摘要
Aging & Test is a critical step in TFT‑LCD manufacturing for early‑failure defect identification, yet capacity and cost constraints make full inspection infeasible, necessitating sampling. This study proposes an Agentic AI–driven dynamic sampling optimization that replaces passive probabilistic sampling with an autonomous, closed‑loop workflow. The agent integrates multi‑stage production history (Array/CF/Cell/Module), inspection results, and repair logs; plans and prioritizes panels under capacity and external cost constraints; executes Aging & Test scheduling; and continuously updates parameters from live outcomes via feedback and memory. Operators specify sampling objectives and constraints via prompts, while stage‑specific agents coordinate sampling and dispatch to accurately target high‑risk panels and adaptively reallocate inspection resources to demand and process variability. Empirically, sampling approximately 20 % of panels was associated with about 0.80 % defect detection in our observational evaluation. Cost analyses were associated with an overall reduction of approximately 30 % across testing levels by optimizing product‑level sampling ratios, and up to 56 % versus random sampling. Overall, the deployment was associated with improved defect capture and more efficient resource allocation under process variability; however, these results are observational and may be influenced by other concurrent factors.