LAUSR

A wavelet-analysis-based singular-value-decomposition (WASVD) algorithm augmented by the current amplitude normalization (CAN) is proposed in this study to tackle the challenge of weak arc-fault detection. First, an experimental platform is prepared for producing bus-current data from normal operations and weak arc faults. Then, the CAN and the wavelet analysis (WA) are performed on collected signals. Moreover, the coefficients in each layer of the WA are exploited to construct a correspondent Hankel matrix for acquiring the WASVD coefficients by the singular value decomposition (SVD). In order to filter the irrelevant components automatically, the coordinates of AC/DC current components are located by the entropy index calculation on WASVD coefficients. The performances of conventional arc-fault detection algorithms with dependence on manual set-up will be severely degraded if loads, current amplitudes or working environments undergo drastic changes. The proposed algorithm avoids this dilemma by subtracting one arbitrary processed normal signal from all signals of the same system to reduce the noise floor instead of manual adjustment. In addition, signals are reconstructed to have features extracted for the support vector machine (SVM) to make the final diagnostic judgment. The effectiveness of the proposed WASVD-CAN algorithm is verified by various experiments including single load and parallel loads with different operating current amplitudes and an arc generator mounted in series with each load respectively generating series or parallel arc faults. The results show that the proposed algorithm, without manually retuning parameters, has achieved outstanding detection accuracies in comparison with traditional weak arc-fault detection methods.