A dual-purpose data-model interactive framework for multi-sensor selection and prognosis

Prognostics for multi-sensor monitored stochastic degrading systems has been attached great importance in the last decade. In this context, it is desirable to achieve the informative sensor selection and data-model interaction between the composite health index (CHI) constructing and the stochastic degradation modeling in a joint framework. To do so, we propose a dual-purpose data-model interactive framework for multi-sensor selection and prognosis. Specifically, based on the CHI extracted from multi-sensor historical data and the associated lifetime prediction via stochastic degradation modeling, a prognosis accuracy-oriented objective function with regularization is constructed to realize the joint optimization of the fusion coefficients and the failure threshold of the constructed CHI. As such, the dual-purpose task for automatic sensor selection and prognosis can be achieved simultaneously. In addition, the Box-Cox transformation is applied to the constructed CHI to endow the proposed method with the ability of handling nonlinear degradation and improving the early prognosis performance. To conduct the prognosis task for the in-service degrading system, the Bayesian method is applied to update the degradation model parameters and the remaining life distribution for prognosis is derived and updated. Finally, we validate the proposed method by multi-sensor data of aircraft gas turbine engines and milling cutters.