Explainable AI and Agentic Human-AI Teaming for Autonomous Manufacturing Systems: XAI Methods, AR-Enhanced Operator Collaboration, and Multi-Agent Production Orchestration

Abstract

The deployment of deep learning models in manufacturing systems—spanning process monitoring, quality prediction, predictive maintenance, and autonomous production scheduling—has been constrained by a fundamental barrier: the opacity of modern neural network architectures. As AI systems assume increasingly consequential roles in manufacturing decisions, the inability to explain their predictions and recommendations to human operators, process engineers, and regulatory auditors has become a critical bottleneck to adoption, trust, and compliance. Simultaneously, the frontier of manufacturing AI is shifting from individual models operating in isolation to agentic AI systems—multi-agent architectures capable of autonomous goal-directed reasoning, collaborative decision-making, and continuous self-improvement in dynamic production environments. This review provides a comprehensive and critical synthesis of two interconnected developments at the frontier of manufacturing AI: explainable AI (XAI) methods that render AI predictions interpretable to human stakeholders, and agentic AI and multi-agent systems that enable autonomous, collaborative manufacturing intelligence. We examine XAI methods including SHAP, LIME, counterfactual explanations, and ante-hoc interpretability for manufacturing applications; AR-enhanced human-AI teaming for operator decision support; and multi-agent architectures for autonomous production planning, scheduling, and self-evolving process optimization. We further connect these advances to industrial sensing technologies, demonstrating how explainable and agentic AI integrate with precision metrology, collaborative robotics, and automated testing to create transparent, trustworthy, and autonomous manufacturing ecosystems. A central contribution is the articulation of an integrated Transparent Autonomous Manufacturing (TAM) framework that unifies XAI, human-AI teaming, and multi-agent orchestration for the next generation of manufacturing intelligence.