Machine learning as a reliable tool for compressive strength prediction in pa6-cf

Machine learning plays a crucial role in recent years in manufacturing systems. Beginning with optimisation to predicting machine failures, machine learning has been widely exploited by the manufacturing sector. However, determining the mechanical properties of fabricated parts is crucial for manufacturing sectors to apply the material in the right areas. Determining the mechanical properties of materials includes costly laboratory setups and a time-consuming process. Hence, there is always a need to look for alternative methods to determine the mechanical properties. In this investigation, Machine learning models, including Linear regression, Decision Tree, Support Vector Regression, AdaBoost, XGBoost and CatBoost, were analysed to predict the compressive strength of additively fabricated specimens. The fused filament fabrication technique has been selected as the additive manufacturing method for fabricating the test samples. Carbon fibre-reinforced nylon filament, widely used for industrial applications, is selected as the raw material to fabricate the specimens. Compression specimens were prepared as per ASTM standards, and an experimental investigation was carried out to obtain the experimental values. The obtained values were split into training and testing data and were used with machine learning models. The models were evaluated based on MSE and R-squared evaluation metrics, and the best-fit model was determined. Optimisation techniques, including Optuna and skOpt, were included for hyperparameter tuning. Results proved that ML models are effective in predicting compressive strength with minimal variations. CatBoost and XGBoost were the most suitable models for predicting the compressive strength of fused-fabricated carbon fibre-reinforced nylon specimens.