Model-Based Diagnosis with Multi-Label Classification Academic Article uri icon

abstract

  • Model-Based Diagnosis (MBD) is one of the leading artificial intelligence approaches that copes with the diagnosis problem. MBD is known as a hard problem and grows exponentially in the size of the system. In this paper, we propose a novel approach that combines MBD with multi-label classification. We propose to build a classifier that maps symptoms of the system to possible faults. The major advantage of this approach is by reducing significantly the online computational complexity; The learning process of the relations between the observation and the diagnosis is performed in advance offline, and then online, by using the classifier, we can immediately return the diagnosis. This paper addresses several challenges: 1) modeling the MBD problem as a classification problem, 2) generating informative samples for the training set, 3) verifying sound and minimal diagnosis. 1

publication date

  • January 1, 2011