Remaining Useful Life estimation for MEMS-based transducers

The use of MEMS-based measurement systems is experiencing great pervasiveness. This is mainly due to the small size that allows for a small footprint and use in previously impossible locations. For these reasons, they are also used in the medical field, within the Human Activity Recognition field, for trials and early diagnosis of diseases such as Parkinson's disease. Given the criticality of such applications, it is necessary to ensure the proper functioning of these devices by accurately defining their useful life. To date, however, there are no unambiguous parameters to estimate the health status of MEMS devices and, thus their Remaining Useful Life (RUL). This work aims to propose an RUL estimate by employing Machine Learning techniques capable of synthesizing a Health Index from the main temporal and frequency features of such devices. The results obtained were satisfactory, allowing the prediction of device failure more than 30 days before the effective breakdown.