Diaz-Gonzalez, Abel.; Coursey, Austin.; Quinones-Grueiro, Marcos.; Biswas, Gautam. (2025).Ìý.ÌýOpenAccess Series in Informatics, 136, 11.Ìý
Predicting how much longer an industrial system will keep working is important for maintenance, because it helps prevent unexpected failures. This study presents a data-driven method for estimating remaining useful life, or RUL, which means how much operating time is left before a system reaches the end of its life. Unlike traditional approaches that try to model every internal state of a machine, this method treats degradation as a random process based on performance measurements and uses a Bayesian particle filtering framework, a probability-based technique that updates predictions as new data arrive. Instead of building a detailed state-space model, the approach directly estimates the distribution of end-of-life timing from observed performance data and also quantifies uncertainty, so it can show not just a prediction but how confident that prediction is. The method further adjusts key filtering settings over time, such as how much random variation to allow and how strongly to correct predictions using new observations, based on both current data and past forecasting performance. The researchers tested the approach on a simulation dataset from an unmanned aerial vehicle, or drone, that includes realistic degradation signals and known performance outcomes, allowing them to evaluate how accurately the method predicts future failure.
