Feature Visualization within an Automated Design Assessment leveraging Explainable Artificial Intelligence Methods

AI systems used to assess manufacturing designs can now be made transparent: The paper demonstrates that explainable AI (xAI) methods can reveal which specific geometric features in 3D designs influence an AI system's decisions, moving beyond "black box" assessments where the reasoning remains hidden from users.

Multiple explanation techniques provide complementary insights: The researchers tested four different xAI methods (sensitivity analysis, layer-wise relevance propagation, Grad-CAM, and LIME) to identify geometric patterns, suggesting that using multiple approaches together may provide more reliable understanding of AI decision-making in design assessment.

Geometric visualization makes AI decisions actionable for engineers: By translating abstract AI assessments (like "this design has gripping surfaces") into visual indicators of which parts of the design the AI is actually analyzing, engineers can verify whether the system is making decisions for the right reasons—critical for safety-critical manufacturing applications.

This approach helps catch when AI systems learn the "wrong lesson": The paper addresses a key responsible AI concern: an AI system might achieve high accuracy while focusing on irrelevant features. By visualizing what the network learned, designers can identify and correct these flawed patterns before deployment.

Explainability enables human oversight of automated decision-making: As manufacturing increasingly automates the automation process itself, making these systems interpretable allows human experts to maintain meaningful control and catch errors that could otherwise propagate through automated workflows.