Steam Trap Acoustic Loss Monitor

The Problem

20-40% of steam traps in a typical plant have failed (blowing or cold). Each blowing trap wastes 50-280 kg/h steam — $3k-$30k/year in fuel + water + carbon tax. Traditional inspection is manual, subjective, and misses:

• Flash steam false positives: Hot downstream but low acoustic → flagged as blowing incorrectly
• Cold traps causing water hammer: Idle lines misdiagnosed, risk catastrophic pipe failure
• Archetype confusion: Disc trap cycling vs F&T continuous quiet vs Bucket clunk+burst require different models
• No dollar quantification: Technician says "failed" but no bounded loss min/exp/max kg/h, no $ impact for prioritization

Result: 3-5% of total site energy lost to failed traps, Scope 1 emissions underreported for EU ETS / GGPPA, maintenance backlog with no ROI ranking.

---

The Solution

Hardware-Agnostic Ingestion

Supports any ultrasonic contact probe (Sonicman, UE Systems, custom). Offline-first mobile UX caches locally, syncs when online. Raw audio S3 optional (WAV 16kHz) for audit trail.

Multimodal Physics-Gated Diagnosis

Physics gates prevent false positives — per ASME PTC 39 and ISO 7841. Cold clamp stops idle line misclassification. Blow-through proof distinguishes flash steam (high Tout, low acoustic) from true blowing (high Tout + high RMS + low subcool).

Bounded Loss & Economics

Not a single point estimate — a range with confidence:

• Min/Expected/Max kg/h via Napier calibrated per archetype (Cd 0.6-0.8)
• Research-validated: 6.35mm @ 1MPa gauge ≈75 kg/h, 10mm @1MPa ≈185 kg/h, 6.35mm @20 bar ≈280 kg/h
• Annual cost: Fuel MMBtu = (kg * h_fg / boiler_eff), Fuel $= MMBtu *$6, Water + Chemical, Carbon tons + $80/t tax
• ROI: Avoided loss - hardware - SaaS. Simple payback <1 month at 2000 traps, 18% failure, $22/t steam.

Fleet Analytics & Carbon Compliance

Dashboard: 4 metric cards (total traps, live loss, annual $, CO2), bar health distribution, pie breakdown, recent diagnoses. Analytics: loss by archetype bar, cumulative economic impact, timeline line chart, usage stats. Export ready for carbon accounting.

Auto Work Order & CMMS

If blowing or cold with confidence ≥85%:

• Auto-create work order with priority: CRITICAL for cold/water hammer, HIGH if loss>50kg/h, MEDIUM otherwise
• Fields: Title, TrapTag, Archetype, State, Confidence, Loss kg/h, Cost $, Rationale, Evidence JSON
• Export CSV for SAP PM M2 notification or Maximo MIF REST (fields: WorkOrderID, Priority, TrapTag, LossKgH, EstimatedCost)
• Status flow: OPEN → IN_PROGRESS → CLOSED after 60s baseline verification scan
• Technician checklist: isolate, replace, verify thermal gradient restored, acoustic baseline new.

---

How It Works — 7 Steps (Visual)

1. Create Site: Enter fuel cost $/MMBtu, carbon tax$/ton, steam cost $/ton, operating hours. Powers$/CO2 calc.
2. Register Traps: Tag, archetype (DISC, FLOAT_THERMOSTATIC, INVERTED_BUCKET, BIMETALLIC, BALANCED_PRESSURE, VENTURI), DN, orifice, pressure, baseline acoustic dB. NFC/QR link.
3. Field Inspection: BLE wand 10s burst → RMS, Peak, Crest, CPM, spectral centroid. Measure Tin/Tout under insulation, Tamb, header P.
4. Diagnosis: Physics-gated inference → state, confidence, bounded loss min/exp/max, annual $/CO2, rationale, warnings, evidence (Tsat, ΔT, subcool).
5. Work Order: Auto-create if ≥85% confidence, priority, CSV export SAP/Maximo, status tracking.
6. Fleet Analytics: Dashboard + analytics, ROI, carbon compliance.
7. Billing: Gatekeeper subscription lookup, balance €, 402 handling, marketplace link.

All steps have real UI screenshots (light-mode, 1280px verified) in Docs page.

---

Ground Truth & Validation

5-Level Pyramid (never train on L5):

• L1 ISO 7841 calorimeter 99%: dual-tank on load cells, 75 kg/h @1MPa 6.35mm
• L2 teardown 90%: wire-drawn seats, ruptured floats, prime loss
• L3 post-replacement recovery 75%: thermal gradient restored
• L4 technician consensus 55%: noisy supervision
• L5 model predictions 0%: never ground truth

Cross-validation: Group-K-Fold by trap_id, Leave-One-Site-Out to prevent acoustic memorization and ambient leakage.

Targets: >92% accuracy, <3% false positives on blowing, per independent Level II blind survey.

---

FAQ — Real Operator Questions

Q: How do you handle flash steam false positives? A: Physics gate: Tout≈Tsat but RMS low (<baseline+10dB) → NORMAL, not BLOWING. Requires high acoustic + low subcool (<3°C) + high ΔP.

Q: Cold trap water hammer risk? A: Cold Proof clamps Tin<Tamb+20°C to COLD, confidence >0.9, zero loss, CRITICAL priority. Immediate isolation per ASME PTC 39.

Q: Hardware-agnostic? A: Yes. BLE handheld 10s burst 8kHz, LoRaWAN hourly RMS, WirelessHART pressure+temp. All normalized to RMS/Peak/Crest/CPM/Tin/Tout/Tamb. Raw audio S3 for audit.

Q: Bounded loss formula? A: Napier: m_dot = 0.0165 * Cd * P_abs_kPa * d_mm². Cd 0.6-0.8 per archetype. Min 0.7x, Exp 1.0x, Max 1.3x scaled by confidence. Economic via IAPWS-IF97 h_fg, boiler eff 0.82, operating hours 8400.

Q: CMMS integration? A: CSV export for SAP PM M2 and Maximo MIF REST. Fields: WorkOrderID, Priority, TrapTag, Archetype, State, Confidence, LossKgH, EstimatedCost. Post-maintenance verification requires 60s baseline scan.

Q: Carbon compliance? A: Annual CO2 tons = (loss_tons * 2.68) with fuel carbon factor. Ready for EU ETS and GGPPA Scope 1 reporting. Fuel MW and $ also calculated.

---

Competitive Edge

---

Roadmap

• Q1: 100-trap pilot in 90 days, independent Level II blind survey, discrepancy teardown
• Q2: S3 raw audio + spectral features, edge quantized LightGBM <2MB ARM Cortex-M33
• Q3: Maximo MIF REST push, SAP PM BAPI, thermal camera integration
• Q4: Digital twin per trap, predictive failure (remaining useful life)

---

Built for: Energy managers, reliability engineers, field technicians in chemical, food, pulp, district heating plants.

Compliance: ISO 7841, ASME PTC 39, EU ETS, GGPPA, Sensifai Marketplace native (Logto, Gatekeeper, Webhook).