An ensemble stacking machine learning model for temperature prediction of extruder nozzle in an additive manufacturing process

Abstract

Additive Manufacturing (AM) technique is preferred over subtractive manufacturing because it creates complex geometric structures with minimal material waste. Even though it has many advantages, a few drawbacks limit its application in the industrial environment. One of the main disadvantages of the process is the extruder nozzle temperature fluctuation which causes nozzle clogging, overheating and failure. Sensorized AM machine control is the most adapted process monitoring strategy to avoid these issues. However, manually managing the process variables with sensorized control becomes complex due to a substantial amount of data and variables in AM. Machine Learning (ML) is a suitable approach for analyzing massive amounts of sensor data with many attributes. This paper considers a fused filament fabrication process and its extruder nozzle temperature prediction. We developed ensemble stacking regressor models using sensor data to predict the extruder nozzle temperature. As a precursor of the study, we applied ML models such as Naive Bayesian, Random Forest, XGBoost, CatBoost, Artificial Neural Network, and Support Vector Machine as stand-alone models. The stand-alone models with better performance measure values based on Root Mean Squared Error (RMSE) and R squared (R2) are considered for developing the ensemble stacking models. The best-performing ensemble stacking model employs XGBoost, CatBoost and Random Forest in layer zero and XGBoost in the meta-model layer. It has the highest R2 value (0.94), indicating the best fit than the stand-alone models.