# AI for AVEVA MES: Ask Questions Across MES, Historian and ERP Without Leaving Your Network > Source: https://netray.co/erp-ai/aveva-mes-ai-integration > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: AI for ERP ## Short answer AI for AVEVA MES, formerly Wonderware MES, connects a private model to MES records held in SQL Server, to AVEVA Historian time-series, and to your ERP, so supervisors and quality staff can ask plain-English questions about orders, genealogy, downtime and quality with the query shown. It is read-only by default and can run on-prem or air-gapped. ## Kind platform ## Erp Family mfg-systems ## Erp Products - AVEVA MES - AVEVA System Platform - AVEVA Historian ## Industries - Food and Beverage - Pharmaceuticals - Automotive - Consumer Goods ## Region Global ## Persona Plant Systems Manager ## Eyebrow AVEVA MES + private AI ## Intro AVEVA MES grew out of Wonderware MES and sits on the AVEVA System Platform (the ArchestrA object model), alongside InTouch HMI and AVEVA Historian. Many plants therefore have a layered picture: real-time objects and alarms on the System Platform, process and tag history in the Historian, and production orders, operations and genealogy in the MES database. The three are connected, but not in a way a supervisor can query as a single thing. The questions that matter cross these layers. Why did a batch fail quality, given the temperatures the Historian recorded during an operation, the material lots consumed, and the equipment used? Why did a line miss its schedule, when the ERP says the order was released on time? Answering today means a systems specialist, several tools, and a spreadsheet. AVEVA offers analytics and AI features across its portfolio, including industrial information tools, and they are worth evaluating first. A private layer is useful when data cannot go to a vendor service, when the same assistant must reach ERP and quality documents, or when you want governed agents that prepare actions for approval rather than act on their own. This page shows how a private AI layer sits on an AVEVA estate, what it reads and where it stops, and the questions to put to any vendor before connecting it to a production historian or a system that supports regulated records. ## Pain Points ### Historian, MES and ERP are separate views **Detail:** Tag history, execution records and orders are stored separately, with different keys and time bases. Correlating them for one batch is manual work. ### Quality investigations are slow **Detail:** Linking a failed result to process conditions, material lots and equipment requires several lookups, and the person who can do it is often the same engineer who is needed elsewhere. ### Alarm and downtime noise **Detail:** Alarm floods and coded stop reasons hide the real cause. Pareto views on codes alone mislead teams about priorities. ### Tribal knowledge in System Platform objects **Detail:** Behavior is embedded in object templates and scripts built over years. New engineers struggle to understand why a status or calculation looks the way it does. ### Data policy limits **Detail:** Recipes, yields and customer formulations are sensitive, and some companies prohibit sending them to external AI services. ## Use Cases ### Batch and lot investigation assistant **Description:** For a failed result, assemble the material lots, equipment, operations performed and key Historian trends around the event, with sources cited. **Erp Objects:** Production orders, operations, lot genealogy, Historian tags, quality results **Outcome:** Cuts the time to assemble investigation context; the root cause remains an engineering judgment. ### Downtime and alarm explanation **Description:** Summarizes alarms and state changes around a stop and proposes likely causes alongside coded reasons. **Erp Objects:** Alarm history, equipment states, downtime codes, maintenance orders **Outcome:** Clearer view of repeat stoppages; accuracy depends on alarm design. ### Shift and order status in plain language **Description:** Supervisors ask where an order stands and receive status combining MES progress, material availability and ERP due date. **Erp Objects:** Production orders, operation status, ERP due dates, inventory **Outcome:** Fewer manual status requests to the floor. ### Genealogy and recall readiness **Description:** Runs forward and backward trace across lots and shipments to test recall readiness or scope an event. **Erp Objects:** Genealogy records, shipments, supplier lots **Outcome:** Faster mock recalls and containment scoping; results reviewed by quality. ### Electronic batch record review support **Description:** Flags exceptions and missing data in a record for QA to review, and drafts a summary of deviations. **Erp Objects:** Batch records, deviations, data collections, signatures **Outcome:** Reduces time spent on routine review; QA retains approval. ### Historian trend search **Description:** Finds periods when a variable behaved unusually using natural-language descriptions and links them to production orders. **Erp Objects:** Historian tags, time ranges, order and batch windows **Outcome:** Faster discovery of drift patterns; tag naming consistency is critical. ### Object and script documentation **Description:** Reads System Platform templates and scripts and generates plain-language descriptions of how key statuses are derived. **Erp Objects:** Object templates, scripts, galaxy export files **Outcome:** Reduces key-person risk; output reviewed by the owner of each area. ## Architecture **Overview:** A private layer reads curated views of the MES SQL Server database, queries the AVEVA Historian for selected tags, and reads ERP data over its supported interfaces. Open-weight models run on local GPUs. The layer has no write access to the System Platform or control network. **Layers:** #### MES read-only access **Detail:** Dedicated SQL accounts on curated views of orders, operations, genealogy and quality results. #### Historian query path **Detail:** Read-only access to selected tags through supported Historian interfaces, limited to agreed tag lists and time windows. #### ERP