One-class classifier based on SBT for analog circuit fault diagnosis

This paper presents an analog circuit fault diagnosis framework, called OCB-SBT (one-class classifier based SBT). In OCB-SBT, one-class classifiers as many as predefined fault classes are constructed from their corresponding fault simulation samples. And they serve the role of fault dictionary and classifier. This approach is simple for Go/NoGo testing because it requires only no-fault circuit simulation samples to provide the result. It will also provide “unknown fault” result if the testing object lies outside the boundary of training data, and bring “or” combined result when the testing sample belongs to the overlapped area among different fault classes. Fault diagnosis example shows that the proposed method is robust and reliable to the effects due to not only component tolerances and nonlinearity but also poorly defined fault classes and low testability.