Abstract:
The problem of predicting the degradation rate of technical system characteristics (life time, LT) is usually solved within the framework of the accelerating testing (AT) paradigm under stress effects. However, statistical methods for evaluating the results may be ineffective for AT when the system performance depends on a very large number of factors. Accelerating testing also has its own specifics at the early stages of experimental design work when, having only a small number of device copies, it is necessary to estimate their potential service life to assess the feasibility of continuing the development. The article analyzes the extent to which modern mathematical and statistical models that form the AT methodology, namely, survival analysis, extreme value theory, allow obtaining forecasts of the service life of the designed devices under real operating conditions at the early stages of development/design. The authors indicate the problems of solving the LT forecasting problem using known machine learning tools, consider and propose a heuristic method for solving the LT forecasting problem in real conditions. As an example, the authors consider the prediction of performance degradation for new solar cell designs whose performance tends to degrade. This heuristic refers to the extraction of the Hondrick–Prescott trend from a nonstationary time series that represents the degradation of a quality characteristic. The applicability of the proposed heuristic to predict degradation in other technical applications, particularly, in computer networks, is discussed and justified.
Fulltext:
PDF file (303 kB)