Nondestructive evaluation of material parameters using neural networks

The general aim is the characterization of microstructural states and the determination of mechanical and technological parameters for 7 different steel grads used in nuclear power plants. In order to reach this goal, toughness sensitive physical measuring quantities must be used which record changes in micro stresses of second and third order as well as changes in the precipitation states. Micromangetic measuring parameters obtained from Barkhausen noies, incremental permeability and longitudinal magnetostriction were investigated. To determine tensile strength, yield strength, hardness and other parameters, it is necessary to combine several independent nondestructive measuring quantities in a multiparameter evaluation approach. This approach approximates an inverse function modelling the relation between parameters. It was realized using neural networks of the backpropagation type. The analysis was successful in separating different microstructure states and proved at the same time its generality for differnt material parameters. The best results were achieved using a very broad approach of up 14 different nondestructive parameters. Neural ents proved to be capable of supporting the optimization of NDE-techniques by objectively evaluating the significance of nondestructive parameters, both as single or as combined parameters. Future work will be directed towards using self organizating maps as nerual nets in order to get more general results using these methods