MachineMetrics + ERP + private AI
AI for MachineMetrics: Connect CNC Machine Data to ERP Jobs, Costs and Answers
Short answer
AI for MachineMetrics joins the machine data it collects (cycle times, part counts, downtime reasons, feed and spindle signals) with ERP jobs, routings and costs, so operations teams can ask why a job ran long or which machines lose the most time and get a grounded, cited answer. MachineMetrics data is exported through its API into a store you control, and the AI layer can run on-prem.
- ERP
- MachineMetrics
- Industries
- Machining, Aerospace, Automotive, Medical Devices
- Written for
- Operations Manager
MachineMetrics is a machine monitoring platform for discrete manufacturers. Edge devices read controllers over protocols such as MTConnect, Fanuc FOCAS, OPC UA and Modbus, and the platform records cycle times, part counts, machine states, operator-entered downtime reasons and, depending on the controller, signals such as spindle load and alarms.
That makes it excellent at the question what is the machine doing right now and why was it stopped. It is much less able to answer the question a plant manager actually asks: did this job make money? That answer needs the ERP's job, routing, standard hours, material and labour cost, and the machine data side by side.
MachineMetrics provides dashboards, OEE views, downtime analysis and an API, and many shops are well served by those alone. The case for a separate AI layer is the cross-system question, the open-ended investigation, and the plants whose security policy limits what machine data can sit in a vendor cloud next to ERP cost data.
This page explains how to bring MachineMetrics data and ERP data together for natural-language analysis, what the data actually contains, and the honest limits of what an AI can infer from machine signals.
What usually gets in the way
The problems we hear most from operations manager teams running MachineMetrics.
Machine time and job cost live in different systems
The ERP knows planned hours and cost per work order operation; MachineMetrics knows actual cycle, setup and idle time per machine. Comparing them is usually a manual spreadsheet exercise per job.
Downtime reasons are inconsistent
Operators select reason codes under pressure, so the same cause appears under several codes and a large share ends up as unclassified. Pareto charts look precise but hide the quality of the underlying entries.
Standards in the ERP are stale
Routing run times were set at quoting and rarely updated. Actual cycle data would correct them, but nobody has time to compare thousands of operations by hand.
Insight stops at the dashboard
A dashboard shows that Machine 14 had low utilisation last week. It does not explain whether that was a changeover, a material shortage visible in the ERP, or a maintenance event.
Cloud data sharing needs review
Some aerospace and defense customers restrict where part, program and process data may be stored, which complicates joining machine data with controlled ERP records in a public AI tool.
Where AI earns its place in MachineMetrics
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Job actual versus standard
Ask why a job overran and get actual cycle, setup and downtime from machine data next to the ERP routing standard and quoted hours.
Touches: ERP work orders, routing operations, standard run and setup hours; machine cycle and state data
Outcome: Job reviews become a few minutes of reading rather than hours of data pulling, and repeated overruns point to bad standards.
Routing standard correction proposals
Compare observed median cycle times with ERP standards and list operations where the gap is consistent, as a proposal for a planner to review.
Touches: Routing master, operation run times, observed cycle time distributions
Outcome: Quotes and schedules improve gradually as standards move toward reality; every change stays a human decision.
Downtime Pareto with context
Rank downtime by reason, then cross-check against ERP shortages, maintenance orders and changeover counts to suggest probable causes.
Touches: Downtime events and reasons, ERP purchase order due dates, maintenance work orders
Outcome: Shifts focus from top-line OEE to the few causes that repeat, with evidence you can inspect.
Capacity and promise date checks
Use observed machine availability and cycle rates to sanity-check scheduled load and promise dates in the ERP.
Touches: ERP finite or infinite schedule, work centre calendars, observed utilisation
Outcome: Fewer promise dates built on nameplate capacity; accuracy depends on how consistently machines are monitored.
Unclassified downtime cleanup
Group similar free-text notes and unclassified stops, then suggest consolidation of reason codes for the continuous improvement team.
Touches: Downtime reason code list, operator notes, shift and machine attributes
Outcome: Cleaner reason codes over time, so the next Pareto is more trustworthy.
Tool and program change correlation
Look for scrap or cycle changes following a program revision, tool change or new material lot recorded in the ERP.
