Aligning quality management and data science with AI agents for process optimization in manufacturing: A survey

Process optimization (PO) in manufacturing is guided by ISO 9000 quality management principles and implemented through methodologies such as Six Sigma's DMAIC cycle. Yet the extent to which data science frameworks (DSFs) align with QM standards, and how AI agents can extend data-driven PO, remains insufficiently understood. This systematic mapping study (SMS) examines 96 publications on: (i) how DSFs reflect ISO 9000 principles, (ii) how CRISP-DM compares with DMAIC, (iii) how data-driven methods support PO tasks, and (iv) the functional roles of AI agents in supporting PO. The SMS reveals that none of the evaluated DSFs fully satisfy ISO 9000 requirements, with customer focus and relationship management most consistently unaddressed. A phase-by-phase comparison reveals that CRISP-DM and DMAIC are complementary but structurally distinct: CRISP-DM is data-centric and exploratory, while DMAIC is process-centric and control-oriented. Across publications, data-driven methods yield cycle-time reductions of up to 5.5%, makespan reductions of 8–11%, predictive accuracies above 90% for quality tasks, and fault detection improvements of up to 30% over baseline methods. Five recurring challenges constrain industrial deployment, including data quality, process complexity, and organizational barriers. AI agents operate across three functional roles: decision-making, knowledge-reasoning, and orchestration, extending DSFs toward autonomous analytics. Building on these findings, this paper develops an agentic process optimization framework (APOF) comprising a business governance layer, an AI agent system, a data system, and a tool system. Unlike prior frameworks, APOF connects both process- and data-centric paradigms to support human experts while enabling continuous data integration, automated model development, and operationalization, with auditable decision trails.