An artificial immune and incremental learning inspired novel framework for performance pattern identification of complex electromechanical systems

Performance pattern identification is the key basis for fault detection and condition prediction, which plays a major role in ensuring safety and reliability in complex electromechanical systems (CESs). However, there are a few problems related to the automatic and adaptive updating of an identification model. Aiming to solve the problem of identification model updating, a novel framework for performance pattern identification of the CESs based on the artificial immune systems and incremental learning is proposed in this paper to classify real-time monitoring data into different performance patterns. First, an unsupervised clustering technique is used to construct an initial identification model. Second, the artificial immune and outlier detection algorithms are applied to identify abnormal data and determine the type of immune response. Third, incremental learning is employed to trace the dynamic changes of patterns, and operations such as pattern insertion, pattern removal, and pattern revision are designed to realize automatic and adaptive updates of an identification model. The effectiveness of the proposed framework is demonstrated through experiments with the benchmark and actual pattern identification applications. As an unsupervised and self-adapting approach, the proposed framework inherits the preponderances of the conventional methods but overcomes some of their drawbacks because the retraining process is not required in perceiving the pattern changes. Therefore, this method can be flexibly and efficiently used for performance pattern identification of the CESs. Moreover, the proposed method provides a foundation for fault detection and condition prediction, and can be used in other engineering applications.