Translating Machine Learning Into Manufacturing Expertise—Actionable Interpretability and Stability for Precision Injection Molding

Industrial adoption of artificial intelligence/machine learning (AI/ML) in injection molding (IM) continues to grow, alongside efforts in trustworthiness and domain knowledge alignment. This research aims to promote explainable AI (XAI) applications with IM domain knowledge. An augmented XAI human‐machine perception language structure in IM was proposed. Serving as a domain‐centric approach, two major IM challenges were addressed: building confidence in AI/ML model deployment and machine fine‐tuning to deal with unpredicted defects. Based on IM simulations, six machine parameters were selected to predict four optical qualities using artificial neural networks (ANNs), which were also utilized for model explainability tasks. To promote trust, an integrated stability analysis was employed to evaluate both the median values and variation of and SHAP contributions across trials and data usage. Results revealed similarity in final saturation but discrepancies in stability and saturation speeds. To provide help with machine fine‐tuning, parallel coordinate plots were proposed to visualize perturbation scenarios that incorporated machine control tolerance and user‐specified fine‐tuning ranges. The interactive selection of multiple machine parameters‐of‐interest allowed engineers to concentrate on potential quality shift from fine‐tuning. The proposed methodologies enhance model explainability across diverse manufacturing fields, potentially facilitating broader AI/ML adoption in various industrial applications.