts of the fault events are not statistically independent, so the conventional fault tree (FT) diagnosis method suffers from its shortcoming, but fault tree is still the best way to express the prior knowledge of system. On the other hand, the data-driven method Bayesian network (BN) without requirement for the accurate mathematical mode has been widely use in fault diagnosis. However, it is extremely difficult to obtain sufficient failure data set in real production line system so that the BN structure only learning from history fault data is always incomplete. To solve this problem, this paper proposed a fusion fault model combining the advantages of FT model and BN model. At first, based on the structure and working principle of production line, the cause and the type of the fault in welding production line were studied, the relationship between the faults was analyzed and the FT model of welding production line was established. Secondly, the fault sample was extracted from the history fault records and a primer incomplete Bayesian network was structured by using the structure expectation maximization (SEM) structure learning algorithm. Finally, to obtain a more complete Bayesian network, the prior knowledge from fault tree model was used to optimize the primer Bayesian network. These two networks were simulated on MATLAB and the result suggests the proposed fault diagnosis algorithm combining Bayesian network and fault tree is feasible.