Injection Moulding Quality and Energy Optimisation Using ML and NSGA-II

An injection-moulding setter chooses melt temperature, injection time and pressure, holding pressure and time and back pressure largely by experience: too little gives short shots, too much wastes energy and cycle time. Using an open 2026 data set of 1,000 real polypropylene batches (Zenodo, CC BY 4.0), this project predicts part viability (G / Y / R), shot weight and energy per batch from the six settings, compares classical design-of-experiments models (quadratic response surface, multinomial logistic regression) with a Gaussian process, bagged trees and gradient boosting on an honest split (DOE grid batches 1-768 for training, space-filling maximin batches 769-1000 for testing), explains the models with SHAP and searches energy- and time-saving settings at full shot weight with NSGA- II. Unlike the related projects (PR4 laser powder bed fusion uses literature tables, PR3 chatter and A12 tool wear use signals), PR6 is the polymer-processing and multi-objective optimisation project of the group. A FastAPI web application with five interchangeable databases serves predictions, what-if sweeps, NSGA-II runs and a review loop for measured trial batches.

Category: Production Engineering

Production Engineering Project PR06 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.