Paper on Federated Learning-based Defect Detection for Additive Manufacturing published in JMS

Our paper titled “Federated learning-based semantic segmentation for pixel-wise defect detection in additive manufacturing” has been published in the Journal of Manufacturing Systems. The paper was co-authored by Manan Mehta and Prof. Chenhui Shao.

Federated learning (FL) is an emerging machine learning (ML) paradigm which allows several participants (manufacturers) to collaboratively train a model while keeping the training data decentralized. FL enables manufacturers to obtain a globally optimized model without uploading their proprietary data to a central database or sharing it with other manufacturers, thus maintaining data privacy without compromising model performance. In this work, we show the application of FL for defect detection in metal additive manufacturing where data availability and data privacy are important bottlenecks for ML at an industrial scale. Through case studies on laser powder bed fusion data, we show that manufacturers can use FL to collaborate with each other and obtain highly accurate pixel-wise defect detection models (with as low as only one image available for training), while also keeping their own data private. This work is among the first to explore the applicability of FL to real-world manufacturing data.

Read the (open access) paper here. The code developed for the paper is available here.