How to find fault detection time for a fault signal using ANN Toolbox

Hello there,

I would like to ask a very particular and detailed way to find the exact fault detection time of a fault signal. What I am doing is gathering a data from the time window of 2 mili-second of the post fault values. Forexample I am generating fault at 0.5 mili-second and capturing a fault window for the 2 mili-second. After gathering data for different fault distance and fault resistances, I am trainning it with the ANN tool box. Now, I have assigned 1 as an response to the fault value and 0 to the healthy signal. I have successfully trained my ANN with the given data. However, I am struggling now to find the fault detection time for any test signal I am giving as an input to my trained ANN. I have so many ways like tic,toc and timeit, but I belive these are not the right ways to detect the time. Please suggest me some ways to deal with my issue. thanks,

Answer 1

[Walter Roberson]

For any given input signal, break it up into 2 millisecond segments, and classify the contents of the segment through the network you developed. If the result is 0, continue on to the next segment. If the result is 1, report back the time of the start of the current 2ms block (start time is (segment number minus 1, times 2 ms)

[zain yousaf]

First of all Sir once again, thanks alot for your answer. But, what if I take DWT of that signal and use it to train my network, this technique still works? What i am doing is taking dB3-level4 decomposition and trainning it with my ANN. Features I am extracted from the dB3-level4 are not in same number number, for example, If I am getting 404 data points for the 2msec window, its decomposition is giving me only 30 segments, how I can use your mentioned technique on my problem?

[ali mustafa]

I am also facing the same problem. I got stuck in this question in a couple of weeks.

[Walter Roberson]

But, what if I take DWT of that signal and use it to train my network, this technique still works?

That sounds possible.

Features I am extracted from the dB3-level4 are not in same number number, for example, If I am getting 404 data points for the 2msec window, its decomposition is giving me only 30 segments

That is not a problem, as long as the same sized input always produces the same size of feature vector. You just concatenate all 30 of the segments into one vector and that is the feature vector for training purposes.