I need to identify a system which is open-loop stable. Now if I want to identify it using neural network,would I need to make the system close-loop stable for collecting the input-output data points ?
If your process is open-loop stable, and you want to model it with a neural network, just collect the data in open loop, of course. Closed-loop stability is not an issue. Then you will have to make various choices for training, depending on the purpose of the model (one-step-ahead predictor, or simulator?) and also depending on the noise in your data (state noise or output noise?).
For a complete design methodology see
Neural networks, methodology and applications, chapter 2 (Springer, 2005)
For didactic examples see
O. Nerrand, D. Urbani, P. Roussel-Ragot, L. Personnaz, G. Dreyfus
Training Recurrent Neural Networks : Why and How ? An Illustration in Process Modeling
IEEE Transactions on Neural Networks 5, 178-184 (1994)