Abstract

Gas-Relief Valve (GRV) plays a role of over-pressure protection to mechanical equipments and its faults will bring unpredictable severe accidents. So it is of important significance to diagnose the faults of GRV effectively. Small failure samples, as well as obtained leak magnitude without overall showing the actual leak situation of GRV, are two factors causing low diagnosis accuracy. A new approach is proposed for fault diagnosis of GRV in this paper. A novel pneumatic circuit acquiring leak upstream of input let of GRV is designed. Using support vector machines (SVM) based on binary tree structure to identify the failure pattern of GRV, is to solve the problem of artificial intelligence methods weak in small samples. The diagnosis results of experiments on 16 GRVs demonstrate the superiority of this approach. Besides leak through leakage hole of GRV, leak caused by deterioration of both fit clearance and mechanical sealing elements, is also acquired successfully. Compared with neural network, SVM achieves a higher diagnosis precision. The influence of the leak on the diagnosis of different running statuses is also discussed.

Paper Details