Touches: Part program revision records, tool life data, ERP lot receipts and quality inspections
Outcome: Faster root-cause leads; correlation is presented as a lead to investigate, not proof.
Shift handover summaries
Draft a summary of what ran, what stopped, and which jobs are at risk, for the next shift lead to confirm.
Touches: Machine state history, open work orders, due dates
Outcome: Handover notes take minutes to review instead of being written from scratch.
Reference architecture
Machine data is pulled through the MachineMetrics API into a store in your environment, aligned to ERP jobs by machine, work centre and work order identifiers, and queried by a locally hosted model through a curated semantic layer.
- 1
Machine data ingest
Scheduled and incremental pulls of cycles, states, downtime events and part counts via the vendor API into a time-series or relational store you control.
- 2
ERP alignment
Mapping of machine IDs to ERP work centres and of job or part-program identifiers to work order operations, with exceptions listed where mapping is missing.
- 3
Semantic layer
Named metrics such as run time, setup time, idle time and first-article scrap defined once, so every question uses the same definition.
- 4
Model serving
An open-weight model on your GPUs or private cloud, using tool calls to query the store and ERP APIs rather than answering from memory.
- 5
Controls
Read-only access, query display, role scoping by plant, and human approval for any change to ERP routing data.
Integration notes for your ERP team
- Use the documented MachineMetrics API for data export and confirm rate limits and data retention windows before planning backfills.
- Agree on a single machine identifier shared between MachineMetrics, the ERP work centre and maintenance records.
- Capture job context by having operators or the ERP shop-floor integration tag the active work order, otherwise cycles cannot be tied to costs.
- Store raw events and derived metrics separately so definitions of OEE or utilisation can be changed without reloading data.
- Join to the ERP using its APIs for work orders, routings and standards; avoid writing to ERP tables directly.
- Control-level data such as spindle load depends on what each controller exposes, so list per-machine signal coverage early.
Deployment options
Hybrid: vendor platform plus on-prem AI
Shops happy with MachineMetrics cloud but unwilling to send ERP cost data out
Machine data is mirrored to your environment; the model and ERP data stay on-prem.
Private cloud analytics
Multi-plant manufacturers with a central data team
A central store collects machine and ERP data from each plant, with per-plant access scopes.
Air-gapped analysis copy
Programs with strict data handling
Periodic exports are moved into an isolated environment for analysis, accepting some latency in exchange for isolation.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
ITAR / EAR
Part, program and process data in the analysis store can be confined to infrastructure the customer controls; the customer's export program governs access.
CMMC / NIST SP 800-171
If cost and drawing-related data is controlled unclassified information, keeping the AI layer inside the enclave avoids adding an external processor.
AS9100 and IATF 16949
Machine and process evidence becomes queryable for audits, while the quality system remains the record.
Vendor terms
Review MachineMetrics data ownership and API terms; the integration uses only documented interfaces.
Where Netray fits
Custom build
The value is in the join between machine data and ERP jobs, which needs a bespoke semantic layer for your machines, jobs and cost model.
ERPray
For the ERP side of the question, ERPray can answer job, routing and cost questions on supported ERPs and be paired with the machine data store.
DataRay
If machine data lands in a database or file share alongside quality records, DataRay can chat across those sources on-prem.
How an engagement runs
Phase 1 . 1 to 2 weeks
Discovery
- -Questions that mix machine and ERP data
- -Machine to work centre mapping audit
- -API access and retention review
Phase 2 . 2 to 4 weeks
Data foundation
- -Ingest pipeline into your store
- -Defined metrics (run, setup, idle)
- -Job context tagging approach
Phase 3 . 3 to 5 weeks
Pilot
- -Job actual versus standard Q&A for one cell
- -Test set with verified answers
- -Feedback from planners
Phase 4 . 3 to 6 weeks
Rollout
- -More machines and plants
- -Standards review workflow
- -Monitoring and runbook
Questions to ask any vendor, including us
A short list that separates real MachineMetrics AI work from a chatbot demo.
- How will machine cycles be tied to ERP work orders, and who maintains that mapping?
- What share of our downtime is currently unclassified?
- Which machines expose which signals, and which are only on/off?
- Where will machine data and ERP cost data be joined, and who can see it?
