Remaining Useful Life estimation for production devices in the automotive industry based on health threshold and cyclical data streaming

Remaining Useful Life estimation based on stochastic methods has been widely used with successful results to forecast the first hitting time of a failure threshold on turbomachinery, as well as in navigation systems of real weapons. Nevertheless, not enough attention has been paid in the implementation in real production environments. Most of the Remaining Useful Life estimation methods implement an online adaption with each new data input, and with this they generate an updated estimation of the remaining useful life. However, given the conditions in the production, in some applications would be easier to generate an estimation based on remaining cycles before failure. Remaining cycles would be helpful to make maintenance decisions based on available produced parts, reducing undesired uncertainties generated by long pauses on the production bounded to human schedules. In this research, we implement Remaining Useful Life estimation on real production devices based on Bayesian prognosis, Wiener process, and Monte Carlo simulations. These methods are adjusted to deliver a result suitable for decision making in production environments. Moreover, a health score is implemented to generate an estimation only after some wear on the device is detected.