Manufacturing

Precision Manufacturer Cuts Unplanned Downtime 40% with €720 Retrofit Sensors

Yalantis developed a non-invasive predictive maintenance system with clip-on IoT sensors, BLE mesh networking, and ML anomaly detection – enabling a legacy CNC fleet to achieve Industry 4.0 connectivity without voiding warranties.

€720

Per-machine retrofit

40%

Downtime reduction

140+

Machines connected

Talk to an Expert

“The client didn’t need us to replace their machines or tear into PLCs. They needed sensors that clip on, connect over BLE mesh, and start detecting bearing wear without anyone touching the control system. That’s what we built.”

– Denys Hukov, Head of IoT Unit at Yalantis

Book a Consultation

Want predictive maintenance without replacing your machines? We build non-invasive retrofit solutions.

€1.8M annual downtime losses on legacy CNC machines that OEMs wanted €8,000+ each to connect

Client:

Precision metal components manufacturer

Headquarters:

Europe

Industry:

Automotive & aerospace supply chain

Fleet size:

140+ CNC machines (60% legacy)

Partnership:

11 months

Team:

Project Manager
Solution Architect
Embedded Engineer (2)
Backend Developer
ML Engineer
Mobile Developer
QA Engineer

Services:

IoT device development
Edge computing
ML/AI
Cloud architecture
Mobile app

A mid-sized European precision metal components manufacturer was bleeding money on unplanned downtime. Their mixed fleet of 140+ CNC machines supplied tier-2 automotive and aerospace clients – customers who don’t tolerate late deliveries.

The problem: 60% of their machines were 10–22 years old. No native connectivity, no diagnostic interfaces, no visibility into what was happening inside the spindles and drives. When a bearing failed or a tool went out of balance, they found out the hard way – after the machine stopped.

Their existing SCADA system only covered newer equipment. For the legacy fleet, maintenance was purely reactive: run to failure, scramble to repair, apologize to customers for missed shipments.

Leadership got quotes from OEM integrators for full PLC integration on the older machines. The numbers came back: €6,500–€12,000 per machine. For 80+ legacy units, that’s over €600,000 – before counting the production loss during installation, which would require extended downtime for each machine.

Worse, deep PLC integration would void warranties on machines still under service contracts and require certified OEM technicians to install.

The client needed a different approach:

Sub-€800 per machine – not €8,000+
Non-invasive: no touching the control bus, no voiding warranties
Installable by in-house maintenance during scheduled stops
ML-based anomaly detection for spindle bearings, tool imbalance, coolant pumps

That’s when they partnered with Yalantis to design a clip-on predictive maintenance system from scratch.

Client:

Precision metal components manufacturer

Headquarters:

Europe

Industry:

Automotive & aerospace supply chain

Fleet size:

140+ CNC machines (60% legacy)

Partnership:

11 months

Team:

Project Manager
Solution Architect
Embedded Engineer (2)
Backend Developer
ML Engineer
Mobile Developer
QA Engineer

Services:

IoT device development
Edge computing
ML/AI
Cloud architecture
Mobile app

Clip-on IoT sensors with edge signal processing, BLE mesh networking, and cloud-based ML anomaly detection

Nordic nRF52833-based sensor module

The clip-on sensor integrates 3-axis accelerometer, current transformer, and temperature sensing. We selected the nRF52833 specifically for its extended temperature range (-40 to +105°C) – critical for sensors mounted on spindle housings near coolant-heated zones where the nRF52832 would drift out of spec.

On-device FFT and envelope analysis

The Cortex-M4 with FPU runs vibration FFT, RMS, and kurtosis calculations on-device. Only extracted feature vectors are transmitted over BLE mesh – not raw 3-axis accelerometer streams – cutting bandwidth by ~95%. Essential for a dense shop floor with 140 nodes.

BLE mesh across metal-heavy RF environment

Metal CNC housings, chip conveyors, and coolant tanks make the 4,200 m² floor an RF nightmare, so coverage came down to antenna engineering. External high-gain antennas on the gateways, tuned transmit power, and placement validated by on-site RF surveys. Direction Finding support on the nRF52833 helped with topology optimization, but the antenna work was the main lever; without it, three gateways wouldn’t be enough for this environment. With it in place, and with the mesh self-healing around obstructions as machines and chip movement shift the RF picture, three was sufficient.

AWS cloud pipeline with ML anomaly detection

Feature vectors flow from gateways to AWS IoT Core, then to Timestream for time-series storage and S3 for model training data. Models were trained on labeled vibration and current signatures gathered during the 12-machine pilot (healthy-baseline runs alongside confirmed bearing-wear, tool-imbalance, and coolant-pump-degradation events) and retrained as the fleet generated more data. The deployed models flag those same three failure modes before they cause downtime.

