ACG Deploys AI Closed-Loop Systems to Lift First-Pass Yield and Cut Defect PPM

Persistent setup variability and knowledge-transfer risk on the shop floor are problems ACG has moved to address structurally, deploying two AI-driven closed-loop systems across its capsule and packaging materials operations. The practical results, improved first-pass yield, reduced defect PPM, and shorter setup cycles, carry direct implications for plant heads evaluating smart manufacturing investments under ICH Q10 continual improvement frameworks.

The first system, Golden Batch, targets first-time-right production in pharmaceutical capsule manufacturing. Machine setup across ACG's capsule lines historically depended on operator expertise and SOPs, a method that struggled against the combinatorial complexity of thousands of colour variants, machines ranging from new to over twenty years old, and more than 60 critical settings per line running at several thousand capsules per minute. Golden Batch resolves this by prescribing optimal values for all 60-plus parameters using an ensemble of machine learning models trained on three years of production, quality, and machine-condition data. Confirmed settings transfer directly to the machine PLC via Industrial IoT and edge computing, closing the loop from model output to machine-level execution without manual transcription. A continuous learning pipeline incorporates production outcomes and operator feedback to refine model accuracy over time.

The second system, Chanakya, addresses a different but related risk: the concentration of operational knowledge in long-serving experts approaching retirement, compounded by decision-relevant data scattered across order management, planning, procurement, production, quality, logistics, maintenance, and finance systems. With more than 5,000 SKUs, micron-level substrate tolerances, and over 30 percent rush orders in its packaging materials business, ACG needed cross-functional insight delivered at shop-floor speed. Chanakya orchestrates 15-plus generative AI assistants aligned to specific functions, returning a unified response that includes a direct answer, supporting analysis, visual substantiation, and recommended next steps. For supervisors, this means actionable guidance on recurring stoppages, material constraints, and maintenance interventions without navigating multiple disparate systems.

For QA directors, the architecture of both systems is worth scrutiny: the continuous learning pipeline in Golden Batch raises questions around model version control and change documentation that would surface during a 21 CFR Part 211 inspection, while Chanakya's cross-functional data orchestration depends on the integrity of every upstream system it queries. Neither challenge is insurmountable, but both warrant inclusion in process validation and data governance reviews before broader rollout.

ACG reports Golden Batch is now deployed across its capsule manufacturing plants, with measurable gains in first-pass yield and defect PPM reduction already recorded; the scale and pace of Chanakya's deployment will be a measurable indicator of how quickly the knowledge-retention gap closes across its packaging materials network.

Source: ACG via acg-world.com, 15 June 2026.