IoT Predictive Railway Maintenance — Sensor Mesh, IEC 62443 and EN 50126 RAMS

The State Railway of Thailand, MRTA and rail transport operators have long used scheduled preventive maintenance — which is safe but budget-wasteful. A Predictive Maintenance system driven by IoT + AI reduces unscheduled downtime by 30-50% and long-term repair costs by 10-25% while maintaining the same safety level. This article summarizes the IEC + EN frameworks that Thai railway operators must understand before they begin investing.

3 Maintenance Strategies — A Serious Comparison

ISO 13374-1 and EN 50126 divide maintenance strategy into 3 levels:

Data from ISO 13374-1 and case studies of European railways (Siemens Mobility, Alstom) 2020-2024

IoT Architecture — Sensor → Edge → Backend → CMMS

flowchart LR
  A1[Vibration
ISO 10816] --> B[Edge Gateway
onboard the train]
  A2[Temperature
IR + RTD] --> B
  A3[Acoustic Emission
AE sensor] --> B
  A4[GPS/IMU
track position] --> B
  B --> C[IEC 61375
Ethernet Train Backbone]
  C --> D[Train Server
data aggregation]
  D --> E[Wireless link
4G/5G + Wi-Fi yard]
  E --> F[Backend Cloud
ISO 13374 stages 1-6]
  F --> G[Predictive Model
AI + threshold]
  G --> H[CMMS Work Order
EN 50126 Phase 10]
  F --> I[IEC 62443
SOC + monitoring]
  H --> J[Maintenance
technician]

ISO 13374 — 6 Stages of Condition Monitoring

ISO 13374-1 + 13374-3 divide condition monitoring work into 6 stages:

1. Data Acquisition (DA) — sensor captures raw signal
2. Data Manipulation (DM) — filtering, FFT, statistics
3. State Detection (SD) — compare against baseline + threshold
4. Health Assessment (HA) — combine multiple variables, compute a health index
5. Prognostic Assessment (PA) — predict the time to failure (RUL = Remaining Useful Life)
6. Advisory Generation (AG) — automatically create a work order into the CMMS

A system that covers every stage and has an interoperability data format (ISO 13374-2 OSA-EAI) can pass data between vendors, reducing vendor lock-in

EN 50126 RAMS — A Lifecycle That Must Be Set From the TOR Onward

EN 50126-1:2017 (the railway RAMS standard replacing the old BS 50126) specifies 14 phases for the rail system lifecycle. In the context of predictive maintenance, the important ones are:

• Phase 4 — Risk Analysis: identify the failure modes the predictive system will detect + set the SIL (Safety Integrity Level 1-4)
• Phase 5 — System Requirements: specify the data sampling frequency, latency requirement, false-positive tolerance
• Phase 9-11 — Operation + Maintenance: actually operate the predictive system + iteratively tune the model
• Phase 12 — Performance Monitoring: measure system KPIs (true positive rate, mean time to detect, missed failure rate)

The key point: specify before you build, not build first and verify later. If you start at the TOR by including EN 50126 + ISO 13374 compliance, the system will have an audit-ready structure from the outset.

IEC 62443 — Cybersecurity That Cannot Be Forgotten

A predictive maintenance system is an OT (Operational Technology) network — not ordinary IT. IEC 62443 (ISA/IEC) is divided into 4 parts:

• 62443-1: terminology, concepts
• 62443-2: policies, procedures
• 62443-3: system security (zones + conduits, SL-T target)
• 62443-4: component security (secure development lifecycle)

For rail, in brief:

• Security Level 1 (SL-1) — protects against casual/coincidental violation. Minimum
• Security Level 2 (SL-2) — protects against intentional violation using simple means
• Security Level 3 (SL-3) — protects against intentional sophisticated means
• Security Level 4 (SL-4) — protects against state-actor level attack

New railway projects in Thailand should start at SL-2 as a minimum, and SL-3 for systems connected to signaling / traction control. EN 50701 (railway-specific cybersecurity) uses IEC 62443 as a baseline and adds railway context

