A Fuzzy Logic-Based Framework for Statistical Process Control in Additive Manufacturing

Additive Manufacturing (AM) research as well as applications are rapidly growing primarily due to the inherent capability to manufacture very complex parts. This is particularly important for metal parts, for example for the aerospace, automotive and biomedical sectors. However, it is still challenging to develop reliable monitoring frameworks that guarantee process quality and stability regardless of part geometry. This is due to the complexity of the process which is based on material-beam interactions followed by layer-based deposition of material. Advanced imaging techniques can provide in-situ information for quality assurance purposes. In this study, we enhance a previously developed process monitoring framework based on Statistical Process Control with Fuzzy Logic-based modelling to calibrate the non-linear relationship between normal process behaviour and part defects. In recent work, it was shown that the monitoring system can be very effective in the identification of defects via the use of thermal imaging and multilinear principal component analysis (MPCA) to characterise process performance with a T-2 metric resulting from statistical process control (SPC). While the model performs sufficiently well in preliminary results it is important to also test for generalisation, hence in this new study we extend the results to assess more parts. We also extend the computational framework to cope with the further complexity in modelling the relationship between monitoring data outliers and defects. This is achieved via the use of 1-)fuzzy c-means (FCM) clustering for the feature clustering which helps to group the similar thermal image groups to improve capability to capture multiple process behaviors in the same modelling framework, 2-) adaptive neuro-fuzzy inference system (ANFIS) to model the expected non-linear relationship between the SPC T-2 metric and actual part defects. A case study in blown-powder laser melting deposition (LMD) of complex geometry is presented. A clear correlation is observed between the predicted outliers and the measured part defects; this is shown for multiple parts.