Predictive Modeling of Compaction Density in Powder Metallurgy Components

This paper presents preliminary results of neural modeling of compaction density variations in powder metallurgy components. The model will be used for calibrating finite element models of compacting complex shapes. This will reduce the uncertainty associated with modeling parameters and speed up the modeling effort by reducing the number of experimentally determined characteristics. Experimental measurements of local density have been conducted using SEM and the results analyzed to identify the degree of complexity required in the neural model. The training effort is also monitored to ensure the neural model is capable of generalizing and is not overtrained. Results show that neural modeling is a feasible approach to predicting compaction density variations. Published in the Transactions of NAMRI/SME, Vol. 38, 2010, pp. 633-638.