A novel methodology for predicting energy and time consumption in additive manufacturing powder bed fusion with electron beam system using neural network

Abstract Evaluating the time and energy consumed by the machine during the process is essential for assessing the efficiency and eco-friendliness of the manufacturing method. However, adapting macro models for conventional manufacturing processes to additive manufacturing fails to adequately capture the intricate relationship between energy/time and process design, particularly for complex processes such as powder bed fusion with electron beam melting. This work highlights the limitations of the current modelling approach. It introduces a new methodology that, relying on the beam behaviour at the individual layer level, considers how build design, machine logic, and layer-wise energy consumption influence total energy consumption. Three typical structures, corresponding to three sets of process parameters, are examined: bulk material, overhang support structures, and net/lattice structures. These structures are incorporated into several real components arranged in 25 builds, each with varying nesting degrees, packing densities, build heights, and projections on the build platform. The manufacturing process time for these 25 builds is simulated using software on an Arcam A2X machine, a powder bed fusion with an electron beam system. Data are then analysed using an empirical literature model to highlight the impact of build design and process parameters on total specific energy consumption. Owing to the limitations of the macro empirical model, an artificial neural network based on a multilayer perceptron is employed to develop a model that predicts manufacturing time and energy consumption at the layer level. This model introduces a novel approach to describing the complexity of the build design, making layer-wise energy consumption a complex function of various geometrical features and process parameters. The neural network is trained, tested, and validated using data from six experimental builds, demonstrating its validity and accuracy. This work provides a practical tool for optimising build layouts in EB-PBF. By predicting energy consumption and processing time for different layouts or nesting, the model helps identify the most efficient configuration—minimising the production times and energy demand. Unlike existing models, it incorporates real machine data to accurately capture the impact of complexity, enabling more efficient and sustainable manufacturing.