Neural Network Modelling of Track Profile in Cold Spray Additive Manufacturing

Abstract

Cold spray additive manufacturing is an emerging technology that offers the ability to deposit oxygen-sensitive materials and to manufacture large components in the solid state. For further development of the technology, the geometric control of cold sprayed components is fundamental but not yet fully matured. This study presents a neural network predictive modelling of a single-track profile in cold spray additive manufacturing to address the problem. In contrast to previous studies focusing only on key geometric feature predictions, the neural network model was employed to demonstrate its capability of predicting complete track profiles at both normal and off-normal spray angles, resulting in a mean absolute error of 8.3%. We also compared the track profile modelling results against the previously proposed Gaussian model and showed that the neural network model provided comparable predictive accuracy, even outperforming in the predictions at cold spray profile edges. The results indicate that a neural network modelling approach is well suited to cold spray profile prediction and may be used to improve geometric control during additive manufacturing with an appropriate process planning algorithm.

Dimensional and Geometrical Quality Enhancement in Additively Manufactured Parts: Systematic Framework and A Case Study

“…Others reflect the difficulty of producing parts with similar mechanical behavior to those obtained by traditional manufacturing processes. Finally, dimensional and geometric quality deficiencies of AM parts have also been highlighted as common disadvantages [4,5], which explains their relevance as research subjects during the last decade [6,7,8,9,10]. Quality improvement is a sine qua non condition for the generalized industrial adoption of AM processes, since cost-per-unit reduction would not be enough by itself.…”

“…Some works just compensate deviations from the theoretical values [16]; others aim to compensate mechanical errors in the machine [17]; some works elaborate complex models to compensate the influence of different parameters upon the overall quality [18]. In recent years, there has been a tendency to apply machine-learning methods to provide error correction in AM [10,19,20]. Some works focus on compensating in-plane shape deformation [20,21].…”

“…Other approaches work with the 3D geometry, mainly compensating thermal deformations that have previously been modelled via finite elements (FE) simulation [22] or by means of virtual manufacturing models [23,24]. These types of works build “predictive” mathematical models that could be based on experimental data [10,20] or build from theoretical models [21,23]. Although Artificial Neural Networks (ANN) are frequently used for building predictive models [10,21,22], alternative mathematical modelling is also used for this purpose (e.g., Gaussian process multi-task learning [20] or particle swarm optimization [25].…”

“…These types of works build “predictive” mathematical models that could be based on experimental data [10,20] or build from theoretical models [21,23]. Although Artificial Neural Networks (ANN) are frequently used for building predictive models [10,21,22], alternative mathematical modelling is also used for this purpose (e.g., Gaussian process multi-task learning [20] or particle swarm optimization [25]. The information provided by predictive models could be later used to change input parameters in order to fulfil tolerances.…”

Special Issue: Advances in Thermal Spray Technology

“…These include, for example, evolution of new torch designs, advanced characterization of coatings, novel approaches for in-flight diagnostics and modelling of coating formation. The use of artificial intelligence/machine learning and data-driven modelling approaches, as illustrated in one of the papers [12], is also destined to play an important role in the future as thermal spray expands into new application domains such as additive manufacturing. Perhaps these can be the focus of a subsequent Special Issue.…”