Abstract
For the industrial fault classification, there are still two important issues ignored by the typical Fisher discriminant analysis (FDA). Firstly, because of inevitable process change, online samples may come from new fault modes. Secondly, information contained in available historical samples may not be enough to represent corresponding fault modes. Therefore, an online learning based FDA is initially proposed in this paper. The online learning is achieved by incorporating a novel online sample selection criterion into incremental learning. The selection criterion is developed to choose appropriate online samples to update the classification model continuously without expert labeling. And it can recognize samples from new fault modes through a max-min deviation rule with online updating capability, as well as ensure the reliability and the information content of the chosen online sample by the weighting sum of its maximum class posteriori probability and information entropy. Experiments on the benchmark TE process and a real industrial air separation unit demonstrate the effectiveness and the superiority of the proposed model.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call