Design of experiments and machine learning to improve robustness of predictive maintenance with application to a real case study

When deploying predictive analytics in a Big Data context, some concerns may arise regarding the validity of the results obtained. The reason for this is linked to flaws which are intrinsic to the nature of the Big Data Analytics methods themselves. In this article a new approach is proposed with the aim of mitigating new problems which arise. This novel method consists of a two-step workflow in which a Design of Experiments (DOE) study is conducted prior to the usual Big Data Analytics and machine learning modeling phase. The advantages of the new approach are presented and an industrial application of the method in predictive maintenance is described in detail.