Reliability assessment for stress relaxation considering heteroscedasticity among accelerating levels

- Stress relaxation is a common failure mechanism for many components and products in long-term loading. Accelerated degradation test (ADT) has been demonstrated to be an effective solution to collect relaxation data and then predict lifetime and assess reliability. Traditional ADT data modeling approaches, however, can only describe the time-varying characteristic of dispersion while ignore the heteroscedasticity between the accelerating variable levels. We have investigated many relaxation datasets and conclude that dispersion of lifetimes enlarges as stress level drops. To well describe this characteristic, a novel methodology is proposed in this paper, where lifetime obtained by general degradation path approach is considered to follow a normal distribution with linear mean and standard deviation (std) functions against temperature by a proper transformation. Maximum likelihood estimation (MLE) for model parameters via genetic algorithm (GA) is further presented to enhance the analysis precision. Empirical results for helical compress spring relaxation data indicate that, compared with the conventional homoscedasticity model, the constructed method can provide a better modeling accuracy and an extensive adaptation.