IIoT Predictive Maintenance: From Sensor Data to Real-Time Field Alerts
Predictive maintenance isn't limited to new plants with built-in IoT sensors. It can be deployed on existing equipment, using their current PLCs, in just a few days—and you'll start receiving relevant alerts as early as the first week.
Why Corrective Maintenance Is So Expensive
Equipment that breaks down in the middle of production triggers a chain reaction:
- Cost of the breakdown: parts + labor (€300 to €2,000)
- Unplanned downtime: 30 minutes to several days
- Loss of production: often 10 to 50 times the repair cost
- Emergency: technician called in at night or on weekends (+50% of the hourly rate)
Preventive maintenance (scheduled inspections) partially solves the problem: maintenance is performed either too early (replacing parts that are still in good condition) or too late (the equipment breaks down before the scheduled inspection).
Predictive maintenance aims for the right balance: taking action when the data indicates deterioration—not before, not after.
Data That Warns of a Failure Before It Happens
Universal Early Warning Indicators
| Parameter | Affected equipment | Failure indicator |
|---|---|---|
| Bearing Temperature | Motors, pumps, compressors | Deviation of +10°C from the usual value |
| Motor Current | All electric motors | Gradual increase (mechanical load) |
| Vibrations | Motors, fans, spindles | Increased amplitude or new frequencies |
| Differential pressure | Filters, pumps, heat exchangers | Increase = fouling |
| Energy consumption | All equipment | Deviation outside the nominal range |
| Cycle time | Presses, conveyors, production lines | Lengthening = mechanical wear or quality issue |
What the machines are already reporting
The good news is that most of these parameters are already measured by your PLCs and drives. They are stored in Modbus registers—but no one is reading them continuously.
Variateur Schneider ATV320 — registres Modbus accessibles :
Register 0x0054: Motor Current (A × 10)
Register 0x0C14: Radiator Temperature (°C)
Register 0x0C16: Operating time (h)
Register 0x0018: Output Frequency (Hz × 10)
Register 0x002C: Line voltage (V)
Register 0x0068: Number of boots
Register 0x2100: Fault code active
Register 0x2101: History of the Last 8 Faults
All of this is available without modifying the PLC program. All you have to do is read it.
IIoT Architecture for Predictive Maintenance
Configure Anomaly Detection
Static Thresholds vs. Dynamic Thresholds
Static thresholds: You set an absolute limit value.
# Simple Alert Based on a Fixed Threshold
alerts:
- tag: temperature_roulement
condition: "value > 85"
severity: critical
message: "Température roulement Moteur-1 critique : {value}°C"
channels: [sms, email]
delay: 0
Dynamic thresholds: An alert is triggered when the value deviates from its usual behavior.
# Drift Alert — Smarter
alerts:
- tag: courant_moteur
condition: "value > baseline_30d * 1.15"
# Alert if current value > 115% of the average over the last 30 days
severity: warning
message: "Courant Moteur-2 en dérive : {value}A (référence 30j : {baseline}A)"
channels: [email]
delay: 30min # Triggers only if the condition persists for 30 minutes
Dynamic thresholds detect subtle changes that fixed thresholds miss—particularly gradual drifts over several weeks.
Real-world example: filter clogging detection
Historical Data — Hydraulic Filter Presse-3:
Semaine 1 : ΔP = 0,8 bar → Normal
Semaine 3 : ΔP = 1,1 bar → Normal (acceptable)
Week 5: ΔP = 1.4 bar → WARNING alert triggered
Week 6: ΔP = 1.8 bar → CRITICAL alert (absolute threshold)
Action: Scheduled filter replacement in Week 5
→ Service call in working hours; part ordered in advance
→ Zero unplanned outages
Without continuous monitoring, this filter would have been replaced during the next monthly inspection—or would have caused a hydraulic failure.
Field Alerts: The Right Message to the Right Person
Notification Hierarchy
| Level | Sending Interval | Recipients | Channel |
|---|---|---|---|
| INFO | Daily (summary) | Maintenance Manager | |
| WARNING | < 5 min | On-call technician | Email + app |
| CRITICAL | < 1 min | Technician + supervisor | Text message + email |
| URGENT | Immediate | Entire team | Text message + phone call |
What Makes an Alert Useful
An effective alert includes:
[EZIWAN — WARNING] Current drift in Engine 2 — Atelier Nord
Equipment: Atlas Copco compressor motor (Atelier Nord)
Current value: 18.4 A (normal: 15.8 A — deviation +16%)
Duration of the issue: 45 min
History: Value has remained stable for the past 3 months, but has been drifting since yesterday morning
Suggested action: Check the vacuum filter and room temperature
→ Direct VPN access: https://app.eziwan.com/sites/atelier-nord
→ Historique complet : https://app.eziwan.com/charts/moteur-2
The technician has all the necessary information to decide whether the service call is urgent or can wait until the next shift.
