Understanding and predicting electronic vibration stress using ultrasound excitation, thermal profiling, and neural network modeling
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Vibration stress is a major source for failure of electronics components. In this study, we used an ultrasonic/infrared system to generate vertical vibration of a microprocessor on a printed circuit board, in order to study the impact of short burst high load vibration stress on component integrity. The objectives of this research were to (1) understand the impact of vibration on electronic components under ultrasound excitation; (2) model the thermal profile presented under vibration stress; and (3) predict stress level given a thermal profile of an electronic component. Research tasks included: (1) retrofit of current ultrasonic/infrared nondestructive testing system with sensory devices for temperature readings; (2) design of software tool to process images acquired from the ultrasonic/infrared system; (3) developing hypotheses and conducting experiments; and (4) modeling and evaluation of electronic vibration stress levels using a neural network model. Results suggest that (1) an ultrasonic/infrared system can be used to mimic short burst high vibration loads for electronics components; (2) temperature readings for electronic components under vibration stress are consistent and repeatable; (3) as stress load and excitation time increase, temperature differences also increase; (4) components that are subjected to a relatively high prestress load, followed by a normal operating load, have a higher heating rate and lower cooling rate. These findings are based on grayscale changes in images captured during experimentation. Discriminating variables and a neural network model were designed to predict stress levels given temperature and/or grayscale readings. Results suggest a 15.3% error when using grayscale change rate and 12.8% error when using average heating rate within the neural network model. Data were obtained from a high stress point (the corner) of the chip. 2005 Taylor & Francis.
