The Client
A leading global supplier of advanced materials and process solutions to the semiconductor and high-tech industries, this organization operates large-scale manufacturing facilities where batch quality and process consistency are critical to customer outcomes. Managing hundreds of process parameters per production batch, the client’s engineering teams are responsible for diagnosing root causes of quality failures quickly and accurately to protect yield, reduce waste, and maintain delivery commitments.
The Challenge
Inefficient root cause analysis processes force highly skilled engineering teams to waste time on manual data review rather than innovation. This reliance on intuition over evidence results in costly operational delays and an inability to systematically optimize production yield.
- Manual and Time-Intensive Root Cause Analysis: Each OCAP event required engineers to manually review hundreds of process parameters, a process taking 30–60 minutes per case across approximately 250 cases per week, consuming around 187 engineering hours weekly.
- Intuition-Based Diagnosis: Root cause analysis was driven by individual engineering judgment rather than data, producing variable outcomes across engineers and making institutional knowledge difficult to capture or scale.
- Delayed Process Improvement: Weekly review meetings were required before root causes could be confirmed and corrective actions initiated, creating a lag between failure detection and process improvement.
- No Unified Data View: Raw materials, process parameters, machine sensors, genealogy, and inspection data existed in separate systems with no common linkage to batch IDs, making end-to-end analysis impractical.
The Impact
- 80–90% Of OCAP analysis automated
- 150+ Engineer hours saved per week
- Minutes Analysis time per case, down from 30–60 min
The Solution
Hoonartek built an AI-powered OCAP intelligence platform on GCP, unifying raw materials, process parameters, machine sensor data, genealogy, and inspection records by batch ID in BigQuery via a Dataflow real-time ingestion pipeline. LASSO, PLS, and XGBoost models were trained and deployed on Vertex AI to identify the process parameters driving quality failures across hundreds of variables, replacing intuition-based diagnosis with ML-driven precision.
A multi-agent AI layer was built on Vertex AI Agent Engine to translate model outputs into actionable intelligence for engineers. An Analysis Agent retrieves ML outputs and extracts key influential parameters; an Explanation Agent converts technical results into plain-language root cause narratives engineers can act on immediately; and a Visualization Agent renders interactive charts, highlights abnormal parameters, and compares batches against normal baselines. The entire platform was deployed on Cloud Run with full CI/CD via Cloud Build, enabling engineers to move from event selection to root cause diagnosis in minutes rather than hours.
Key Benefits
- Automated Root Cause Diagnosis: 80–90% of OCAP analysis is now automated end-to-end, freeing engineering capacity for higher-value optimization and process improvement activities.
- Engineering Productivity: Over 150 engineer hours saved per week, equivalent to reclaiming nearly four full-time engineers’ capacity from manual analysis.
- Faster Defect Resolution: Per-case analysis time reduced from 30–60 minutes to minutes, enabling corrective actions to be initiated in the same shift as the failure event.
- Consistent, Data-Driven Analysis: ML-driven diagnosis replaces variable intuition-based methods, producing consistent, auditable root cause outputs regardless of which engineer reviews the case.
- Continuous Process Improvement: Real-time analysis cycles replace weekly review meetings, enabling continuous feedback loops that accelerate batch yield improvement over time.