A Novel Subspace-Based Observer for Servo Systems Fault Prediction

In the course of system operation, minor faults are inevitably encountered, which typically do not affect the normal output of the system but may introduce certain safety risks. Therefore, it is necessary to promptly diagnose these minor faults that occur in order to make predictions regarding potential faults. In this paper, a novel observer design method that combines model-based and data-based approaches and is sensitive to minor faults is proposed based on the concept of Stable Kernel Representation (SKR). Firstly, an application form of the system residual observer is presented, utilizing coprime factorization techniques. Subsequently, historical data is decomposed into distinct subspaces to identify the SKR of the system dynamics. Following this, by establishing a fault sensitive system, the observer output is expanded into a two-dimensional space. This expansion enables the output to not only reflect the current operational state of the system but also exhibit heightened sensitivity to minor variations in model parameters. Ultimately, the effectiveness of the proposed approach is substantiated through simulation studies involving a turntable system. In contrast to conventional observers, the improved observer, eliminating the need for a known system model, demonstrates outstanding dynamic performance and high sensitivity to minor faults. Therefore, it possesses the capability to diagnose minor changes in parameters, thereby contributing to the effective prediction of potential critical failures within the system.