Insights Hub + private AI
AI for Siemens Insights Hub (MindSphere): Asset Data Joined to ERP, Answered in Plain English
Short answer
AI for Siemens Insights Hub reads asset models and time series through the Insights Hub APIs, joins them to ERP work orders, routings and material data, and lets engineers ask questions in plain English with the query and source shown. For plants that cannot send process data to a public model, the language model runs in the customer's own environment while Insights Hub remains the system of record for telemetry.
- ERP
- Siemens Insights Hub, MindSphere
- Industries
- Manufacturing, Automotive, Machinery, Electronics
- Written for
- Industrial IoT Lead
Siemens Insights Hub, the platform most people still call MindSphere, collects machine and sensor data from PLCs and edge devices, organises it as assets with aspects and variables, and stores it as time series. Plants that adopted it usually have a few dozen machines streaming well, a handful of dashboards built in the Monitor or Mendix apps, and a long list of questions nobody has time to wire up.
The gap is context. A vibration trend or a spindle load curve tells you very little until it is joined to what the machine was running: the work order, the routing operation, the part number, the tool, the shift. That context lives in the ERP and sometimes the MES, not in Insights Hub. Engineers end up exporting CSV from one side and pasting into a spreadsheet beside an ERP report.
An AI layer closes that gap only if it respects both sides. It must query Insights Hub through its published APIs with proper service credentials, query the ERP read-only under real roles, and show the engineer exactly what it fetched. It must also respect that Insights Hub is delivered as a cloud service, so the architecture question is what leaves the plant and what stays.
This page describes where AI genuinely helps Insights Hub users, how the integration works against the Time Series and Asset Management APIs, and where Siemens' own tooling is the better answer.
What usually gets in the way
The problems we hear most from industrial iot lead teams running Siemens Insights Hub.
Telemetry without production context
Insights Hub knows a machine's spindle load and temperature. It does not know which work order, part or operator produced the anomaly. That join sits in ERP shop-floor tables and is done by hand.
Asset model sprawl
Asset types, aspects and variables are named by whoever onboarded each line. Two identical machines can expose different variable names, which makes fleet-wide questions painful.
Dashboards answer yesterday's question
Monitor dashboards and custom apps cover anticipated views. A new question, such as which machines drifted before last month's scrap spike, means a data engineer and a notebook.
Cloud data residency concerns
Insights Hub is a managed cloud service. Defense, aerospace and customer-NDA-bound plants often cannot also send the same data to a public LLM, so an AI layer has to run privately.
Alert fatigue with no triage
Threshold events accumulate in Event Management. Without maintenance history and open work orders beside them, technicians cannot tell a known issue from a new one.
Where AI earns its place in Siemens Insights Hub
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Scrap spike root cause across telemetry and ERP
Engineer asks which machines and parts were involved in last week's scrap increase. The agent pulls time series windows from Insights Hub and joins them to production orders and scrap postings.
Touches: Production orders, operation confirmations, scrap postings, routing operations; Insights Hub assets and time series
Outcome: Hours of CSV joining shortened to a reviewable answer in minutes, with the underlying queries shown for verification.
Fleet comparison across identical machines
The agent maps variable names across inconsistent asset models and compares like machines on load, temperature or cycle behaviour.
Touches: Insights Hub asset types, aspects, variables; ERP equipment master
Outcome: Outlier machines found without a custom notebook per question; accuracy depends on a one-time variable mapping.
Maintenance planning with live condition
Summarise condition trends beside open maintenance orders and spare availability so planners can bring forward or defer work.
Touches: Maintenance orders, equipment master, spare parts stock; Insights Hub events
Outcome: Better-informed deferral decisions; the agent proposes, the planner approves any order change.
Energy per unit reporting
Combine energy variables from assets with produced quantity to report energy per part by product and line.
Touches: Production order quantities, work centers, cost centers; energy time series
Outcome: Per-product energy figures that previously needed manual alignment of timestamps and orders.
