ENBIS Spring Meeting 2019 in Barcelona

13 – 14 June 2019 Abstract submission: 15 January – 20 May 2019

Remaining Useful Lifetime Prediction: A case study in combining statistics and machine learning

13 June 2019, 16:55 – 17:20


Submitted by
Alessandro Di Bucchianico
Alessandro Di Bucchianico (Eindhoven University of Technology), Yingjun Deng (Eindhoven University of Technology), Mykola Pechenizkiy (Eindhoven University of Technology)
Calculation of Remaining Useful Life (RUL) is an important new trend in maintenance. It is an essential element to go from diagnostic analytics to predictive or rather prescriptive analytics, to use data science terminology. In Industry 4.0 settings, we have to determine RUL in data rich environments with an abundance of sensor values and degradation indicators but few failures. In such settings machine learning techniques may be useful to get predictions when complex relations between sensor values exist that make statistical modelling not feasible. However, machine learning techniques typically fail short of providing indications of uncertainty. To combine the best of two worlds, we propose an approach that combines machine learning and statistics. Recurrent neural networks are used to provides predictions from degradation values, and we use stochastic degradation models to add uncertainty to the prediction.

The proposed approach to RUL prediction will be illustrated with an industrial use case within the EU Horizon 2020 project Prophesy.
View paper

Return to programme