Latent Fault Diagnosis for Liquid Launch Vehicle Using Belief Rule Base With State Miner

Abstract

A new latent fault diagnosis (FDs) method is developed for a liquid launch vehicle. The proposed method aims to solve three challenges: lack of failure data, limited expert cognition, and new system latent state triggered by faults. As an interpretable method, the belief rule base (BRB) model can both combine the data and knowledge that can solve the first two problems. It provides a basis for FDs of the vehicle. However, when the vehicle fails, its internal mechanism changes, and the new system state may exist. Limited by the output framework of BRB, it cannot detect these latent faults. Hence, a new BRB with state miner (BRB-M) is proposed with an adaptive discernment framework. It can mine the new system states by the combination of output propositions. Then, the traceability analysis of BRB-M is conducted based on the transparency of its modeling process, and the influence of each input characteristic is analyzed quantitatively. To improve the diagnosis accuracy, an optimization model is put forward for BRB-M. To illustrate the performance of the proposed method, an experiment of vehicle is conducted. In the experiment, the diagnosis accuracy is 97.00%, and increases 29.11%, 23.17% compared with the fuzzy theory and BP neural network.