Engineering Copilot: Subsystem Analysis

Interpreting subsystem telemetry against design limits, physics models, and history to answer 'is this subsystem healthy and why'.

Structured subsystem interrogation

Subsystem analysis answers a bounded question: given a subsystem and its recent telemetry, is it within design intent, and if not, what is the mechanism. The Engineering Copilot decomposes the question into tool calls — pull the relevant telemetry window from L2, retrieve the design limits and the physics model, run a twin subsystem module, and compare.

Example: breeder magnet system

For the breeder's REBCO magnets (16.84 T peak field, 8 T on-axis) the copilot compares measured strain, temperature margin, and delta-T across pancakes against the quench-onset model, cross-checks with the strain and quench sensing telemetry, and reports the temperature margin to the current-sharing temperature with its uncertainty.

def analyze(subsystem, window):
    tele  = L2.query(subsystem, window)          # bounded read
    spec  = rag.retrieve(f'{subsystem} design limits')
    model = twin.module(subsystem)               # e.g. thermomech
    pred  = model.predict(tele.inputs)
    resid = tele.measured - pred
    margin = spec.limit - tele.worst_case
    return report(margin, resid, drivers(resid), cite=[tele, spec])

What a good analysis contains

• Measured value, model prediction, and residual with uncertainty
• Margin to the nearest design limit, not just a pass/fail
• Physical attribution of any residual (drift, degradation, miscalibration)
• Citations to the exact telemetry window and design document

The copilot distinguishes measurement problems from physical problems by cross-checking against the L2 signal-validation verdicts and calibration-drift tracking before attributing a residual to hardware. It never asserts a fault without evidence, and it labels low-confidence conclusions explicitly. Subsystem analysis feeds directly into anomaly triage when a residual crosses a threshold, and into maintenance reasoning when a trend indicates degradation. On the burner, the same machinery covers the plug and throat coils, the DEC train, and the expander thermal load.