Online Fault Diagnosis and Tolerance Based on Multiexpert Joint Belief Rule Base for Sensor Failures of Vehicles

As sensor readings are used for vehicle flight control, their reliability directly affects flight performance. This article develops a new fault diagnosis and tolerance (FDT) method for sensor failures of vehicles by addressing three problems: unavailable faulty data, difficulty in establishing the analytical system, and the inconsistence of expert cognitive ability. For the purpose, a new belief rule base model with multiexpert joint (BRB-ME) is proposed. The first two problems are handled by combining the small size of observation data and the uncertain knowledge from multiple experts in BRB-ME. For the third problem, a new multiexpert joint strategy is proposed in the BRB-ME model. The experts first construct their own models, and then, the models are fused with different weights according to the experts' ability, such as research fields and the working time. Then, a new FDT framework is developed based on BRB-ME for detecting vehicles' sensor failures, where the sensor failures are tolerated by the reconstruction strategy for faulty sensor output. Moreover, in order to address the influence of uncertain expert knowledge, an optimization model is constructed for obtaining the optimal solutions for the framework. An experimental illustration is conducted for accelerometer failure. The diagnosis accuracy is 97.50%, and the developed framework can ensure the navigation accuracy of the vehicle under accelerometer failure.

Automated summary

This article presents a new fault diagnosis and tolerance (FDT) method for sensor failures in vehicles. The method addresses three main problems: unavailable faulty data, difficulty in establishing the analytical system, and inconsistency of expert cognitive ability. A new belief rule base model with multiexpert joint (BRB-ME) is proposed to handle these problems. The BRB-ME model combines observation data and knowledge from multiple experts to address the first two problems, and introduces a multiexpert joint strategy to handle the third problem. An experimental illustration is conducted for accelerometer failure, with a diagnosis accuracy of 97.50% and ensured navigation accuracy under accelerometer failure.