ROI measured over 6 months
Results observed at an electrical panel manufacturer (45 machines, 3 workshops):
| Metric | Before | After 6 months |
|---|---|---|
| Unplanned outages per month | 3.2 | 0.8 |
| Unplanned downtime (hours/month) | 28 hours | 6 hours |
| Average cost of an emergency breakdown | €1,200 | €800 (preventive) |
| On-call technician visits | 12/month | 3/month |
| Estimated monthly savings | — | €5,500 |
The deployment (6 gateways, configuration, training) paid for itself in 3 months.
Where to Start
- Select 3 critical pieces of equipment — those whose failure is most costly
- Identify the available Modbus registers — typically 10 to 20 per piece of equipment
- Deploy a gateway as a pilot — 1 day of technician time
- Monitor for 2 to 4 weeks — build your baselines
- Configure the first alerts — start with the obvious critical thresholds
- Refine gradually — add drift alerts over time
Vibration Analysis: Advanced Monitoring for Pumps and Motors
Vibration is the earliest indicator of mechanical degradation—often detectable 6 to 12 weeks before a failure. For critical rotating equipment (pumps, compressors, fans), vibration analysis is the natural complement to Modbus monitoring.
Fundamental Vibration Indicators
| Indicator | Meaning | Fault Signal |
|---|---|---|
| RMS (Root Mean Square) | Total vibration energy | Gradual increase > 20% of the nominal value |
| Crest Factor | Peak-to-RMS ratio | > 3.5: early signs of bearing damage |
| Peak | Peak value | Occasional shocks, impacts |
| Kurtosis | Detection of periodic shocks | > 4.5: incipient bearing failure (balls) |
Real-world example — centrifugal pump:
Semaine 1 : RMS = 2,8 mm/s Crest = 2,1 Kurtosis = 2,8 → Normal
Semaine 5 : RMS = 3,4 mm/s Crest = 2,3 Kurtosis = 3,2 → Surveiller
Week 9: RMS = 4.1 mm/s Crest = 3.2 Kurtosis = 5.8 → WARNING: Bearing needs to be inspected
Week 11: RMS = 6.2 mm/s Crest = 4.7 Kurtosis = 8.4 → CRITICAL Alert: imminent failure
The deterioration was visible 6 weeks before the failure in the kurtosis and crest factor metrics, whereas the RMS alone would not have triggered an alert until week 9.
Industrial IoT Vibration Sensors
For IIoT deployments that do not require extensive cabling, wireless vibration sensors transmit data via MQTT or RS-485:
- SKF Enlight Collect IMx-1: BLE + Wi-Fi/Ethernet gateway, built-in RMS/kurtosis calculation
- Samsara VT100: Built-in LTE, GPS + vibration, 7-year battery life
- Bosch CISS: BLE, multi-parameter sensor (vibration + temperature + humidity)
- VibraWave WM-100: RS-485 Modbus, magnetic mounting, suitable for industrial cabinets
On existing equipment without easy access for wiring, a magnetic vibration sensor with MQTT, Wi-Fi, or LoRaWAN transmission can be mounted on the motor housing in 30 seconds. The data is transmitted to the Eziwan Gateway via its local MQTT broker.
