An adaptive ARX model to estimate an asset remaining useful life
MetadataShow full item record
Condition-Based Maintenance (CBM) is a maintenance policy that addresses the problem of managing production equipment through the monitoring of variables associated with functional failures. Using CBM prognostic techniques, the Remaining Useful Life (RUL) of an asset can be forecasted, enabling better maintenance decision making. In this field, a wide variety of CBM methodologies have been proposed to estimate RUL; however, most of them require a significant amount of pertinent historical failure data to train the correspondent models. This situation is not always feasible in practical applications. Moreover, techniques that address this issue often ignore exogenous variables that affect the degradation process. To address this situation, this research evaluates the prognostics capabilities of an adaptive autoregressive with exogenous variable model (ARX) and a non-linear Autoregressive model (NLAR) to estimate the RUL of an asset. Performance was evaluated using a simulated aluminum alloy crack-growth model under different scenarios considering processes with constant parameters as well as gradual and stepped changes in the parameters. In addition, to illustrate their application, ARX and AR models were compared with real data from the IEEE 2012 Data Challenge. Results showed that, when RLS were used to update parameters, ARX performed better when degradation process is linear, and AR and NLAR are more suitable for non-linear degradation trends. Furthermore, it was found that RUL estimation improved significantly after reaching the 75% of degradation threshold, in every scenario. Maintenance practitioners might find this approach suitable to manage equipment where no a priori information is available to estimate the RUL of an asset.