Dampak Loss Early History dan Missing Failure Data pada Model Keandalan Weibull dan Crow Amsaa

Abstract

The accuracy of Remaining Useful Life (RUL) estimation is an important component in optimizing Condition Based Maintenance (CBM) strategy determination, however, the RUL model prediction results are often affected by the phenomenon of incomplete historical data in the industry, especially on legacy assets. This study investigates the quantitative impact of two phenomena of incomplete historical data which include early history loss which will impact the start date determination and the phenomenon of missing failure data which can reduce the number of failure sequences on the performance of the 2-parameter Weibull reliability model and Crow-AMSAA. The study was conducted with an experimental approach based on data simulation, with the main data set which also serves as a benchmark varied using scenarios of early history truncation and data elimination (25%, 50%, and 75%) with variations in simple random sampling for the start date determination and data elimination scenarios. The modeling results show that the phenomenon of early history loss and missing failure data can distort the shape parameter and scale parameter values, causing deviations in the results of the Weibull reliability curve and Crow-AMSAA cumulative failure. However, the Weibull Distribution model is proven to be more robust to start date uncertainty in the intact data scenario (loss of 0% early historical data). This is because the Weibull reliability curve is modeled based on the Time Between Failures (TBF) interval, so changes in the start date will only affect the first TBF calculation and will not affect the TBF in subsequent data.