Electrical schematic and control-cabinet design generation (EPLAN)

In German machine and plant building, electrical engineers spend many days per machine drawing circuit diagrams and compiling the electrical BOM, mostly by copying and adapting earlier projects. Two things are mature: macro- and rule-based configuration (EPLAN macros, Cogineer) and automatic checks of cross-references and parts lists. Generative drafting of schematics from a free-text function description is early: it exists in vendor pilots and research, not as a proven product. The realistic gain today is a hybrid: the LLM selects and parameterises approved macros and checks the result, the deterministic EPLAN engine builds the pages, and the engineer reviews every page.

Problem & value

• Today: a machine builder (Maschinenbau) draws the circuit diagrams for each machine variant in EPLAN Electric P8, usually by inserting macros and reworking an earlier project. The electrical BOM, terminal plans and cable lists follow from the drawing, and errors surface late at panel assembly or commissioning. The analysis behind this entry estimates about 60 hours per project for circuit diagrams and electrical BOM in a typical machine-building firm.
• AI value: a first draft in hours instead of days, consistent use of company standards, fewer cross-reference and BOM errors, and know-how from retiring engineers captured in macros and rules.
• Evidence: EPLAN markets its rule-based generator Cogineer with "days or even weeks" of saved engineering time on large, repeatable projects; this is a vendor claim without published data [2]. I found no independent measurement for generative (LLM) schematic drafting. The 20 to 50 percent range in the front matter is an estimate for the engineering hours covered, higher for repeat variants and low for one-off machines.

Variants

• Configurable series machines (variant engineering): the best case. Macros plus a configurator (Cogineer) generate most pages from options; AI helps with the long tail and with requirement-to-option mapping.
• One-off special machines: little reuse; AI suggests macros from past projects and drafts sections, but the engineer draws most of it.
• Panel builders (Schaltschrankbau) for customers: the AI works from a customer circuit diagram or specification and derives component list and layout input for the cabinet (EPLAN Pro Panel).
• Check-only mode: no generation; an assistant reviews a finished project for cross-reference gaps, duplicate device tags, missing articles, and deviations from the company standard. Feasible now.
• Private/maker use: hobby tools (KiCad, LLM for netlists) exist but are not comparable to industrial cabinet design.

How it works

1. function description / I/O list / options
2. requirement parsing (LLM maps wording to functions, drives, safety circuits)
3. macro and rule selection (retrieval over macro library, configurator, past projects)
4. schematic build (EPLAN engine: macro placement, auto-connect, device tags, PLC I/O)
5. electrical BOM from Data Portal / article master
6. automated checks (cross-references, tags, terminals, BOM, company rules)
7. engineer review of every page (human-in-the-loop)
8. release: documentation, parts list to ERP, input for Pro Panel layout

• Deterministic core: EPLAN places macros, connects and numbers them, and derives reports. The LLM should not draw lines itself; its job is to choose and parameterise building blocks and to explain findings. A third-party MCP server that lets an AI assistant drive a running EPLAN instance exists, but I could not verify its vendor or maintenance status, so treat it as an experiment, not a product [1].
• Macro-based configuration: EPLAN Cogineer builds configurators from partial circuits (macros), value sets and rules, groups them into typicals, and generates project documentation for Electric P8 [1][2]. It is available only in certain module packages.
• Checks: comparing cross-references, device tags and BOM against rules is exactly what LLM-plus-rules assistants do well, and is the lowest-risk entry point.
• Human-in-the-loop: a qualified electrical engineer is responsible for the design (protection, wire sizing, safety functions) and signs it off.

Data preparation

Which data: the macro library, configurator rules, device and article master data (EPLAN Data Portal, manufacturer data, own parts master), approved reference projects, company standards, and I/O or function lists from mechanical and software teams.

Where it lives: EPLAN project folders and macro projects, a PDM/PLM or file share, the ERP article master, and Excel lists and heads of senior engineers.

Typical gaps: macros with inconsistent device tags or missing article numbers, several uncontrolled variants of the same circuit, projects copied and modified over years, articles in the schematic not matching the ERP master, and company standards that exist only as PDF or as habit.

