A novel strategy for fault diagnosis of analog circuit online based modified kernel fuzzy C-means

Focusing on the issue of analog circuit performance online evaluation, the arithmetic speed and the evaluation reliability should be considered simultaneously. A novel online faults diagnosis strategy based on modified kernel fuzzy C-means (β-MKFCM) is proposed based unsupervised learning algorithms of analog circuit faults diagnosis for the known faults and unknown faults online. More specially, the kernel fuzzy C-means itself can decrease the train samples and eliminate wild values, in this way the training speed and precision of classifier can be improved. In addition, one of the crucial points of the fault diagnosis is to confirm exact class center from the data of known faults. Then, depending on the fault data of each class to obtain the mean value, meanwhile, setting this mean value as the threshold for judging fault and then each data point issued with a class label. During the whole data processing, each data will be compared with the threshold, the high similarity data fall into the known fault class, and while the low similarity data is labeled as unknown fault. Experiment takes the Sallen Key low-pass filter as the diagnosis circuit to prove the effectiveness of the β-MKFCM algorithm. For proving the validity, another RBF fault diagnosis method is employed here. Numerical simulations reveal that the proposed method β-MKFCM has the higher recognition capability than the RBF method for the known fault and unknown fault. Meanwhile, the fault diagnosis speed and precision of the β-MKFCM are all superior to that of the traditional supervised mechanism, precision.