Validation of a wavelet algorithm implemented in a fixed-point DSP for detection and analysis of electrical transients

In this work we develop an on-line algorithm to detect and classify transients produced in capacitor buses during connection and disconnection. The algorithm was implemented in a Blackfin ® BF-535 Processor. Transients are modeled as an oscillation modulated in amplitude by an exponential decay. The detection process consists in detecting the transient's onset, while the classification consists in measuring the oscillation frequency and decay parameters. Two validations were performed, one against a database of simulated transients and the other with a database of experimental signals. In the last case, as no gold standard is available, we assessed the degree of agreement against an off-line algorithm implemented in a PC. The algorithm performed good in onset detection, where the absolute error was under 1 ms. Also exhibited good performance in frequency classification, where the global relative error was under 1.8%. In contrast, the τ parameter classification relative error only reached an overall range under the 22%. This motivated future improvements in this aspect.