Prediction of thermally induced failure for electronic equipment based on an artificial olfactory system

Abstract The failure of electronic equipment causes serious consequences and even catastrophic fires. Abnormal thermal signals are one of the main characteristics of the failure of electronic equipment. Thus, a new method for recognizing and predicting the thermally induced failure states of electronic equipment was proposed, based on an artificial olfactory system (AOS). The AOS recognizes the state of the volatile components released during the early stages of thermally induced failure and uses it to predict the state of health of the electronic equipment. Some typical electronic devices, such as microcomputer units, electronic rectifiers, transformers, and battery modules, were tested with the AOS to recognize the failures indicated by abnormal thermal accumulation. Compared with infrared thermal imagers and gas analyzers, the PEN3 electronic nose was utilized to monitor the status of the devices under different thermal failure scenarios. It was found that infrared thermal imaging was only able to monitor the local surface temperature, and the air temperature in the device chamber changed slowly with the surface temperature of the electronic modules. However, the AOS was able to detect the abnormal change in the whole chamber. Linear discriminant analysis (LDA) and principal component analysis (PCA) were then adopted to investigate the features of thermally induced failure for different thermal states. The results showed that the models obtained both from LDA and PCA were able to distinguish the different states of the electronic devices. Furthermore, a support vector machine model was built, based on the AOS data, to recognize and predict the thermally induced failure processes. All the failure states of the electronic devices caused by thermal simulations were recognized successfully and the prediction accuracy was above 95%. Hence, the experimental results of this research proved that using the AOS, it is feasible to predict the thermally induced failure states of electronic equipment, and the failure of electronic devices can be forecast in advance, before the obvious temperature rise and smoke release. Moreover, the method proposed in this research can also be applied to the prediction of, and warning about, electrical fires, indoor fires, and other thermally induced accidents.