Preparation steps: 1. Inventory and clean the macro library: one macro per function, complete article data, consistent tagging (PREP-009). 2. Link macros and reference projects to approved variants and revisions (PREP-010). 3. Synchronise article master between EPLAN and ERP (PREP-005). 4. Write down company standards as machine-readable rules (PREP-013, PREP-016). 5. Build a test set of 10 to 20 past projects with known-good BOMs for evaluation (PREP-018). 6. Restrict access to customer projects (PREP-023).

Tools: EPLAN macro and project management, spreadsheets, the ERP export.

Rough share of effort: 40 to 50 percent of the project. Macro and master-data clean-up is the real work and also the foundation for classic Cogineer automation. See GUIDE-003.

Implementation paths

Which path for whom

• SMB: Path A, because macro libraries, EPLAN's own configurator (Cogineer) and EPLAN checks are the proven part and need no AI infrastructure. Add a check assistant from a service partner where available.
• Enterprise: Path B, because linking EPLAN automation interfaces, PLM and ERP with a cloud LLM is custom integration work and the engineering data justifies it.
• Developer: Path C, because service providers and tool builders can combine a local LLM with the EPLAN API and open parsers for experiments.

Path A - Buy (recommended for SMBs)

• EPLAN Electric P8 with macro technique and Cogineer for rule-based generation [1][2]; EPLAN Pro Panel for the 3D cabinet layout; EPLAN Data Portal for manufacturer device data. EPLAN is a German vendor (Friedhelm Loh Group, Monheim) with a large regional partner network.
• Alternatives: WSCAD (DE), Zuken E3.series (JP/DE), SEE Electrical, Siemens/other ECAD with their own automation.
• Generative features: as of late 2026 I could not confirm a generally available generative schematic function from the vendor. Ask your EPLAN contact for roadmap and pilots before relying on it.
• Choose A when: you already use EPLAN, have repeat variants, and are willing to invest in macros before AI.

Path B - Build on cloud APIs

• Use an LLM from an EU provider or an EU-region frontier model with the EPLAN API (scripting/actions) to place macros, fill attributes, create reports and run rule checks. The LLM proposes, the EPLAN engine executes, a rules engine validates.
• Use retrieval over macros, past projects and standards (embeddings and vector search).
• Choose B when: you need a custom assistant tied to your ERP/PLM, accept project-specific build effort, and have clean macros.

Path C - Self-host (open source)

• An open-weight reasoning or code LLM on HW2 for requirement parsing and check explanations, with retrieval over exported macros and project XML. EPLAN itself remains a commercial, Windows-based tool.
• Suitable when customer projects are under strict confidentiality. The effort is high and the tooling immature.

Path D - Custom training

Research setting. Training a model to produce schematics requires thousands of cleaned, labelled circuit projects, which almost no Mittelstand firm owns, and circuit logic is safety-relevant. Not recommended outside vendors and research.

State of the art (2026)

• Mature: macro-based projects, rule-based generators (Cogineer), data-portal-based BOMs, and EPLAN's own checks [1][2].
• Emerging: LLM assistants that choose macros from a function description, translate I/O lists into PLC I/O pages, and explain check findings. Visible mostly as vendor pilots, service-provider prototypes and community tools; there is no standard benchmark.
• Open: fully generative schematics (free-form circuit generation) are research. Reasoning about protection, selectivity, wire sizing and safety circuits is unreliable without verification, so a generated circuit must always be checked by a qualified engineer.
• Limits and failure modes: plausible but wrong wiring, invented article numbers, ignored company standards, silent changes in macros, and designs that look complete but violate standards.

Shared resources & combinations

• Upstream: DOC-013 (reading existing drawings, PDF schematics from customers) turns legacy paper into structured data for reuse; DOC-014 for parsing data sheets; SCI-013 for requirements.
• Downstream: CODE-016 (PLC code generated from the same I/O list and function description), SCI-023 (mechanical production), SCI-024 (CAD/drawings), AGENT-027 (quote from the draft BOM) and KNOW-020 (find similar earlier projects and reuse them).
• Adjacent: SCI-016 (PCB/EDA) solves a related design problem at board level; SCI-025 (FMEA) for risk review of safety functions.
• Biggest combined effect: one structured function description and I/O list feeding schematics (SCI-027), PLC code (CODE-016) and quote (AGENT-027) removes triple data entry.
• Shared infrastructure: PLM/PDM or file store, article master, a vector index over macros and projects, and a workflow engine.

