From raw vibration and temperature to policy-grade maintenance and shutdown decisions
Problem → Fit
Problem. Machines do tell the truth — but only as numbers.
What humans need is:
• Is this safe?
• Is this drifting?
• Does this require slowdown, inspection, or shutdown right now?
Without shared meaning and timestamped policy, maintenance becomes guesswork and post-failure blame.
Fit. SSMDE turns each measurement into a policy-bearing safety state with:
• value (the raw reading),
• align (bounded stability / risk dial in (-1,+1)),
• band (actionable human status),
• manifest_id (declared safety rulebook),
• stamp (tamper-evident time and order proof).
The invariant dial (align)
Hardware telemetry can be noisy. But align is always bounded, fusable, and order-invariant:
a_c := clamp(a_raw, -1+eps_a, +1-eps_a)
u := atanh(a_c)
U += w * u
W += w
align := tanh( U / max(W, eps_w) )
a_rawis how far this reading is from safe baseline, declared in the manifest.wmay depend on load, runtime hours, strain, or other declared weighting.- Output is always
-1 < align < +1.
No spike overwhelms the history. No “rolling average magic.” No silent reinterpretation.
Collapse parity stays intact: phi((m,a)) = m
Meaning: the value stays exactly the measured value, always.
Mapping to policy bands
Declare band cutpoints and actions in the manifest:
bands := [
{ name:"GREEN", range:"[+0.35, +1.00)", action:"normal operation" },
{ name:"AMBER", range:"[-0.20, +0.35)", action:"slowdown + schedule inspection" },
{ name:"RED", range:"(-1.00, -0.20)", action:"controlled shutdown + notify maintenance <= 15m" }
]
band is not UI color.
band is an operational instruction.
What a hardware-state SSMDE record looks like
Minimal (dashboard-friendly)
{
value: { bearing_vibration_rms: 0.037 },
band: "AMBER",
manifest_id: "PLANT_A_BEARING_SAFETY_v7"
}
Full (safety, insurance, audit, and warranty-grade)
{
value: { bearing_vibration_rms: 0.037, rpm: 3400 },
align: -0.28,
band: "AMBER",
manifest_id: "PLANT_A_BEARING_SAFETY_v7",
stamp: "SSMCLOCK1|2025-11-05T10:14:55Z|sha256=8fe3...|prev=71bc..."
}
Reading this record:
- The vibration measurement is exact.
align = -0.28means drifting, risk emerging.band = AMBERmeans slowdown + inspection required.manifest_idtells you which safety policy defined AMBER.stampproves this statement existed at that time in that order.
No dashboard interpretation.
No dispute.
No blame post-failure.
Where a_raw comes from (declared per asset)
Choose one, declare in the manifest, and do not change mid-run:
a_raw := stability_score_from_baseline(vibration_fft, baseline_fft)
a_raw := -strain_ratio(actual_strain / max_safe_strain)
a_raw := -delta_temp_over_allowed((T_now - T_nominal) / margin)
a_raw := health_signal_from_model(soh_battery_pct)
Semantics MUST be declared:
positive = stable / predictable
negative = unstable / approaching failure mode
Producer workflow (sensor or gateway)
[ ] Measure raw value(s) exactly
[ ] Compute a_raw using declared baseline/logic
[ ] Run clamp → atanh → fuse → tanh to compute align
[ ] Determine band via manifest cutpoints
[ ] Emit SSMDE record with manifest_id and stamp
Consumer workflow (maintenance / control system)
IF band == "GREEN" → normal operation
IF band == "AMBER" → slowdown + inspection window
IF band == "RED" → trigger controlled shutdown + notify human
No hidden heuristics.
No “interpretation layer.”
The data itself gives instructions.
Fleet and vendor interoperability
Two plants, two vendors, two sensor OEMs — no problem.
The manifest makes the rulebook public and stable:
manifest_id: "PLANT_A_BEARING_SAFETY_v7"
Anyone can replay it later.
This is what makes insurance, warranty, maintenance contracts, and regulator audits finally operate on shared, provable truth.
Copy-ready real-world example
{
value: { pump_temperature_K: 341.7, load_pct: 82.1 },
align: -0.61,
band: "RED",
manifest_id: "PLANT_B_PUMP_HEAT_v11",
stamp: "SSMCLOCK1|2025-11-05T11:02:09Z|sha256=b73c...|prev=e221..."
}
Interpretation:
- This pump was already in failure approach territory at
11:02:09Z. - The threshold was not guessed — it is declared in manifest v11.
- No one can later say “we didn’t know.”
One-line takeaway
SSMDE makes machine telemetry actionable, explainable, and replayable — turning raw sensor values into safety-grade operational decisions with proof.
Navigation
Previous: SSMDE – AI Routing and Escalation (3.2)
Next: SSMDE – Chemistry / Process Safety (3.4)
Directory of Pages
SSMDE – Table of Contents