XAI-driven digital twin for cobot dynamic error compensation

Process and product fingerprints (FP) have been approved as effective parameters to reveal the principal contributing factors towards functionality in smart manufacturing processes. Though AI-driven methods outperform other approaches for fingerprint extraction, the lack of explainability in its black-box style predictions leads to misconceptions and trust issues among stakeholders. In this study, a novel explainable-AI (XAI) approach is proposed to identify mathematical fingerprint expressions by formulating them as graphs using the QLattice algorithm, inspired by path integral formulation. Here, the Qlattice model identifies explainable and human-comprehensible FP expressions for cobot dynamic error based on accelerometer signal features. The discovered symbolic model is subsequently applied to a digital twin which successfully tracked and compensated for dynamic errors autonomously in real time.