Pilot Strategy — Start at the Traction Motor Bearing

For a fleet of 50-100 trains, rolling everything out at once carries high risk — a phased approach is recommended:

1. Phase 1 (months 1-3): pilot 5-10 trains, focus on 1-2 asset types (traction motor bearing + axle box bearing). Install vibration + temperature sensors, edge gateway, basic cloud dashboard
2. Phase 2 (months 4-9): tune thresholds + ML model from real data. Add Acoustic Emission for early warning. Integrate with CMMS for automatic work order generation
3. Phase 3 (months 10-18): roll out to the whole fleet of 50-100 trains. Add track monitoring (strain gauge, GPS, IMU)
4. Phase 4 (months 19-24): optimize ROI, integrate with the company's asset management dashboard

Why start at the traction motor bearing? Because:

• The failure cost is high (the whole train stops for 4-12 hours)
• The failure modes are well-known (inner race, outer race, ball, cage)
• The vibration signature at the axle housing is easy to measure
• ISO 13373 + 10816 standards cover the algorithm

6 Procurement Guidelines

1. The TOR specifies EN 50126 + ISO 13374 + IEC 62443 SL-2 as a minimum — forbid vendors from designing outside the framework
2. Data ownership clause — state clearly that the raw sensor data + processed analytics belong to the user, not the vendor
3. Open data format — mandate ISO 13374-2 OSA-EAI or the MIMOSA standard for portability
4. Pilot before fleet-wide rollout — the TOR specifies phased deployment + KPIs that must be met per phase
5. Cybersecurity audit — Type Test per IEC 62443-4-2 (component level) + 62443-3-3 (system level) before acceptance
6. Training + knowledge transfer — the in-house team must be able to operate + tune the model after the warranty expires — written into the acceptance criteria

Summary

Predictive Maintenance for Thai railways is an investment in the 30-60 million baht range for a fleet of 50-100 trains, with a payback of 2-4 years and a 5-year ROI of 200-350% compared with preventive-only. The core frameworks are EN 50126 RAMS (lifecycle), ISO 13374 (condition monitoring stages), IEC 62443 (cybersecurity), and IEC 61375 (train communication backbone). Start the pilot at the traction motor bearing because the failure cost is high + the sensor signature standard is mature

Sahawatthanakit has more than 20 years of experience working with the State Railway of Thailand (SRT) on rolling stock + signaling — our engineering team helps design specifications, perform EN 50126 hazard analysis, and select vendors for predictive maintenance projects from the pilot phase to fleet rollout

Frequently Asked Questions

How does Predictive differ from Preventive? Preventive follows a time/distance schedule — wasting the cost of parts still in good condition. Predictive uses sensors + AI to predict when a part is near failure, reducing unscheduled downtime by 30-50% and spare-parts cost by 10-25%. ISO 13374 divides it into 6 stages from data acquisition to decision support

How long until ROI pays off? A system for a fleet of 50-100 trains, investment 30-60 million baht, payback 2-4 years, 5-year ROI 200-350% vs preventive-only. Start with a pilot before the whole fleet to reduce risk

Is IEC 62443 really enforced in Thailand? In Thailand it is not yet enforced by law, but foreign EPCs mandate it via contract clauses for MRT/MRTA/SRT — SL-2 minimum, SL-3 for systems connected to signaling. EN 50701 uses IEC 62443 as a baseline

Which sensor type suits a wheel/bearing/axle? A vibration sensor at the axle housing measures the bearing via signature (ISO 10816). Acoustic Emission detects crack initiation earlier than vibration. A temperature sensor + IR thermography measures wheel overheating. Track: a strain gauge + tilt sensor measures rail defects + cant deviation

At which stage is EN 50126 RAMS done? EN 50126-1 has 14 phases — predictive maintenance work falls in Phase 9-11 (operation + maintenance). Hazard analysis (Phase 4-5) must identify the failure modes the sensors will detect + set SIL 0-4

Which asset is best to start the pilot on? Traction motor bearing + axle box bearing — high failure cost, standard vibration signature, ISO 13373 algorithm ready to use. Avoid piloting on signaling or catenary, where integration is complex