SPARSITY: using data from sparse measurements for predictive maintenance

pVehicles, manufacturing and printing equipment, consumer and medical devices, robots, wind turbines: they are examples of systems that could benefit from Predictive Analytics (PA). For instance, PA can leverage data science to improve service life through predictive maintenance and to build better ones via predictive design. An outstanding challenge for PA is the need of large amounts of data to develop, identify and validate algorithms. One way to overcome data scarcity is via synthetic data generation from so called Digital Twins (DT), that is dynamical mathematical models that are used in lieu of the physical system, even before it is built. Anyway, DT models development still need data. We plan to solve this conundrum by extending modelling and identification methodologies in order to cope with a particular kind of data scarcity: data sparsity. We will target cases were data is not continuously available over time, and fragmented over a large population of similar devices. This is relevant for mass-produced systems where practical reasons prevent data to be collected, transmitted and processed continuously for the entire population. This novel approach will be validated in an industrial Proof of Concept for the diagnosis and prognosis of real automotive components./p