Session: SYMP 5-3: SHM for Extreme Load Applications

Paper Number: 139950

139950 - Online Health Monitoring of Electronic Components Subjected to Repeated High-Energy Shock 

Electronics experiencing high-rate dynamic events, such as shock and vibrations, can lead to adverse effects on the internal microstructures and delicate contacts, ultimately compromising the overall performance of electronic systems. A range of fault detection and mitigation schemes have been developed for high-performance electronics to combat such effects. According to preliminary investigation, potential failures in electronic components resulting from shock and vibration are classified into solder-joint failures, pad cratering, chip cracking, copper trace fracture, and underfill fillet failures. In a high-rate dynamic environment, claims suggest a direct correlation between the duration of shock loading and the resulting damage to electronic components. However, the exact relationship of damage due to shock remains ambiguous.  Further research to investigate the effects of shock duration on electronic systems survivability is required.

This work undertakes a study of the feasibility of online high-rate damage detection in electronics components subjected to impact, where the induced damage is detected in situ during the impact event. Experimentation has indicated that damage to electronic contacts can be quantified through the in situ monitoring of the impedance of the electric connections. Changes in impedance correlate to alterations in the physical properties of electronic components which indicate the occurrence of damage. On this basis, a comprehensive dataset is created to monitor the impedance changes of a daisy-chained connection through repeated high-energy shocks. Meanwhile, the shock response of the electronic components is captured using accelerometers, enabling a detailed analysis of the effects of high-rate shock on the components' performance. A dataset is developed to encompass 30 repeated impacts of average maximum acceleration of 39046 m/s2, and an average half-sine time of 322 µs.

In an effort to properly categorize connection damage during the high-energy shock event, long short-term memory(LSTM) networks are trained using the shock response of the electronics along with their corresponding impedance measurement.   Long short-term memory is a type of recurrent neural network that excels in modeling time-series events. In this work, an LSTM model to estimate an electric component’s health is developed and trained using a supervised learning framework. The network trains by taking in accelerometer data of high-rate dynamic events and categorizes damage of the ball grid chip, into two health states according to an impedance measurement: 1. healthy; 2. damaged. For online health updating, the LSTM model takes the acceleration time series data as input at individual timesteps and produces a health state estimate for each timestep. The training process incentivizes the model to develop a rapid prediction of the chip’s health state and update this prediction as more information is revealed through the signal. The output of the model is the probability of the chip being in one of the two health states. Prior work revealed that such networks achieved a prediction latency of 500 μs showing potential for high-rate applications. Using this type of network, the methodology for a data-driven approach to online health monitoring of electronic components subjected to repeated high-energy shock is developed. 

The contributions of this work are two-fold. First, a dataset of electronic components subjected to repeated high-energy shock is provided. This dataset serves as a valuable resource for researchers and engineers in the field of fault detection and mitigation. Second, a data-driven methodology for training an online state estimating algorithm using LSTM networks is proposed. This methodology enables the prediction of the health state of electronic circuits, enhancing their survivability when exposed to high-rate dynamic events.

Presenting Author: Joud N. Satme University of South Carolina

Presenting Author Biography: Jude Satme is a mechanical engineering graduate research assistant at the University of South Carolina with an undergraduate degree in electrical engineering. Satme's research expertise lies in the field of signal processing and UAV-deployable sensing platforms. His current research focuses on edge computing systems and online machine learning state estimation for high-rate dynamic applications.

Authors:

Joud N. Satme
Daniel Coble
Austin R. J. Downey