join **Detail:** Production order, material and lot keys mapped to the ERP via read-only connectors, reusing existing MES to ERP integration keys. #### Private model serving **Detail:** Open-weight models via vLLM or Ollama on local GPUs, with no external inference calls. #### Governance **Detail:** Role-based access, complete logs of questions and queries, and approval before any action is proposed to a person or system. ## Deployment Options ### On-prem in the plant or data center **Fit:** Most AVEVA estates with segmented networks **Detail:** GPU servers connect to MES, Historian and ERP over approved IT-side paths. ### Private cloud **Fit:** Multi-plant companies with centralized data **Detail:** Replicated or extracted data in a private cloud account you control. ### Air-gapped **Fit:** Defense and highly sensitive formulations **Detail:** Isolated deployment with controlled data transfers and offline updates. ## Compliance ### The layer is read-only decision support; records and signatures remain in validated systems. Document intended use and risk with your quality unit. **Framework:** 21 CFR Part 11 and EU GMP Annex 11 ### AI summaries are drafts for human review, with source records cited. **Framework:** FDA QMSR / ISO 13485 (devices) ### Trace queries can support your traceability plan using your own records. **Framework:** FSMA 204 / lot traceability ### Access from the IT side via defined conduits, read-only, with no path into control networks. **Framework:** ISA/IEC 62443 ### Operator identifiers can be pseudonymized and personal-data scope is agreed up front. **Framework:** GDPR / data protection ## Integration Notes - Read the MES through curated views and avoid direct writes; document which release the views were built against. - Limit Historian queries to agreed tag lists and use summary or retrieval queries rather than raw high-frequency pulls. - Use consistent tag naming conventions in the semantic layer; inconsistent naming is the most common cause of poor trend search. - Reuse the keys already moving between MES and ERP to join the two, and check the timing of confirmations and backflush. - Export System Platform object metadata for documentation tasks rather than connecting to the live galaxy. - Keep any alerting or notification as a proposal to a person; do not connect to alarm acknowledgement. ## Product Fit ### Custom build **Why:** Historian correlation, genealogy agents and object documentation require a bespoke design per plant. ### ERPray **Why:** Provides ERP-side facts such as order due dates, inventory and costs alongside MES status. ### DataRay **Why:** Queries quality documents, SOPs and databases alongside MES data on-prem. ## Engagement ### Discovery **Duration:** 2-3 weeks **Deliverables:** - Map of MES, System Platform, Historian and ERP - Priority investigation and status questions - Security and data review ### Access and semantic layer **Duration:** 3-5 weeks **Deliverables:** - Curated MES views - Historian tag scope and naming map - ERP join design - Evaluation set ### Pilot **Duration:** 4-6 weeks **Deliverables:** - Investigation and status assistants live - Accuracy report - Logging and approval flow ### Scale and handover **Duration:** 4-8 weeks **Deliverables:** - More lines and plants - Runbook - User training - Support plan ## Buyer Questions - Which interfaces does the AI layer use for the MES, Historian and ERP, and with what privileges? - How do you limit Historian load? - How are tag names mapped and kept current? - Can every answer show its sources and queries? - How is the layer kept out of control networks? - How do you handle validated-system boundaries for Part 11 or Annex 11? - What do AVEVA's own analytics already cover? ## Faq ### Is AVEVA MES the same as Wonderware MES? **Answer:** AVEVA MES is the current branding of the product that was Wonderware MES, running on the AVEVA System Platform. Naming of modules has shifted over releases, so we confirm your exact version and modules in discovery rather than assuming. The AI approach does not depend on branding, only on the data you hold. ### Can AI query the AVEVA Historian safely? **Answer:** Yes, with limits. We use read-only access to an agreed tag list and bounded time windows, rely on summary retrieval instead of raw high-frequency scans, and test load during the pilot. The Historian team keeps control over access and can revoke it at any time. ### Does it replace AVEVA's analytics? **Answer:** Not necessarily. If AVEVA's tools cover your need and data policy permits them, use them. A private layer adds value when data must stay on-prem, when one assistant must span MES, Historian, ERP and documents, or when you want governed agents with approval steps. ### Can it help with electronic batch record review? **Answer:** It can flag exceptions, missing data and deviations, and draft a summary for QA. It does not approve or sign. Treat it as review support and document the intended use and risk assessment in your quality system before relying on it. ### What if our tag names are inconsistent? **Answer:** This is common. We map tags to business terms for the agreed scope, and you can expand coverage over time. Where naming is too inconsistent, we will say so in discovery rather than produce unreliable trend answers. ### Which models run on-prem? **Answer:** Open-weight models such as Llama, Qwen, Mistral, Gemma or gpt-oss class, served with vLLM or Ollama on your GPUs. We choose by evaluating on your own questions and hardware, not by public benchmark claims. ### How long is a pilot? **Answer:** Usually 9 to 14 weeks for one line or area including discovery. The variables are security approvals, the number of Historian tags in scope, and how clean the MES to ERP keys are. ## Search terms aveva mes ai, wonderware mes ai integration, aveva system platform ai, aveva historian ai, private llm aveva, mes historian erp ai, aveva mes genealogy, electronic batch record ai review, aveva mes erp integration, on-prem ai mes historian, aveva mes batch investigation, wonderware mes natural language query --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json