Web dashboard with IBM Maximo integration

Maintenance planners see real-time machine health scores, trend analysis, and predicted failure windows. CMMS integration automatically generates work orders in IBM Maximo when anomaly thresholds trigger, routing the right technician to the right machine.

Mobile app with AR-assisted installation

Technicians use the iOS/Android app for sensor commissioning and health checks. AR overlay guides optimal sensor placement on each machine type, ensuring consistent mounting positions for accurate vibration analysis across the fleet.

Technical deep dive: why the Nordic nRF52833?

We evaluated multiple Nordic SoCs for the sensor node. Here’s why the nRF52833 won:

1. Extended temperature range (-40 to +105°C)

Critical for sensors mounted on spindle housings and near coolant-heated zones. The nRF52832 is capped at 85°C and would drift out of spec in these locations. This is the same reason Nordic’s Product Guide flags the nRF52833 for industrial environments.

2. BLE mesh + Direction Finding at lower BOM cost than nRF52840

The client doesn’t need USB, QSPI, or 1 MB flash. The nRF52833’s 512 KB Flash / 128 KB RAM comfortably fits the FFT buffers and mesh stack. The nRF52840 adds capabilities the product doesn’t require, raising unit cost unnecessarily.

3. Cortex-M4 with FPU and DSP instructions

Enables running vibration FFT and envelope analysis on-device. Only extracted features go over BLE mesh – not raw 3-axis accelerometer streams – cutting bandwidth by ~95%. Essential on a dense shop floor with 140 nodes.

Why not the nRF5340?

The nRF5340’s dual-core architecture and security features (TrustZone, CryptoCell-312) don’t justify the price premium for a sensor that transmits only feature vectors, not IP. Overkill for this application.

How Yalantis delivered a 140-node predictive maintenance network in 11 months

The project moved through three phases: discovery and hardware prototyping, pilot deployment on 12 machines, and full rollout with ML model tuning. Maintenance technicians installed every sensor themselves during scheduled stops – no outside contractors required.

Step 1

Discovery and hardware prototyping (4 months)

We started by surveying the shop floor: machine types, age distribution, mounting points, temperature ranges, RF environment. The goal was understanding exactly what conditions the sensors would face.

Initial prototypes validated the nRF52833’s on-chip DSP capability for FFT processing. We benchmarked power consumption across duty-cycle profiles until we hit the 2+ year battery life target on 2×AA lithium cells with 90-second measurement intervals and event-driven uploads.

Step 2

Pilot deployment on 12 machines (3 months)

We selected 12 machines representing the fleet’s diversity: different ages, manufacturers, spindle types, and criticality levels. Sensors were installed by in-house technicians during scheduled maintenance windows, guided by the AR-assisted mobile app.

The pilot validated BLE mesh reliability in the metal-heavy RF environment and began training the anomaly detection models. Two bearing failures were predicted during the pilot – both confirmed by vibration analysis before any production impact.

Step 3

Full rollout and ML model tuning (4 months)

With pilot success confirmed, we rolled out sensors across all 140+ machines. Three gateways provided full coverage of the 4,200 m² facility. The IBM Maximo integration went live, automatically generating work orders when ML models detected anomaly thresholds.

The final months focused on ML model tuning – reducing false positives while maintaining sensitivity to genuine bearing wear, tool imbalance, and coolant pump degradation. By month 11, the system was predicting failures with enough lead time for planned maintenance during scheduled stops.

The system went live, delivering measurable results.

€720 per-machine retrofit cost

vs. €6,500–€12,000 quoted by OEM integrators.

40% reduction in unplanned downtime

within 18 months of deployment.

140+ machines connected via BLE mesh

with only 3 gateways covering 4,200 m².

2+ year battery life achieved

on 2×AA lithium cells with 90-second measurement cycles.

Zero warranties voided

non-invasive clip-on installation doesn’t touch control systems.

In-house installation

y maintenance technicians during scheduled stops – no OEM contractors.

IBM Maximo CMMS integration

automatically generates work orders from ML predictions

Key milestones of our R&D collaboration.

Be our next success story. Share details about your project and book a call with us to discuss your goals.

Book a consultation

Certifications:

27001 2022 / 27001 2015 / 13485

Advanced IoT Core Delivery / Advanced RDS Delivery / Advanced Tier Delivery

From medical devices to industrial automation — we deliver complete enterprise solutions with regulatory compliance built-in. Everything under one roof.

Learn more

Our offices