Automated HAZOP from STID and Simulation

This workflow connects technical-document access, plant-data context, and NeqSim process simulation to generate a first-pass IEC 61882 HAZOP worksheet. It is a screening and preparation workflow for a chaired HAZOP team, not an automatic approval mechanism.

The implemented MCP bridge is runHAZOP. It accepts a standard NeqSim process JSON definition plus optional HAZOP nodes, failure modes, safeguards, evidence references, and a barrier register. The runner builds and runs the base process, generates equipment-failure scenarios, runs those scenarios on process copies, and returns HAZOP rows with simulation evidence and report markdown.

Scope

Use this workflow when an agent or engineer has one or more of these inputs:

STID/P&ID documents with equipment, lines, valves, instruments, and safeguards.
C&E charts, SRS documents, inspection reports, or operating procedures.
Historian tag mappings or representative operating data.
A NeqSim process model built from JSON, Python, Java, UniSim conversion, or manual modeling.
A barrier register for PSFs, SCEs, SIFs, alarms, relief devices, or procedural safeguards.

Human review remains mandatory. The generated worksheet proposes nodes, deviations, consequences, and safeguards, but a competent HAZOP team must verify node boundaries, causes, safeguard independence, action ownership, and the final risk ranking.

Workflow

Retrieve relevant documents into the task folder.

Use the STID/document retrieval workflow to place source documents under task_solve/.../step1_scope_and_research/references/. Record document IDs, revisions, page references, figure coordinates, and extraction confidence.
Extract process topology and safeguards.

From STID/P&ID and C&E sources, extract HAZOP nodes with nodeId, designIntent, equipment, safeguards, and evidenceRefs. Extract any PSF/SCE/SIF/relief evidence into a barrier-register JSON block.
Build or retrieve a NeqSim process model.

Use a standard runProcess JSON definition. The HAZOP runner accepts either processDefinition, processJson, or top-level fluid plus process.
Select failure modes.

Use failureModes to focus the study, or enableAllFailureModes for a broad equipment-failure sweep. Supported modes include valve stuck open/closed, cooling loss, heating loss, compressor trip, pump trip, blocked outlet, instrument failure, external fire, power failure, and loss of containment.
Run runHAZOP.

The runner uses AutomaticScenarioGenerator to inspect the ProcessSystem, map equipment failures to HAZOP-style deviations, apply scenarios to copied processes, run the modified simulations, and capture key pressure, temperature, mass-balance, and equipment KPIs where available.
Review HAZOP rows.

Each row includes node, equipment, guideword, parameter, failure mode, cause, consequence, safeguards, recommendation, document evidence references, and scenario-simulation status.
Link to barrier management and LOPA/SIL/QRA.

If barrierRegister is supplied, runHAZOP calls the barrier-register runner and includes validation plus LOPA, SIL, bow-tie, and QRA handoff blocks. Only barriers with valid evidence and performance standards should be credited as protection layers.
Generate report output.

Use reportMarkdown directly in task reports, or pass the whole output to reporting tools. The report should include assumptions, source documents, generated rows, simulation caveats, and the human-review gate.

Input Contract

Minimal structure:

{
  "studyId": "HAZOP-COMP-001",
  "runSimulations": true,
  "failureModes": ["COOLING_LOSS", "COMPRESSOR_TRIP"],
  "processDefinition": {
    "fluid": {
      "model": "SRK",
      "temperature": 298.15,
      "pressure": 10.0,
      "mixingRule": "classic",
      "components": {
        "methane": 0.90,
        "ethane": 0.07,
        "propane": 0.03
      }
    },
    "process": [
      {
        "type": "Stream",
        "name": "feed",
        "properties": {"flowRate": [5000.0, "kg/hr"]}
      },
      {
        "type": "Compressor",
        "name": "1st Stage",
        "inlet": "feed",
        "properties": {"outletPressure": [30.0, "bara"]}
      }
    ]
  },
  "nodes": [
    {
      "nodeId": "Node-01 First-stage compressor",
      "designIntent": "Compress feed gas to intercooler pressure",
      "equipment": ["1st Stage"],
      "safeguards": ["Anti-surge control", "Discharge pressure trip"],
      "evidenceRefs": ["EV-COMP-101"]
    }
  ]
}

A complete example is available through getExample with category safety and name hazop-study. JSON schemas are available through getSchema for run_hazop input and output.

Output Contract

The runHAZOP response contains:

Field	Purpose
`summary`	Counts for nodes, generated rows, enabled failure modes, and simulations
`process`	Baseline process name, equipment count, warnings, and report JSON
`nodes`	Node-level text reports generated from `HAZOPTemplate`
`hazopRows`	Machine-readable HAZOP worksheet rows
`scenarioResults`	Per-scenario simulation status, timing, and captured KPI values
`qualityGates`	Human-review, process-model, evidence, and generated-row flags
`barrierRegisterHandoff`	Optional barrier validation plus LOPA/SIL/bow-tie/QRA handoff
`reportMarkdown`	Report-ready markdown section

Quality Gates

Before accepting the generated worksheet:

Verify every node boundary against the STID/P&ID source.
Confirm all extracted equipment names match NeqSim process unit names.
Check that scenario perturbations represent the intended failure mechanism.
Review failed or non-converged scenario simulations manually.
Confirm safeguards are independent, specific, effective, auditable, and available.
Do not credit a barrier in LOPA unless it has traceable evidence and a performance standard.
Record chaired HAZOP team decisions, action owners, and due dates separately from generated output.

Implementation Notes

runHAZOP currently uses these components:

ProcessSystem.fromJsonAndRun(...) for the baseline simulation.
AutomaticScenarioGenerator for equipment-failure discovery and scenario execution.
ProcessSafetyScenario for applying perturbations to copied process models.
HAZOPTemplate for node worksheets and text reports.
BarrierRegisterRunner for optional evidence-linked safeguard handoff.

The runner is intentionally conservative. It proposes rows and captures scenario outputs, but it does not claim that a row is fully risk-assessed until the barrier register, LOPA/SIL/QRA, and human review are complete.