Application of neural networks in robotized peg-in-hole insertions using a pre-process learning method