Monitoring Profiles by Equipment Type
Centrifugal Pump (Comprehensive Monitoring)
# Eziwan Profile — Centrifugal Pump
equipement: pompe-centrifuge
collecte:
modbus:
- registre: "courant_moteur" # Variateur ATV320 : 0x0054
seuil_warning: "+15% baseline"
seuil_critical: "+30% baseline"
- registre: "pression_aspiration" # 4–20 mA Sensor → 0–10 bar
seuil_warning: "< 0.3 bar"
- registre: "pression_refoulement" # 4-20 mA Sensor
seuil_warning: "< setpoint × 0.9"
- registre: "temperature_palier" # Pt100 → sonde roulement
seuil_warning: "> 75°C"
seuil_critical: "> 85°C"
capteur_vibrateur:
- parametre: "rms_radial"
seuil_warning: "> 4.5 mm/s"
seuil_critical: "> 7.1 mm/s" # ISO 10816, Classe II
- parametre: "kurtosis"
seuil_warning: "> 3.5"
alertes:
- condition: "pression_aspiration < 0.3 AND courant_moteur > 120%"
message: "Risque cavitation — pompe {id}"
canal: [sms, email]
priorite: critical
Screw compressor (current and temperature monitoring)
equipement: compresseur-vis
collecte:
- courant_phase_L1_L2_L3 # Phase imbalance → winding fault
- temperature_huile # > 100°C: cooling problem
- pression_sortie # Abnormal pressure drop: clogged air filter or leak
- heures_fonctionnement # Scheduled Maintenance Trigger
- nombre_demarrages # Starter motor wear if > N/day
alertes:
- condition: "heures > 3000" # Vidange huile constructeur
message: "Entretien 3000h requis — Compresseur {id}"
canal: [email]
Common Mistakes to Avoid
1. Trying to Monitor Everything Right from the Start The temptation to connect all devices at once is real—but counterproductive. Start with three critical devices. Learn how to interpret the data, calibrate your baselines, and then expand gradually. A well-executed pilot on three machines is better than a chaotic rollout on 30.
2. Neglecting the baseline phase Predictive maintenance relies on comparing the current state to a baseline. If you set up alerts without allowing time to collect normal operating data (at least 2 to 4 weeks), you’ll generate false alerts—and teams will ignore the system.
3. Confusing Raw Data with Actionable Information A Modbus register that returns "18423" means nothing without its unit, scaling factor, and context. For each variable, document the physical unit, normal range, and the meaning of high and low values. The time invested in initial configuration pays for itself tenfold over the following years.
4. Ignoring data quality alerts If a sensor fails or an RS485 cable breaks, your system will report zeros or erratic values. Set up alerts for communication quality (no Modbus response for N minutes)—so you don’t mistake a silent sensor for “healthy” equipment.
5. Overly Sensitive Alerts = Ignored Alerts If the team receives 20 alerts a day, they’ll ignore them all—including the real ones. Calibrate your thresholds so that the false-alert rate is less than 2–3 per week during your initial deployment period. Refine them gradually.
Frequently Asked Questions
Do I need to modify the PLC program to collect data? No. The Eziwan gateway reads Modbus registers in passive (read-only) mode. It queries the PLC or drive as a Modbus master, without requiring any program modifications. Only the device’s Modbus address and the registers to be read need to be configured in the Eziwan dashboard.
Can data be collected from devices that do not have an RS485 port? Yes, there are several options: (1) if the device has an Ethernet port, Modbus TCP is often available; (2) 4–20 mA transmitters with RS485 converters allow you to connect analog sensors; (3) pulse counters can convert digital signals into Modbus data. Contact our team for a compatibility audit.
How do I know which Modbus registers to read on my equipment? The equipment's Modbus communication manual (often called the "Communication Manual" or "Modbus Register Map") lists all available registers. For common devices (Schneider ATV, Siemens SINAMICS, ABB ACS, WEG CFW), Eziwan provides preconfigured profiles available directly in the dashboard.
Does predictive maintenance work for older electromechanical equipment? Yes—often even better than on newer equipment, because older equipment has slower and more predictable degradation cycles. The key is having access to operating parameters, even via an external current sensor (current transformer on the power cable) if the equipment lacks a native communication protocol.
What is the difference between predictive maintenance and condition-based maintenance? Condition-based maintenance triggers an intervention when a condition is met (e.g., temperature > 85°C). Predictive maintenance analyzes trends to anticipate deterioration before the critical threshold is reached (e.g., an alert is triggered if the temperature has been rising by 3°C per week for the past 3 weeks, even if it is still at 72°C). Conditional maintenance is reactive; predictive maintenance is proactive.
Further Reading
- Guide: Industrial Multi-Site Monitoring — Complete Architecture
- Blog: Modbus RS485 Data Collection to the Cloud — Protocols Explained
- Blog: ROI of Industrial IoT — How to Calculate It
- Docs: RS485 / Modbus RTU configuration on the Eziwan Gateway
- Use Case: Predictive Maintenance for OEMs
Get started with your predictive maintenance driver → · Modbus and RS485 documentation →
Additional Resources
- IIoT Predictive Maintenance — predictive maintenance solutions for the connected industry
- Industrial Equipment Monitoring — monitor your machines and equipment in real time
- Industrial Remote Diagnostics — remote diagnostics for your industrial equipment
- Cloud-Based Industrial Data Logger — Long-Term Data Archiving for Trend Analysis
- Guide: IoT Predictive Maintenance — A Comprehensive Guide to Deploying IIoT Predictive Maintenance
