I am working on ANN toolbox, but i m not getting idea as in what will be the input and what will be the target value???

I am working on like to detect the vibration signal with help of wavelet and to classify whether it is normal,outer race fault or inner race fault with help of ANN. In such case after applying wavelet I am extracting the features like energy,kurtosis,skewness,crest factor and RMS value which will be acting as input to my ANN. but how to exactly apply those input am not getting.

[aditi patil]

Thanks for your reply.

According top above text I have applied input[0 0.5 1;0 0.11 1;0 0.5 1]and target as [1 0 0;0 1 0;0 0 1], but now the further step is if i want to train the network then next what that tool box ask u that whether you want to restore the default setting like tarinlm,learngdm,mse,nuber of layers 2,transfer function tansig. If I say yes it trains the network but what is the internal calculation is been done on my input i m not getting that. Can anyone show me what excat calculation are done with help of that TRAINLM training function and LEARNGDM adaptation learning function.

[Walter Roberson]

http://www.mathworks.com/help/nnet/ref/trainlm.html#zmw57dd0e40856

http://www.mathworks.com/help/nnet/ref/trainlm.html#moreabout

[aditi patil]

Thanks for the reply sir. I am still not getting on what basis we should select the training function and learning function. Can someone explain me with some application or some example so that I can apply it my application.

Answer 1

[Greg Heath]

If you have N input/target pairs for I-dimensional inputs to be classified into one of c-classes characterized by class indices 1:c, the input and target matrices have sizes

 [ I N ] = size(input) 

 [ c N ] = size(target)

where target columns are c-dimensional {0,1} unit vectors from the unit matrix eye(c).

The relationship between the true class indices and the target matrix are

 target           = ind2vec(trueclassindices) 
 trueclassindices = vec2ind(target)

Similarly, the relationship between the estimated class indices and the classifier output are

 estimatedindices = vec2ind(output);

Then correct classifications and errors are obtained from

 errors = (estimatedindices ~= classindices);

 Nerrors = sum(errors)

 Pcterr = 100*Nerrors/N

Class and trn/val/tst division breakdowns can be obtained by knowing the datadivision indices obtained from the training record tr

 [ net tr output ] = train(net, input, target);

You can probably find several examples by searching the NEWSGROUP and ANSWERS using

 greg patternnet

Hope this helps.

Thank you for formally accepting my answer

Greg

[aditi patil]

Thanks for replying sir.

Sir can you guide me when to use which algorithm and which adaptation learning rule????

[aditi patil]

Suppose sir if I use backpropagation in which i use delta learning rule then in such case i need to set the threshold value to classify my output. So in such case is there any rule or some formula to set that threshold value. Or according to my information the input my applying which is crest factor,skewness...I have normalized those value and after that from that normalized data I have set my threshold value according to them.Is this correct method??

[Greg Heath]

I have already told you to search for my PATTERNNET posts.

But first, see the example in the documentation:

 help patternnet

 doc patternnet

Write back if you have anymore questions regarding the above.

Hope this helps.

Greg

Answer 2

[Walter Roberson]

Create a 2D array, with one row per sample, and with the columns being energy, kurtosis, skewness, crest factor, and RMS value. Call this "input"

Create a 1D column vector, ind, one row per sample, which is 1 for samples that are normal, 2 for samples that are outer race fault, 3 for samples that are inner race fault.

Call <http://www.mathworks.com/help/nnet/ref/ind2vec.html ind2vec()) passing in that column vector, and saving the result as the variable "target".

Now pass the variables input and target to your neural network routine.

[Greg Heath]

Incorrect.

Those matrices will have to be transformed to a 5 X N input and a 3 x N target.

Answer 3

[Sandeep Nayak]

Dear guys,

I want to know how to write code in matlab function in order to detect the fault using wavelet transform. I have four output signal data. But after fault are varying randomly but at the time of fault the respective signal have some discontinuity which others havenot. How to detect that particular instant in matlab code. I used wenergy command but that didnot work in simulink. The error was like this,"can`t generate the code". Please guide me. It is very essential