Flexible architecture for data-driven predictive maintenance with support for offline and online machine learning techniques

Predictive maintenance requires the constant monitorization of equipment and the accumulation of data captured from sensors, industrial equipment, and existing management software. This data must be cleaned and processed before being used to train machine learning models that will generate different outputs of interest, such as fault prediction, fault detection, estimation of an equipment's remaining useful life, among others. Considering these requirements and the different technologies needed to accommodate them, we present an architecture for predictive maintenance, based on existing standard architectures for Industry 4.0, that not only supports the implementation of all stages of predictive maintenance, but is flexible enough to be applied in distinct industrial scenarios. Moreover, the architecture is capable of accommodating both offline and online data pre-processing and machine learning techniques