NAM-CAM: Neural-Additive Models for Semi-analytic Descriptions of CAM Simulations

Computer-Aided Manufacturing (CAM) is an iterative, time- and resource-intensive process involving high computational costs and domain expertise. The exponentially large CAM configuration space is a major hurdle in speeding up the CAM iteration process. Existing methods fail to capture the complex dependency on CAM parameters. We address this challenge by proposing a new element for the engineer’s design workflow based on an explainable artificial intelligence method. Using Neural-Additive Models (NAMs), we create a semi-analytic model that improves guided search through the configuration space and reduces convergence time to an optimal CAM parameter set. NAMs allow us to visualize individual parameter contributions and trivially compute their sensitivity. We demonstrate the integration of this new element into the CAM design process of a blade-integrated disk (blisk). By visualizing the learned parameter contributions, we successfully leverage NAMs to model the dependency on CAM parameters.