Semiparametric constant stress accelerated life tests based on artificial neural networks

Artificial neural networks (ANNs) are powerful predictors that have demonstrated remarkable performance among a range of applications, including reliability engineering. In this study, semiparametric modeling for constant stress accelerated life test (CSALT) based on the P-spline approach and its ANN representation under the type-II progressive censoring scheme (T-II PSC) is provided. In the proposed model, failure times follow the generalized exponential (GE) distribution, and relationships between accelerated stresses and characteristics of the lifetime distribution are nonparametrically and additively modeled by the B-spline approach. A penalty component within the loss function serves to regularize the model, and one or more smoothing parameters control the amount of regularization. A gradient-based optimization method, such as the Adam algorithm, is employed for estimating the parameters of the model, and the smoothing parameters are selected by minimizing the Akaike information criterion (AIC). The performance of the proposed model is investigated by a simulation study and a real data set analysis. Furthermore, the partial derivatives (PaD) method is used to conduct sensitivity analysis in the real data set.