Event triage with history
For an incoming event, retrieve the last similar events, linked maintenance records and open orders, then draft a triage note.
Touches: Insights Hub Event Management; maintenance order history
Outcome: Faster triage and fewer repeated investigations of known issues.
Cycle time versus routing standard
Compare actual cycle behaviour from machine signals to routing standard times to flag stale standards.
Touches: Routing operations, standard times, work center capacity; cycle signals
Outcome: A ranked list of operations where standards no longer match reality, for engineering review.
Plain-language asset documentation
Generate readable descriptions of asset models, variables and units from metadata so new engineers can explore the data.
Touches: Asset Management API metadata; item and equipment master
Outcome: Onboarding time reduced; descriptions flagged for human confirmation before publishing.
Reference architecture
A private orchestration service sits between users and two read paths: Insights Hub APIs for telemetry and the ERP for production context. A locally hosted open-weight model plans queries, a policy layer enforces roles, and every answer shows the calls made.
- 1
Insights Hub access layer
Service credentials scoped to read assets, aspects and time series via the Asset Management and Time Series APIs; rate limits and paging respected; selected series optionally replicated to a local store.
- 2
ERP context layer
Read-only connection to ERP production, routing and maintenance data through APIs or reporting views, with the user's ERP role applied.
- 3
Semantic mapping
A maintained map from Insights Hub asset and variable names to ERP work centers and equipment IDs, reviewed by an engineer.
- 4
Private model serving
Open-weight models served with vLLM or Ollama on customer GPUs or a private cloud region; no prompts or telemetry sent to a public model API.
- 5
Audit and approval
Logged queries, sources and answers; any write-back such as a maintenance request requires explicit human approval.
Integration notes for your ERP team
- Authenticate with a dedicated service credential and scope it to read-only asset and time series access.
- Use the Asset Management API to enumerate asset types, aspects and variables, then build the variable mapping from that metadata.
- Query the Time Series API with explicit time windows and limits; do not let the model request unbounded ranges.
- Join to ERP on a maintained key such as equipment ID or work center, not on fuzzy name matching alone.
- Treat Event Management data as read-only context; do not acknowledge or close events from the AI layer.
- Check current tenant limits and API versions with Siemens documentation before sizing, as services have evolved since the MindSphere name.
Deployment options
Customer VPC with API pull
Plants already running Insights Hub who want AI inside their own cloud account
Orchestrator and model run in the customer's tenancy and pull from Insights Hub APIs over TLS; only the model host is private.
Local replica of selected series
Defense or NDA-bound plants
Chosen time series and asset metadata are synchronised to an on-prem store, and the model queries the replica and the ERP only.
Edge-adjacent deployment
Plants using Siemens Industrial Edge
Lightweight analysis near the line, with the language layer on a plant server; limited by edge hardware.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
IEC 62443
AI service is deployed in a defined zone with read-only conduits to telemetry and ERP; no path to controllers.
ITAR / EAR
Technical data stays within customer-controlled infrastructure; the customer defines which series may leave Insights Hub.
GDPR
Operator identifiers are minimised or pseudonymised before reaching the model; retention is set by the customer.
ISO 27001 customer controls
Credentials in a customer-managed secret store, logged access, and least-privilege API scopes.
How an engagement runs
Phase 1 . 1-2 weeks
Discovery
- -Asset and variable inventory
- -ERP join keys identified
- -Data residency decision
- -Priority questions list
Phase 2 . 2-3 weeks
Mapping and access
- -Service credentials and scopes
- -Variable to equipment mapping
- -Read-only ERP views
- -Test data window
Phase 3 . 3-5 weeks
Pilot
- -Two or three use cases live
- -Query transparency in the UI
- -Evaluation set with engineers
- -Audit logging
Phase 4 . 2-4 weeks
Hardening
- -Role enforcement tests
- -Runbook and monitoring
- -Rollout plan to more lines
- -Cost and GPU sizing review
Questions to ask any vendor, including us
A short list that separates real Siemens Insights Hub AI work from a chatbot demo.
