Using Machine Learning with Supplemented NC Code to Predict Machining Energy

Manufacturing in the United States contributes a significant amount of money to its economy. Simultaneously, it consumes nearly one-third of the total energy produced within the nation. Computer-aided technologies have been developed to aid in streamlining the development of products created within the sector. Despite this, energy-efficiency processes are largely ignored by technologies developed by third-party vendors, within the subtractive manufacturing industry. Such benightedness sparked research into the development of mechanistic and data-driven models that attempt to accurately predict energy consumption. Unfortunately, the models largely fail to reflect the processes for which they are supposedly suitable due to poor experimental validation. The variables monitored within the literature are lacking and neglect to account for the inherent complexity of the processes. A model needs to be developed which properly accounts for the complexities associated with CNC machining. The variables such a model employs must account for variations in operations observed during manufacturing, as supplied via the process. The research conducted herein explores the development of an energy-predicting deep-learning model built upon data from data collected during complex CNC machining. Specifically, the model makes predictions by accepting supplemented numerically controlled programs and processing instructions sequentially, via a recurrent neural network layer. Four variants of the model are created to provide insights into the inclusion of supplementary information. Namely, the comparison of monitored material removal variables. The variables are depth of cut, width of cut, material removal rate, and the volume of material removed per NC instruction.