Effort, cost & KPIs

• Path A: weeks for Cogineer-style setup per product family, plus months of macro clean-up; licence for ECAD modules and partner services.
• Path B: 3 to 6 months for an assistant with EPLAN integration and checks.
• Path C: similar, with more in-house effort.
• KPIs: engineering hours per project, share of pages generated without manual change, BOM error rate at assembly, rework and commissioning changes, lead time from order to released documentation, share of projects using standard macros.

People & change

Involve early: management and the head of electrical design (Leiter Elektrokonstruktion), the engineers who maintain macros, IT for the EPLAN and ERP interface, the data owners of the article master, and sales if the configurator touches quoting. Add legal/compliance for conformity responsibility (GUIDE-010).

Works council: depends. Logging of who drew, changed or accepted which page can make the system suited to monitor performance under §87(1) no. 6 BetrVG; §90(1) no. 3 requires information about AI-supported work. A short works agreement that excludes individual evaluation usually suffices (GUIDE-004).

AI literacy: Art. 4 AI Act applies (since the 2026 Digital Omnibus as a duty to take measures that support AI literacy, not to guarantee a competence level); train engineers on typical AI errors (invented articles, wrong protection) and on checking every generated page (GUIDE-006).

Communication and pilot: engineers fear loss of craft and responsibility. Show that the engineer stays accountable and that macros turn senior know-how into a shared asset. Pilot with one product family and two engineers, measure hours per project and error rate (GUIDE-008, GUIDE-009).

Risks, regulation & compliance

• EU AI Act: minimal risk for an offline design assistant (Annex III not applicable). AI literacy (Art. 4, as recalibrated by the 2026 Digital Omnibus) applies. If AI-generated logic became part of a safety function of a machine, review again.
• Machinery safety and standards: DIN EN 60204-1 (electrical equipment of machines) is the base for the CE declaration of a machine; DIN EN 61439 (low-voltage switchgear assemblies) adds requirements for the cabinet, such as type and routine verification, and both partly overlap. The documentation duties (wiring diagrams, parts lists) fall to the manufacturer [3][4]. The Machinery Regulation (EU) 2023/1230 applies from January 2027. AI output does not lighten these duties.
• Liability: the machine builder or panel builder remains responsible; document that a qualified person reviewed AI-drafted designs, and keep the audit trail.
• GDPR: schematics normally contain no personal data; engineer names in logs and cloud tools are the exception.
• IP and confidentiality: customer circuit diagrams are often under NDA. Check whether cloud tools retain uploaded project data; prefer EU hosting or on-prem processing.
• Mitigations: deterministic generation through the EPLAN engine, rule-based checks on every output, mandatory review, versioned macro library, no free-form AI changes to safety circuits.

Getting started

1. Measure the baseline: hours per project for circuit diagram and BOM, share of reused macros, error types found at assembly.
2. Start with checks (cross-references, tags, BOM vs. article master): low risk, fast benefit.
3. Clean up the macro library and standardise one product family; evaluate Cogineer-style configuration first.
4. Pilot an LLM macro-selection assistant on past projects, compare against the engineer's solution.
5. Define the review rule: every generated page is approved by a named engineer before release.

Sources

1. EPLAN Infoportal - Cogineer introduction (configurators from macros, value sets and rules) - https://www.eplan.help/en-us/Infoportal/Content/Cogineer/2.9/Content/htm/cogineer_intro.htm
2. Design World - New EPLAN Cogineer automates generation of schematics (vendor claim of days or weeks saved) - https://www.designworldonline.com/new-eplan-cogineer-automates-generation-schematics/
3. Siemens - IEC 60204-1 vs. IEC 61439-1/2 for control panels - https://www.siemens.com/de-de/industries/industrial-machinery/control-panels/iec-standards/
4. WEKA - Schaltschränke und Maschinen richtig in Verkehr bringen (DIN EN 61439-1 and DIN EN 60204-1) - https://www.weka.de/produktsicherheit/schaltschraenke-maschinen-richtig-in-verkehr-bringen/