Identifying Changes in Degradation Stages for an Unsupervised Fault Prognosis Method for Engineering Systems

The maintenance strategy known as Condition-Based Maintenance (CBM) has become increasingly popular as it optimizes asset availability by minimizing maintenance downtimes and reducing overall maintenance costs. To do so, it analyses asset monitoring data to forecast the degradation and prevent failure before it occurs, a process called fault prognosis. This process generally comprises four basic steps: data acquisition, construction of a Health Indicator (HI), identification of the Health Stages (HS), and prediction of the Remaining Useful Life (RUL). Nevertheless, it is usually dependent on prior knowledge of a failure threshold, thus enabling the prediction of the RUL. In cases where this information is not available, a different prognosis approach is required. Therefore, rather than predicting the RUL, the proposed method intends to indicate the proximity of the failure occurrence based on the premise that during the development of the fault, breakpoints associated with the acceleration of the degradation rate occur. In this way, evaluating only the HI behavior, without considering previously monitored data, the proposed method could be applied to machines whose faults of interest had not yet been observed. To validate the method, it is applied to synthetically generated HI data with different behaviors over time. Results show that the method has the potential to be used in scenarios where there is no previous information on the degradation pattern.

Keywords: Fault prognosis, Unsupervised prognosis, ARIMA, Adaptive prognosis, CBM.

Download PDF