A new strategy for automotive off-board diagnosis based on a meta-heuristic engine

• Authors: A. Azarian;A. Siadat;P. Martin
• Venue: Engineering Applications of Artificial Intelligence
• Year: 2011

The automotive industries lack of solutions for accurately, comprehensively and efficiently fault localization. However, such services in the after-sales networks are crucial to the brand value of automotive manufacturer and for client satisfaction. In this paper, a new approach for the off-board diagnosis is presented, with significant improvements compared to the current technologies usually based on inference rules. A more robust approach that allows, per the additions of functional modules, to enhance traditional computer aided diagnostic systems towards a global diagnostic engine reasoning on different sources of knowledge with their uncertainties. Once the design of a new vehicle has begun, information like the dependencies between the components could be re-used for the models dedicated to the diagnosis task. Moreover, the economic pressure leads to a high degree of innovation with a massive use of electronics in safety, comfort and entertainment (OCC'M Software GmbH, 2010). This dramatically increases the amount of data to manage for the testing of E.C.U. (Electronic Control Unit) functionalities. The complexity of the subsystems leads to breakdowns that need qualitative symptom description for the fault localization. Finally, a feedback engine automatically completes the expensive models for the diagnosis and returns critical dysfunctions to the design department. In this paper, we present our research on a new diagnosis strategy for complex mechatronics systems. It encompasses the needs and requirements of automotive manufacturer. The results are presented with data obtained from low, middle and luxury class vehicles. They demonstrate the performance in real field conditions of our approach. They are based on the interpretation of observations, the fault localization and isolation, the evaluation of feedbacks for model auto-completion. The novelty in this approach is based on the reasoning of different sources of knowledge (construction and design knowledge, expert knowledge, return of experiences) which leads to an efficient diagnosis. The approach approximates the optimal path from the observations toward the fault isolation with the help of a meta-heuristic engine. These experiences show the potential of our proposed approach for the automotive off-board diagnosis task.