Sensor Based ANN System for Monitoring Die Fill in Forging

Abstract The North American forging industry has started to implement robotic automation due to ongoing employee shortages. However, most installations are based on Industry 3.0 technology and require human intervention to perform advanced functions related to process control and defect detection. However, Artificial Neural Networks (ANNs) have made it possible to develop intelligent control systems for forging processes. In the present paper, a 6x2x2x1 ANN model was developed to predict when fill defects would occur in a forged aluminum part using 5 discrete inputs. An ablation study showed that preform geometry, furnace temperature, and die temperature had the most effect on classification accuracy, while maximum press load was found to have a marginal effect. Physical trials showed that the ANN model was 95% accurate in differentiating between fill and no fill occurrences using virtual and physical process data but dropped to 75% when simulated training data was used to classify physical parts. Though further investigation is warranted to understand the accuracy difference between simulated and physical process conditions, ANNs represent a useful tool that could be coupled with appropriate sensors to develop intelligent forging automation.