A System for Detecting Failed Electronics Using Acoustics

Abstract

Industry-produced printed circuit boards (PCBs) used by the United States Navy and Marine Corps are typically coated with a layer of “conformal coating” made of silicone or polyurethane in order to protect electrical and electronic components on the board. Conformal coating has to be removed every time board troubleshooting and maintenance are performed, and must be reapplied after board maintenance is complete. This can be an expensive and time-consuming process. This paper describes an effort to develop a non-contact solution to detect failed components on a PCB without having first to remove the conformal coating. This patent-pending technique detects density changes in the physical makeup of circuit board components due to failure. By analyzing ultrasonic reflections from the components at 2MHz, the authors were able to distinguish between working components and failed components with varying degrees of accuracy. The authors applied this technique to $1K\Omega$ resistors and three types of transistor-to-transistor logic (TTL) integrated circuits (ICs). Overvoltage faults were induced in these components in order to generate observable density changes. To reduce human error, a measurement rig was built which incorporated an automated X-Y-Z plotter system, in order to process dozens of components at a time without human interaction. The data gathered by this system was processed to isolate only the acoustic reflections of components on a circuit board. Time-domain and frequency-domain features were then extracted. These features were used to train neural networks to distinguish between working components and components with over-voltage faults that were not readily observable by eye. Each type of component or chip needed to have its own associated trained neural network. For $1K\Omega$ resistors, the system has demonstrated seventy to eighty percent accuracy in distinguishing components with over-voltage faults. For two of the TTL ICs, eighty to eighty-five percent accuracy has been achieved. For one IC type, a fifty-five percent accuracy was measured. The authors have demonstrated that low-cost acoustic measurements in the megahertz range can be used to detect failures in ICs and other common circuit board components.