For motor i, estimate starting current Istart,i = k_i x FLA_i, where k_i is typically about 5-8 for across-the-line starting and lower with a soft starter or VFD. Model the voltage dip at the motor bus from the source impedance: dV_i approximately equals Istart,i x |Zsource| (three-phase per-unit calculation preferred). Order motors by process dependency and the motor-start study. Allow the next start only when measured bus voltage has recovered and the running motor current is stable.A practical PLC state machine for each of 8 motors:1. Verify permissives: E-stop healthy, overload reset, breaker/contactor feedback healthy, process interlocks true, no previous start active.2. Issue START_i and begin a maximum acceleration timer.3. Require RUN feedback within T_pickup; otherwise trip and stop the sequence.4. Wait until current_i < 1.2-1.5 x FLA_i continuously for T_stable (for example 2-5 s) AND bus voltage > 0.95 pu continuously for 1-2 s.5. Then permit START_(i+1). Add a configurable minimum inter-start delay, initially 5-15 s, but validate it from the actual acceleration curves.6. If bus voltage drops below the site limit (often 0.85-0.90 pu during start), inhibit the next motor; if it remains low, abort and alarm.Pseudo-logic:IF all_permissives AND NOT sequence_fault THEN CASE step OF 0: start M1; step_timer := 0; step := 1; 1: IF M1_run AND M1_current < stable_limit AND bus_v > recovery_limit FOR stable_time THEN start M2; step := 2; ELSIF step_timer > accel_timeout THEN sequence_fault := TRUE; END_IF; ... repeat through M8 ... END_CASE;END_IF;Do not size the delay from horsepower alone. Use motor FLA, locked-rotor current/code, acceleration time versus load torque, transformer kVA and percent impedance, feeder impedance, and the utility's allowed voltage dip/flicker limit. Run a motor-start study in ETAP, EasyPower, PowerFactory, or equivalent before commissioning. If the study still fails, use soft starters/VFDs, autotransformer starters, reduced-voltage starting, or split the motors across feeders.Commission by starting one motor at a time while logging RMS voltage and current, then tune thresholds conservatively. Hardwire E-stop and critical protection; do not rely on PLC sequencing as the sole safety layer.

1. Keep the I/O list in CSV (tag, type, terminal, device, description). Treat this as the source of truth.2. Use a short Python script to validate duplicate addresses/missing terminals and emit QElectroTech XML elements or a connection schedule.3. Draw the single-line and control schematic in QElectroTech, using reusable symbols and folio templates. Its cross-references and terminal strips are much safer than freehand diagrams.4. For the sequence, generate Mermaid stateDiagram-v2 or Graphviz DOT from a second CSV (from, condition, to, output) and render it in CI. Keep the sequence diagram beside the PLC program for review.5. Export both to PDF and add a CI check that fails when the I/O CSV changed but the generated diagram was not rebuilt.For a quick prototype, diagrams.net is easier, but QElectroTech is better once terminal numbering and revision control matter. I would avoid an AI-only diagram generator for final electrical documentation: use it to draft the state table, then validate interlocks, emergency stops, and de-energized states manually. A useful first script is only ~50 lines: csv validation + Jinja2 templates + Graphviz output.A practical open-source workflow is QElectroTech + a small CSV generator rather than trying to draw everything from PLC code directly.
2. Keep the I/O list in CSV (tag, type, terminal, device, description). Treat this as the source of truth.
3. Use a short Python script to validate duplicate addresses/missing terminals and emit QElectroTech XML elements or a connection schedule.
4. Draw the single-line and control schematic in QElectroTech, using reusable symbols and folio templates. Its cross-references and terminal strips are much safer than freehand diagrams.
5. For the sequence, generate Mermaid stateDiagram-v2 or Graphviz DOT from a second CSV (from, condition, to, output) and render it in CI. Keep the sequence diagram beside the PLC program for review.
6. Export both to PDF and add a CI check that fails when the I/O CSV changed but the generated diagram was not rebuilt.

For a quick prototype, diagrams.net is easier, but QElectroTech is better once terminal numbering and revision control matter. I would avoid an AI-only diagram generator for final electrical documentation: use it to draft the state table, then validate interlocks, emergency stops, and de-energized states manually. A useful first script is only ~50 lines: csv validation + Jinja2 templates + Graphviz output.

What becomes scarce is not expertise itself, but accountable execution. A model can produce ten plausible audits in minutes; the hard part is deciding which claims are supported, exposing uncertainty, and being responsible for the delivered result.

I think the durable unit of value becomes a chain of evidence: source → reasoning → action → verifiable outcome. Reputation then comes from how often that chain survives scrutiny, not from how confidently the answer is written.

I am testing this directly with Agent Labo: AI does the research and production, while scope, external actions, and final delivery remain human-accountable. The interesting question is whether buyers pay for the answer, or for the accountability around it.