You should think about which features represent your data. For example, for voltage and current you may use the mean value, the maximum and the minimum deviations, and maybe the frequency of oscillation. These features concentrate the information of the signal into few values that form a pattern. Then, you link every pattern with is proper classification and train the network with both patterns and classifications.

Good luck!

Article Improved Shape Parameter Estimation in Pareto Distributed Cl...

current and voltage are time series waveforms , so you may want to look at recurrent neural networks, not just a simple feed forward NN

@ José Raúl Machado Fernández ....I have current and voltage waveforms. Now i want to train these waveforms using ANN,s to check for Fault or No Fault situation.

@Vikas Ramachandra:

Thank you for such a useful answer.

I already planned to use Muti layer Feed forward ANN,s with back-propagation or recurrent Neural networks. But recurrent Neural  may sometimes be prone to instability.

In MATLAB,I am thinking of using NARX net in NNTOOL.

But Problem is how can we prepare training data for feeding these ANN,s.

My data file is attached.Having 6 inputs and 1 output(Fault or no fault)

Apply the real waveform in MATLAB ( Similar to sound recording). then create wave file with appropriate Sampling freq..then you can use the file for any tools 

Initializing ADC, after zero crossing, with taking samples at fixed interval and storing in sequential memory location, could generate required data string of the waveform....I did it for online monitoring, using 8085 module, refer my papers.....,could be regenerated, by out outing the string via DAC.......