This paper presents a novel quality monitoring method for additive manufactured surfaces combining machine learning and light scattering. The proposed method aims to monitor undesired topographical modifications of additive manufactured surfaces by detecting changes in a scattering pattern using an autoencoder, which is an unsupervised machine learning model, trained with datasets directly measured from reference surfaces with desired surface topographies. Given the unsupervised learning nature of the autoencoder, training with datasets acquired from surfaces with deviations is not necessary, which makes the proposed method appealing, as there is no need to retrieve defective surface samples to train the autoencoder. More importantly, the autoencoder can be updated when datasets from a new type of surface with desired but different topographies are available. As scattering patterns related to new topographies are relatively easy to obtain by experiment, we demonstrate that our autoencoder can be retrained with new scattering patterns and learn to address a wider variety of surfaces, showing superior performance with respect to machine learning solutions adopting a static model, trained only once on the initially available information. Experiments performed on laser powder bed fusion surfaces show that the proposed method is effective. The relatively simple and low-cost setup of the measurement system also makes the proposed method appealing for implementation on commercial additive manufacturing machines.

Intelligent quality monitoring for additive manufactured surfaces by machine learning and light scattering

Senin N.;
2021

Abstract

This paper presents a novel quality monitoring method for additive manufactured surfaces combining machine learning and light scattering. The proposed method aims to monitor undesired topographical modifications of additive manufactured surfaces by detecting changes in a scattering pattern using an autoencoder, which is an unsupervised machine learning model, trained with datasets directly measured from reference surfaces with desired surface topographies. Given the unsupervised learning nature of the autoencoder, training with datasets acquired from surfaces with deviations is not necessary, which makes the proposed method appealing, as there is no need to retrieve defective surface samples to train the autoencoder. More importantly, the autoencoder can be updated when datasets from a new type of surface with desired but different topographies are available. As scattering patterns related to new topographies are relatively easy to obtain by experiment, we demonstrate that our autoencoder can be retrained with new scattering patterns and learn to address a wider variety of surfaces, showing superior performance with respect to machine learning solutions adopting a static model, trained only once on the initially available information. Experiments performed on laser powder bed fusion surfaces show that the proposed method is effective. The relatively simple and low-cost setup of the measurement system also makes the proposed method appealing for implementation on commercial additive manufacturing machines.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11391/1553453
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? 4
social impact