AI and Feature-Vector Based Damage Monitoring and Remaining Useful-Life Assessment for Electronics Assemblies in Mechanical Shock and Vibration

• Main Authors: Lall, Pradeep, Thomas, Tony
• Format: Conference Proceeding
• Language: English
• Subjects: Electric shock; Microelectronics; Micromechanical devices; Mission critical systems; Predictive models; Time measurement; Vibrations
• ISSN: 2833-8596
• DOI: 10.1109/EuroSimE56861.2023.10100747

In order to ensure uninterrupted operation over time-in-service, progressive assessment of damage accrual and remaining useful life is needed to allow for early identification of impending failure. This paper proposes Prognostics Health Management methods for feature vector-based assessment of damage initiation and progression in electronic systems. The methods can be used to detect impending failures in mission-critical electronics while they are in operation or to assess mission readiness before deployment. Previous research in the field has focused on using fuses and canaries to detect impending failure. Prior implementation of data-driven methods frequently disregards failure mechanics. A method for assessing evolving damage in complex systems with nonlinear material behavior has been presented. For various mechanical shock and vibration levels, feature vectors have been identified for predicting remaining useful life. Damage progression was investigated in test vehicles made from various solder interconnects, including SAC105 and SAC305, with the same assembly geometry and architecture. The changes in feature vectors caused by differences in solder material were also investigated. In order to create meaningful data vectors for correlation with the underlying damage progression in pristine and aged assemblies, feature-vector engineering has been pursued. The feature vector selected from the strain signal's time and frequency domain analysis is modeled using the Long Short-term Memory (LSTM) deep learning technique to predict the packages' remaining useful files during the drop. Validation cases for the feature vectors identified in the study have been presented by correlating the plastic work predicted by finite element simulation with that predicted by LSTM. Furthermore, the prediction of remaining useful life has been correlated with experimentally measured time to failure.