Using a Neural Network for Predicting the Value of Retained Austenite in Ni-Hard4 Cast Iron

The paper presents a neural-network-based inverse mapping technique to predict the value of retained austenite in Ni-Hard4 cast iron (NiHCI) from the output signal of an absolute eddy-current probe. The network utilizes the back propagation method and the normalized resistive and inductive components of the probe impedance are used for training. The corresponding database is established by producing various sample tests. This is done by employing different destabilizing heat treatments to produce various microstructures in several reference NiHCI blocks. The actual values of retained austenite in each case is obtained by an optical microscope together with a commercial image analysis software. The validity of the proposed technique is demonstrated by comparing the actual and predicted measured values of retained austenite in a number of NiHCI blocks that are not used in the training stage and noisy ones. It is also shown that the proposed technique is more accurate than the previously reported method based on regression analysis of the normalized values of probe impedance and the actual values of retained austenite of reference blocks.

کلیدواژه ها:

Eddy current ، Neural Networks ، Microstructural characterization ، Retained austenite ، Ni-Hard4 cast iron