Answer extracted from The Manufacturing IT Podcast — listen to the full episode below.
How do manufacturers handle dirty MES data before feeding it into AI systems?
MES data isn't clean or static after initial recording—production counts get updated, test results arrive after delays, and corrections happen continuously over time. Manufacturers must correct this data before feeding it into AI systems, ensuring that machine learning models train on accurate information rather than incomplete or outdated records.
The real-world lifecycle of manufacturing data
Manufacturing Execution System data exists in a constant state of refinement. Quality results don't arrive all at once; they trickle in over hours or days as tests complete in the lab. Production counts change as operators reconcile initial estimates with actual output. Rework, scrap, and corrections introduce further variations. Data that was recorded as "complete" yesterday may be outdated today.
This reality creates a fundamental problem for AI applications. Machine learning algorithms learn from the data they consume. If that data contains stale information, partial records, or uncorrected errors, the AI model will internalize those flaws. The result is a system that makes decisions based on false patterns rather than ground truth, a point elaborated in this episode where manufacturing data experts discuss the mechanics of real-time data correction.
Why data quality is non-negotiable for AI success
The stakes are higher when artificial intelligence is in the loop. A human operator can spot an anomaly or question a report; an AI system trained on bad data will confidently produce bad recommendations. Correcting data upstream—before it reaches the AI system—is the only way to ensure reliable machine learning outcomes.
Manufacturers implementing AI for predictive maintenance, quality control, or production optimization must establish workflows that capture corrections as they occur. Late-arriving test results, updated production counts, and corrected data points must flow back into the data pipeline. Without this discipline, AI systems become unreliable decision-making tools rather than strategic assets, a challenge explored in depth in the discussion on data integrity.
"Many manufacturers struggle with getting clean MES data into AI systems. MES data isn't static after initial recording—quality results change, tests run after 24 hours, production counts get updated, and various corrections happen over time. The key challenge is correcting data that goes up to AI so that AI can learn from accurate data instead of bad data."
Tom Heckman — Founder and CTO at Sepasoft, with over 20 years in manufacturing and automation. Heckman started his career in the 1980s at an Allen Bradley distributorship's engineering department, focusing on manufacturing data collection and PLCs, and later founded Sepasoft in 2010 to build modular MES solutions on open platforms.
Tom Heckman's perspective reflects a decade and a half of building manufacturing software systems. The transition from traditional monolithic MES platforms to AI-ready systems requires not just new technology, but a fundamental shift in how manufacturers think about data governance and continuous correction. Want to hear more about how modular MES architecture enables this kind of real-time data refinement? The full episode covers how Sepasoft's approach to data modularity supports downstream AI applications.
Key takeaways
- MES data changes continuously after initial recording as test results arrive, production counts are reconciled, and corrections are applied.
- AI systems trained on incomplete or outdated manufacturing data will learn flawed patterns and make unreliable decisions.
- Manufacturers must establish workflows that capture and apply data corrections before feeding information into AI systems.
- Data quality discipline upstream is the only way to ensure AI-driven manufacturing decisions remain accurate and actionable.
See also
How did Stone Brewing improve operational efficiency after implementing MES software for brew house control?
Stone Brewing, a Southern California brewer recently acquired by Sapuro, deployed Sepasoft software to control their entire brewing process and beer production, achieving significant efficiency gains through integrated system control.
What are the key benefits of integrating MES directly with HMI and SCADA platforms on a single architecture?
By running modules on Ignition, both HMI SCADA and MES exist on the same platform, enabling simple controls like interlocks without requiring separate systems or complex data bridges.
How did modular MES architecture transform the manufacturing software market compared to legacy monolithic systems?
Legacy vendors sold huge monolithic systems that took years to deliver value and were highly complex. Sepasoft's modular approach provides flexibility, faster deployment, and the ability to adopt features incrementally.