A visualization approach for unknown fault diagnosis

Since visualization can provide useful information to control engineers about the state of the process, visualization has become an dispensable item in the condition monitoring toolbox. The objective of this paper is to propose a visualization approach and apply it to unknown fault isolation. First, the data-driven parity space (PS) technique is used to identify the stable kernel representation (SKR) of a linear time invariant dynamic system. Then, the signature directions (SDs) and the current directions (CDs) are defined, based on which the detection and isolation rules are proposed for diagnosing both the known faults (KFs) and the unknown faults (UFs). Finally, a visualization approach is provided for projecting high-dimensional fault information onto a lower dimensional and drawable space. This approach maintains the fault isolability so that engineers will be able to diagnose the faults more reasonably. The proposed visualization approach is applied to a vertical take off and landing (VTOL) aircraft model and a glass tube manufacturing process.