Title: Neural network as approach for detection of non-compliant semi-finished additive manufactured parts

Authors: Gianluca D'Urso; Mariangela Quarto

Addresses: Department of Management, Information and Production Engineering, University of Bergamo, Via Pasubio7/b, 24044, Dalmine (Bergamo), Italy ' Department of Management, Information and Production Engineering, University of Bergamo, Via Pasubio7/b, 24044, Dalmine (Bergamo), Italy

Abstract: The integration of artificial intelligence and algorithms for the elaboration of data with the new advanced technologies represents an interesting topic for the definition of reliability and repeatability of the process. The present work combines these elements for improving the process chain of metal-material extrusion (metal-MEX), developing an algorithm able to predict which parts will be non-compliant after the debinding and sintering processes necessary for obtaining the final parts. Specifically, considering a database containing historical data collected by the authors in their previous research, an artificial neural network (ANN) was trained to be applied immediately after the printing stage for detecting non-compliant parts. In this way, it is possible to avoid the post-printing treatment (debinding and sintering) for parts that do not respect the design requirements. Considering the validation of the model, the ANN can discriminate which parts will be able to satisfy the requirements, supporting the operator in the selection of compliant and non-compliant parts.

Keywords: geometrical accuracy; prediction; additive manufacturing; ANN; artificial neural network; quality assessment; non-compliant detection.

DOI: 10.1504/IJMMS.2023.133396

International Journal of Mechatronics and Manufacturing Systems, 2023 Vol.16 No.2/3, pp.261 - 279

Received: 09 Feb 2023
Accepted: 17 May 2023
Published online: 14 Sep 2023 *