Process stability and quality surveillance in the metas injection moulding process - towards zero rejection based on an artificial neuronal network

In the present research, it was established that the weight properties of an already sintered MIM part which are influenced by the injection moulding machine parameters, could be predicted by adapting a back propagation artificial neuronal network. With a statistical evaluation of experimental data gathered from 68 process parameters of each prepared samples and the related final weight of the parts, an artificial neuronal network can be trained to generate a mathematical model for quality prediction. The correlation analysis revealed that the proposed model almost consider strong complex interactions between the single process variables and the weight of the moulded parts. This generated model permits an online quality control of metal injection moulded parts and the detection of any variations in the process conditions. Further investigation are necessary to expand the prediction to multiple part quality properties. The aim is to use this tool of an artificial neuronal network at the MIM process on the earlier possible stage in the process chain to achieve an appropriate on-line control during the injection moulding process. Detected failed green parts can be recycled and do not have to underlay the long time and energy consuming debinding and sintering process.