Applying Principal Component Analysis for Multi-parameter Failure Prognosis and Determination of Remaining Useful Life

As society moves forward into the Fourth Industrial Revolution, new requirements and benchmarks are being established for industry in general. Over time, the goals of Industry 4.0 become clearer and adapting processes to meet the new demands is a matter of survival. In this scenario, Condition-Based Maintenance (CBM) is increasingly applied in several sectors and many tools and methods have been developed to support it. However, one of the most relevant aspects for CBM, the estimate of the Remaining Useful Life (RUL) of machines and equipment, has not always received the attention it deserved. Despite several articles addressing Fault Detection and Diagnosis (FDD) from multiparameter approaches, as a clear adaptation to new times and the need to work with increasingly complex systems, few works address the prognosis issue in the same context. Therefore, RUL estimate is generally carried out based on single parameter analysis. To fill this gap, this paper proposes a RUL estimate approach based on the classic multiparameter method Principal Component Analysis (PCA). In order to demonstrate its feasibility, a case study considering simulated data from a hydrogenerator is presented. From the selection of a specific failure mode, the RUL of the equipment is estimated considering the new proposed approach, which uses results obtained with PCA, as well as the traditional approach, performed directly from the monitored parameters. The results demonstrate that the RUL estimate from PCA results was not only possible, reducing the processing time and effort, but also proved to be more accurate and reliable, since the time to failure estimates always occurred in anticipation of the real event.