Life testing is vital for determining whether a new consumer or capital good will perform its prescribed function without failure for a desired period. However, many modern products are manufactured to endure for years or even decades, making performing such tests an often tedious and unproductive task. Researchers, such as professor of Engineering Management and Systems Engineering Thomas Mazzuchi, are looking to Bayesian methods for accelerated life testing (ALT) to overcome this obstacle.
In reliability engineering, an environment is defined by stress, such as temperature, vibration, voltage, humidity, use rate, and so forth. New products are typically tested in a normal-use environment; however, obtaining sufficient failure data for products in these conditions is what makes life testing of high-reliability products so costly and time-consuming. ALT can induce early failure and shorten test time by testing in more severe environments, created by applying or fluctuating one or more stressors at accelerated levels.
“Accelerated life testing has been shown to be an effective method for reducing test time, and when the environment can be manipulated using more than one stress, the test time can be further shortened while maintaining the integrity of the failure data,” Mazzuchi stated.
Despite the ability of some Bayesian approaches for ALT to accommodate multiple stressors, most models manipulate only one stress since it is mathematically easier. In their study, “A Bayesian generalized Eyring-Weibull accelerated life testing model,” Mazzuchi and co-authors, Neill Smit and Lizanne Raubenheimer from South Africa Rhodes University, and Refik Soyer from GW’s School of Business, extended the literature on these models by developing a novel Bayesian approach to the generalized Eyring model with Weibull distribution that can incorporate more than one stress.
The study, published in “Quality and Reliability Engineering International” in February, introduced a Bayesian dual-stress ALT model with one thermal stressor, usually temperature, and one non-thermal stressor such as humidity. Being able to manipulate multiple stressors simultaneously is important for reliability testing as it allows engineers to examine the interactions between stressors and explore a wider range of failure modes. Another benefit of the Bayesian approach is that it allows for the inclusion of expert beliefs from those engineers on the parameters of the ALT model.
“By using Bayesian methods and incorporating subjective data or expert judgment into the analysis, we were able to reduce the number of required tests, thus reducing the cost, and produce better statistical results from testing,” said Mazzuchi.
Gathering system and component data via life testing is important for estimating reliability and warranty claims for new products. By developing a new Bayesian ALT model that can incorporate multiple stressors, Mazzuchi and co-authors are not only making this process more efficient and cost-effective but also providing the industry with a more applicable model.