DMAIC-IAD multi-agent system improves anomaly detection by 37.76%
Researchers Yongzi Yu, Ao Li, and Le Wang propose DMAIC-IAD, a "Plan First, Judge Later" multi-agent system for industrial anomaly detection that outperforms agentic baselines by 37.76% across four modalities.
Score breakdown
DMAIC-IAD demonstrates that structuring LLM agent workflows around explicit planning and execution-free strategy evaluation yields a 37.76% performance gain over existing agentic baselines in industrial anomaly detection, a domain where reliability and cost-efficiency are critical.
- 01DMAIC-IAD is proposed by Yongzi Yu, Ao Li, and Le Wang, published on ArXiv on 2026-06-03.
- 02The system is inspired by the DMAIC quality-management framework and adopts a 'Plan First, Judge Later' design philosophy.
- 03Existing LLM-based IAD agents are criticized for focusing on execution while under-exploiting strategy formulation.
Yongzi Yu, Ao Li, and Le Wang present DMAIC-IAD (DMAIC-inspired Agentic Industrial Anomaly Detection), a multi-agent framework designed to make LLM-based industrial anomaly detection (IAD) more reliable and cost-effective in high-stakes manufacturing contexts. The paper argues that existing LLM-based IAD agents over-index on execution while neglecting strategy formulation, leaving them ill-equipped to handle heterogeneous data modalities in a unified way.
The system draws its structure from the DMAIC (Define, Measure, Analyze, Improve, Control) quality-management framework, adopting a "Plan First, Judge Later" philosophy.
The system draws its structure from the DMAIC (Define, Measure, Analyze, Improve, Control) quality-management framework, adopting a "Plan First, Judge Later" philosophy. Before any strategy is generated, DMAIC-IAD distills heterogeneous reference data into standardized operating procedures (SOPs). A pre-trained, execution-free judge model then evaluates and ranks candidate strategies, eliminating the need for expensive runtime trials. Extensive experiments across four modalities demonstrate that DMAIC-IAD improves average detection performance over applicable agentic baselines by 37.76%.
Key facts
- 01DMAIC-IAD is proposed by Yongzi Yu, Ao Li, and Le Wang, published on ArXiv on 2026-06-03.
- 02The system is inspired by the DMAIC quality-management framework and adopts a 'Plan First, Judge Later' design philosophy.
- 03Existing LLM-based IAD agents are criticized for focusing on execution while under-exploiting strategy formulation.
- 04DMAIC-IAD distills heterogeneous references into standardized operating procedures (SOPs) before strategy generation.
- 05A pre-trained, execution-free judge model ranks candidate strategies without costly runtime trials.
- 06Experiments span four modalities and show a 37.76% improvement in average detection performance over applicable agentic baselines.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 9, 2026 · 17:05 UTC. How this works →