Multi-Model Predictive Maintenance: Overview and A Linear System Perspective

Multi-model predictive maintenance is a recently trending approach where information from different inputs and models are processed in a unified fashion in order to provide better failure prognostics. Multi-model systems aim to make clever use of labeled or unlabeled sensor observations, knowledge bases and device specific constraints. In this paper we present an overview for both single and multi-model approaches and provide a linear system perspective. The proposed system leverages linear classification and prediction methods such that the measurement space is partitioned via linear boundaries into nominal and failure operation regions and the distances from boundaries are tracked for providing remaining useful life calculations. The adopted linear approach simplifies the design process and it can efficiently incorporate different modalities in order to enhance the prediction performance. Also, the proposed method is simulated in generated dataset.