Explainable AI for Correct Root Cause Analysis of Product Quality in Injection Moulding

Different explainability methods can lead to wrong conclusions: The study demonstrates that various AI explanation techniques produce different results when analyzing the same manufacturing problem, potentially leading technicians to identify incorrect root causes for product defects. This highlights the importance of carefully selecting explanation methods rather than assuming all approaches are equally reliable.

Explainability matters for real-world adoption: While the machine learning models achieved high accuracy (less than 0.05% error), the researchers found that black-box models without explanations are impractical for quality control. Explainable AI enables factory workers to understand why a product failed and take corrective action, making the technology actually usable in production environments.

Complex interactions between machine settings require sophisticated explanation methods: The research found that injection moulding machine settings don't work in isolation—they interact with each other in ways that affect product quality. Simple explanation approaches may miss these interactions, while better feature attribution methods can reveal the true causes of quality problems.

Model-agnostic explanations provide flexibility: Rather than being limited to specific types of AI models, the study shows that explanation methods can work across different algorithms (random forests and neural networks). This flexibility allows manufacturers to choose the best-performing model without sacrificing interpretability.