Generalisable Predictive Maintenance in Additive Manufacturing Using Machine Learning

D printing, also known as Additive Manufacturing (AM), translates simulation-built models into physical objects. However, the inherent vibration of 3D printers makes it challenging to distinguish between abnormal vibrations, thereby hindering predictive maintenance and significantly affecting product quality. Delays in predictive maintenance can also lead to component failures in the printer, resulting in a prolonged production suspension with financial consequences. To address the above-mentioned problem, this paper performs an anomaly detection study on an Upgraded Creality Ender 3 3D printer using an MPU-6050 motion-tracking sensor attached to the printer bed. The devised framework uses the Adaptive Synthetic algorithm to balance the dataset, aiding the support vector machine algorithm to attain a perfect score of 1 in recall metric and an accuracy score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 2 5 5}$</tex>. Additionally, the paper uses the Shapley-Additive Explanation technique to provide insight into the most important features used for classification. Afterward, a subsequent deployment analysis was carried out for five different deployment boards, quantifying the deployability of the anomaly detection framework. The Alif Ensemble microcontroller provides inference within 1 ms with features like low power and cost, making it suitable for deployment in manufacturing settings to achieve real-time anomaly detection in 3D printing, paving the way for accurate predictive maintenance in AM.