Estimating the Reliability of Consumer Electronics from User Survey Data and Test Data

When a manufacturer develops a new product, assessing its reliability is challenging because it has insufficient historical data. However, reliability assessment is needed to avoid releasing a product with low reliability, resulting in financial loss due to returns, replacements, or repairs, and may be dangerous to users in safety-critical products. Collecting reliability data from returned products, as a common approach, is time-consuming. Reliability data from user surveys can be collected faster than data about returned devices. Reliability assessment based on user surveys has several difficulties, however. First, using the calendar time for reliability assessment ignores the effect of different stress profiles, which describe the frequency and intensity of use over time. Therefore, there is a need to define an equivalent usage time (or cycle) for products that undergo different stress profiles. Second, survey responses may be biased, and using this data may result in an inaccurate reliability estimate. This paper proposes a mathematical framework for analyzing new product’s reliability using the user survey data and reliability test results. A stress index (SI) model is introduced to combine different stress adjustors (i.e., the conditions or activities that increase or decrease the impact and damage from the applied stress or enhance stress absorption). A stress-life model is then used to convert the SI values into the usage time or cycles and estimate the time-to-failure distribution. The bias of the survey data is estimated by minimizing the Kullback-Leibler (KL) divergence value between the time-to-failure distribution estimated from the unbiased reliability results such as reliability test data and the distribution estimated from the user survey data. The remaining parameters of the stress-life model are approximated by integrating the bias-removed survey data and reliability test data through the hierarchical Bayesian method. This paper presents a case study that applies the proposed approach to an electronic device using simulated user surveys and reliability test datasets. To validate the approach, it is assumed that the simulated test data follows a known reliability model, which is called the “true model”. The analysis shows that the true model and the estimated reliability models from the test and survey results are close.