Gaining Physical Insights from Degradation Data

Abstract

Problem: The current study was motivated by the need to identify and explore the mechanisms leading to performance degradation and the ultimate failure of a newly designed battery to be used as a backup source for uninterruptible power supplies (UPS) applications. It was also desired to estimate how long (in terms of number of discharge cycles) the battery could serve its intended function before failing to sustain a discharge cycle for the required duration during a power outage.

Approach: An accelerated test (AT) was undertaken to investigate battery performance and degradation under elevated power, temperature, and cycling rate. The AT resulted in both lifetime and degradation data on test batteries. Standard maximum likelihood methods for censored data were used for lifetime analysis and modeling. In addition, a nonlinear mixed-effects model was developed to study the effects of acceleration variables on degradation.

Results: The lifetime-data analysis of the data from the AT provided good life estimates because the resulting data involved little censoring. Nevertheless, the further careful study of the degradation data provided valuable added knowledge by suggesting some underlying physical mechanisms and yielding insights that helped product-reliability improvement. It is generally recognized that degradation measurements provide more information than lifetime data. This holds particularly when there are only a few observed failures and the data are heavily censored. This case study illustrates another important benefit of degradation data even when lifetime data are nearly complete. Such data allow modeling of physical failure mechanisms more directly than lifetime data and, therefore, yields insights not possible through traditional lifetime analysis.

• Arrhenius Relationship
• Battery Testing
• Censored Data
• Failure Mechanism
• Log-Normal Distribution
• Nonlinear Mixed-Effects Models
• Power Law
• Weibull Distribution

Additional information

Notes on contributors

Necip Doganaksoy

Dr. Doganaksoy was a Principal Statistician in Software Sciences and Analytics when this paper was authored. He is currently a Principal Statistician at GlobalFoundries in Malta, NY and an Adjunct Professor at Union Graduate College. He is a Fellow of ASQ. His email address is [email protected].

David B. Hall

Dr. Hall is a Chemical Engineer in Energy Storage and Conversion Materials. His email address is [email protected].