- Which Insights Hub APIs and scopes does the AI layer use, and can they be read-only?
- How are asset variable names mapped to ERP equipment, and who maintains that mapping?
- Does any telemetry leave our tenancy to reach a model, and which model runs where?
- How does the system avoid unbounded time series queries and API limit breaches?
- Can it show the exact calls behind every answer?
- What happens when Siemens changes an API version?
- Where is the approval step for any write-back?
- When is Siemens' own analytics or Industrial Edge tooling enough?
Frequently asked questions
Is Siemens Insights Hub the same as MindSphere?
Insights Hub is the current name for the platform previously branded MindSphere. Existing documentation, integrations and community material often still use the MindSphere name, so both terms are useful when searching. Capabilities and API details have changed over time, so always confirm against current Siemens documentation for your tenant before building integrations or sizing an AI layer around it.
Can an LLM read Insights Hub time series directly?
Not safely as raw input. Time series can be very large, so the model should plan bounded queries through the Time Series API and receive summaries or small windows. A tool-calling pattern with explicit time ranges, series limits and unit handling keeps cost low and answers verifiable. The model should never hold broad credentials or free-form access.
Does this replace Insights Hub dashboards?
No. Dashboards are the right tool for known monitoring views. The AI layer is for ad hoc questions that cross telemetry and ERP data, such as linking a drift period to work orders and scrap. If your questions are stable and single-source, building a dashboard is simpler and cheaper.
Can this run fully air-gapped?
Insights Hub itself is a cloud service, so a fully air-gapped setup means working from a local replica of selected series or from edge data, not from the live cloud tenant. The model, orchestrator and ERP access can all be private. The trade-off is data freshness and which series you choose to replicate.
How is the join to the ERP done reliably?
Through a maintained mapping between asset IDs or variable names and ERP equipment or work center identifiers, reviewed by an engineer. Fuzzy matching is used only to propose candidates. Time alignment also matters: production order confirmations carry timestamps that must be reconciled with signal time zones and clock drift.
Will it write to the ERP or Insights Hub?
Read-only by default. If you want it to draft a maintenance request or an event note, it prepares the content and a person approves it. Controller or PLC writes are out of scope and should stay outside any language-model path.
Who is this for?
Plants with Insights Hub already streaming data and an ERP holding the production context, where engineers spend real time joining the two. If you have few connected machines, or only need standard monitoring, native Insights Hub applications are likely sufficient and a separate AI layer is not justified.
Related guides
AI 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.
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.
SAP PM/EAM + on-prem AIAI for SAP Plant Maintenance: Notifications, Work Orders, and Predictive Signals
AI grounded on SAP PM data that triages notifications, drafts work orders, standardizes failure codes, and flags predictive maintenance signals, on-prem.
Maintenix + on-prem AIAI for IFS Maintenix in aviation MRO
AI over IFS Maintenix maintenance records, AD/SB compliance, and task cards for MRO, deployed on-prem so tech records never leave your network.
On-prem AI, any ERP, A&DOn-Prem AI for ERP in Aerospace, Defense, and Electronics Manufacturing
A hub guide to on-prem AI across SAP, Infor LN, Costpoint, IFS, and Oracle EBS for aerospace, defense, and electronics manufacturers under ITAR, CMMC, and AS9100.
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 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.
Free ToolGPU Sizing Calculator for LLM Inference
Work out how many GPUs you need to serve a given open-weight model to your user base, based on memory footprint and token throughput.
GuideIoT Sensor to ERP Data Integration for Manufacturing
Build reliable IoT sensor to ERP data pipelines for manufacturing. Covers MQTT, edge processing, data normalization, and ERP API ingestion patterns.
GuideEdge Computing for Manufacturing ERP Integration
Deploy edge computing for manufacturing ERP integration. Covers edge gateway selection, local processing, latency reduction, and offline-capable ERP operations.
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.
Talk it through with an engineer who knows Siemens Insights Hub
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.