Multi-source data fusion technique for parametric fault diagnosis in analog circuits.

Input test signal plays important role in testing of analog circuits. Single type of input stimulus cannot maximally reveal the state of the circuit. To combat this shortcoming, this work proposes to integrate information from the output responses corresponding to different input stimuli and to use the combined information to improve the accuracy of fault diagnosis in analog circuits. The circuit under test is excited with different input signals and wavelet features are extracted from the output responses of the circuit. Ultimate fault features have been defined by applying data fusion algorithm to the sets of wavelet features obtained from individual output. Data fusion has been performed in two steps, data whitening and Principal component analysis (PCA). Fused features are used to train SVM (Support Vector Machine) classifier for fault diagnosis. The proposed approach is validated with three types of filter circuits, i.e. Sallen-Key band pass filter, four OPAMP high pass filter and elliptic low pass filter. The average accuracy of the proposed fault classification method has been found greater than 99.8% for all the test circuits. The proposed technique offers improved classification accuracy, lower computational burden and lesser implementation complexity. • This work proposes to integrate information from the output responses corresponding to different input stimuli to improve the accuracy of fault diagnosis. • Wavelet energy vectors have been used as pre-processor. • Fault features have been constructed by applying multi-source data fusion algorithm. • Data fusion has been performed in two steps, data whitening and Principal component analysis (PCA). • The fused features are used to train SVM (Support Vector Machine) classifier for fault diagnosis. • The proposed technique offers improved classification accuracy, lower computational burden and lesser implementation complexity. [ABSTRACT FROM AUTHOR]