Part quality investigation in fused deposition modelling using machine learning classifiers

In fused deposition modeling process the staircase and beading effect results in poor surface roughness and wrapping results in less dimensional accuracy. In the present study, dimensional deviation and surface roughness are investigated in relation to fused deposition modeling process parameters. The process parameters considered are layer height, wall thickness, infill pattern, infill density, print speed, bed temperature, nozzle temperature, and fan speed. The responses considered are length-wise deviation, breadth-wise deviation, height-wise deviation, surface roughness in the vertical direction, and surface roughness in the horizontal direction. Results were analyzed by machine learning classifier models namely, logistic regression, Gaussian Naïve Bayes (GNB), decision tree, and Support Vector Machines (SVM). Model adequacy has been checked in terms of accuracy, misclassification rate, true positive rate, false positive rate, true negative rate, and precision. Based on the analysis of results, layer height, print speed, and nozzle temperature are the major process parameters that affect dimensional deviation and surface roughness of parts. Surface roughness in the vertical direction is minimum i.e., along the print direction, and surface roughness in the horizontal direction is maximum due to the staircase effect. The dimensional deviation is low at minimum layer height and print speed along with moderate nozzle temperature. Based on the model adequacy check parameters GNB is the most suitable model for length-wise deviation and height-wise deviation. SVMs are best for breadth-wise deviations. For surface roughness in the vertical and horizontal directions, the decision tree & SVM evolved the most suitable model.