- Can the answer show the metric definition and the underlying rows?
- How are changes to ERP routing standards approved?
- What does MachineMetrics already do that we should not rebuild?
Frequently asked questions
Does MachineMetrics have its own AI?
MachineMetrics provides analytics, OEE and downtime views and continues to add features, so check the current product first. Where a separate AI layer helps is the cross-system question, such as comparing machine actuals with ERP cost and standards, or where policy requires the combined data to stay on infrastructure you control.
Can AI tell why a machine stopped?
Only to the extent the data supports it. Controller alarms, operator reason codes and ERP context such as shortages can suggest probable causes, but the AI should present these as leads with evidence. It cannot see problems that no signal or person recorded.
Do we need to replace MachineMetrics dashboards?
No. Dashboards are the right tool for live monitoring on the floor. The AI layer is for investigation, cross-system questions and summaries, and sits beside them, using the same underlying machine data through the API.
Which ERPs does this work with?
Any ERP with a usable API or reporting interface. We have documented patterns for Infor SyteLine, Epicor Kinetic and Prophet 21, NetSuite, SAP and others. The mapping of work orders, routings and work centres is where most of the effort goes, not the ERP brand.
Can it update ERP routing times automatically?
We recommend against automatic updates. The AI can propose changes with supporting cycle data, and a planner or process engineer approves them. Write-back, if used, goes through the ERP's supported APIs with an audit trail.
What about machines with no controller data?
Older machines may expose only on/off or a signal from a current sensor. Questions about cycle time and utilisation can still work, but diagnostics will be shallower. We list per-machine signal coverage during discovery so expectations are realistic.
How long before we get useful answers?
Usually a few weeks after data is flowing and machine IDs are mapped to work centres. The slowest part is typically job context, because cycles cannot be tied to cost unless operators or the shop-floor integration tag the active work order consistently.
Related guides
AI for Epicor Kinetic, Beyond What Prism Covers
Add AI to Epicor Kinetic beyond Prism: private LLM over BAQs, BPM data, and REST v2, on-prem or private cloud, with honest guidance on when Prism already covers you.
ProShop + on-prem AIAI for ProShop ERP in AS9100 Machine Shops
Add AI on top of ProShop ERP to speed up NCR/CAPA drafting, FAI review, and document control, without moving AS9100 quality records off your network.
Infor SyteLine / CSI + AIAI for Infor SyteLine and CloudSuite Industrial
Add grounded AI to Infor SyteLine or CloudSuite Industrial: natural-language answers, agents over IDOs and ION, on-prem or CloudSuite deployment.
RAG + SQL + permissionsA Private LLM Grounded on Your ERP Data
How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.
Agents + approval gatesAI Agents for ERP, Running On-Prem
A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.
Opcenter Execution + private AIAI for Siemens Opcenter Execution MES, Running Inside Your Own Network
Add private, grounded AI to Siemens Opcenter Execution MES: ask questions over lots, work orders, nonconformances and genealogy without leaving your network.
Plan it with numbers
OEE Calculator (Overall Equipment Effectiveness)
Break your Overall Equipment Effectiveness into availability, performance, and quality, and see how far you sit from the 85% world-class benchmark.
Free ToolCycle Time & Capacity Calculator
Translate cycle time, parallel stations, shift pattern, and realistic utilization into hourly, daily, and annual production capacity.
Free ToolManufacturing Downtime Cost Calculator
Combine lost contribution margin, idle labor, and absorbed overhead into a defensible monthly and annual downtime cost - with recovery effects modeled.
GuideSmart Factory ERP Integration: Industry 4.0 Connectivity Guide
Connect your smart factory systems to ERP with this Industry 4.0 integration guide covering MES, SCADA, IoT platforms, and real-time production data pipelines.
GuideAI Predictive Maintenance with ERP Integration
Integrate AI predictive maintenance with your ERP system. Reduce unplanned downtime by 45% using sensor data, ML models, and automated work order generation.
GuideOn-Prem LLM Deployment Architecture: Reference Guide
Reference architecture for on-prem LLM deployment: inference servers, GPU sizing, RAG pipelines, and security zones for regulated manufacturers.
Talk it through with an engineer who knows MachineMetrics
Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.