Predicting dimensional accuracy in 3D printed polydimethylsiloxane‐carbon nanotubes composites via machine learning

Abstract

AbstractDirect ink writing (DIW) is a rapidly expanding additive manufacturing (AM) technique valued for its cost‐efficiency and material adaptability. While DIW offers substantial process flexibility, achieving precise manufacturing demands a thorough understanding of its printability parameters. Key process variables such as printing speed, nozzle‐bed gap, pulses for extrusion, and extrusion multiplier wield significant influence over the shape and dimensions of printed components. In this study, a machine learning model based on k‐nearest neighbors is employed to predict dimensional accuracy. The model is trained using measured width and thickness data from printed rings, with flow and printing‐speed tests defining parameter boundaries for printing. The final model demonstrates an accuracy of 81% and 74% for two PDMS‐CNT composite ink compositions. The principal components analysis underscores the pivotal roles of the extrusion factor and printing speed in achieving superior geometric fidelity. Additionally, retraction‐residual tests were conducted, revealing that the initial 10 seconds of extrusion hold critical importance as they can lead to distortions when traversing printed regions or perimeters, necessitating consideration in part design. To evaluate layer adhesion strength, a crucial aspect of AM, T‐peel tests were performed. Furthermore, rheological tests provided insights into extrudability, printable reinforcing percentages, shape retention, and yield stress.Highlights Rheological tests yield insights into minimum reinforcing% and shape retention. Developed k‐NN model predicts accuracy using measured printed rings' data. PC analysis shows key roles of extrusion factor and print speed. Retraction‐residual tests reveal initial 10 seconds are critical. Model validation results show 83.45% and 77.95% accuracy, close to prediction.