Process Parameters Prediction of Hybrid Manufacturing using Machine Learning

Abstract

In manufacturing industries, hybrid manufacturing (HM) emerged as a paradigm shift since it provides numerous benefits which include lead time reduction, cost savings and waste minimization. In this thesis, the concept of HM as an integration of Fused Deposition Modelling (FDM)-an Additive manufacturing (AM) process, and milling being a traditional process was utilized. To maximize the benefits of HM, it is essential to predict the behaviour of process parameters effectively. This thesis proposes the framework of predictive analysis in HM. Firstly the experiments were conducted by considering AM parameters: layer thickness, print speed, and extrusion temperature. Three levels of each parameter were considered. Surface roughness was measured. Then milling was carried out with two levels of feed rate and again surface roughness was measured to observe the effects of milling. After this, compressive testing was performed. Then experimental data was fed to Microsoft Azure Machine Learning Studio to get numerical results. In this software, neural network regression was performed to explore the behaviour of process parameters. Hyperparameter tuning was also done to increase the efficiency of the model. The actual, predicted, and hyper scored outcomes were compared. It was observed that experimental and numerical results agreed. It was concluded that for HM, to improve the surface finish, layer speed and extrusion temperature need to be kept low while print speed needs to be increased. For compressive strength, layer thickness should be kept maximum while print speed and extrusion temperature should be kept low.