Industrial IoT Case Study | Kawach Technology

Client Overview

Nordholt Manufacturing GmbH

Nordholt Manufacturing runs three precision-component factories across Germany, and for years its maintenance philosophy was the industry default: service every machine on a fixed calendar interval, regardless of how much actual wear that specific machine had accumulated. It's a reasonable-sounding approach until you realize it means perfectly healthy machines get serviced unnecessarily while othe...

Industry

Precision Component Manufacturing · Manufacturing Software Development

Business Size

3 factories, 900+ employees

Location

Stuttgart, Germany · Software development in Germany

Business Model

B2B Industrial Manufacturing

Project Duration

12 months

Existing Challenges

Maintenance was scheduled on fixed calendar intervals regardless of actual machine wear.
Unplanned breakdowns halted production lines with no early warning.
Each factory kept its own paper-based maintenance logs, with no cross-site pattern detection.
Plant managers had no real-time dashboard of machine health across sites.
Machine sensor data, where it existed, was not centrally collected or analyzed.

The Challenge

What We Were Up Against

Nordholt Manufacturing runs three precision-component factories across Germany, and for years its maintenance philosophy was the industry default: service every machine on a fixed calendar interval, regardless of how much actual wear that specific machine had accumulated. It's a reasonable-sounding approach until you realize it means perfectly healthy machines get serviced unnecessarily while others fail well before their scheduled maintenance date.

Those unplanned failures were the real cost. A single unannounced breakdown could halt an entire production line with no warning, and because each factory kept its own paper-based maintenance logs, there was no way to spot patterns — a machine that failed the same way twice, three months apart, at two different sites, looked like two unrelated incidents rather than one identifiable pattern.

Plant managers had no real-time way to check machine health beyond physically walking the floor, and where sensor data existed at all, it sat locally on individual machine controllers rather than being collected anywhere central. Making this harder still, any system touching workforce shift data in Germany has to take GDPR seriously from day one — this wasn't a system we could bolt privacy onto after the fact.

Our Solution

How We Built It

We started on the factory floor, literally — walking each production line with maintenance staff to understand which machines were the highest-value candidates for sensor retrofitting, since instrumenting all 600+ machines on day one wasn't realistic or necessary. Vibration, temperature, and power-draw sensors were fitted to priority machines first, streaming data over MQTT to a central ingestion pipeline.

That data feeds a predictive maintenance model trained specifically on Nordholt's own historical failure patterns rather than a generic industry model, since failure signatures genuinely differ by machine make, age, and workload. The model flags machines showing early signs of the wear patterns that have historically preceded failure, giving maintenance teams days or weeks of lead time instead of a floor manager discovering a stalled line.

Because any system touching German workforce data has to take GDPR seriously, we designed the shift-scheduling and maintenance-assignment features to anonymize or aggregate any employee-linked data at the point of collection wherever the analysis didn't actually require an individual identity — machine health monitoring almost never needs to know which specific technician was on shift, for instance.

We piloted the sensor retrofit and predictive model on a single production line for several weeks, comparing its predictions against what actually happened, before expanding to the other lines and eventually all three factories. That validation period mattered — it caught a few false-positive patterns in the early model that would have eroded trust with the maintenance team if they'd shown up during full rollout instead.

Key Modules Delivered

IoT Sensor Integration

Vibration, temperature, and power-draw sensors on priority machines, streaming data via MQTT.

Predictive Maintenance Engine

A machine-learning model trained on Nordholt's own historical failure data to flag early warning signs.

Central Machine Health Dashboard

Real-time machine status across all 3 factories in one view.

Maintenance Scheduling & Work Orders

Digital work orders generated automatically from predictive alerts.

Cross-Factory Reporting

Pattern detection across sites to catch recurring failure types before they repeat.

Alerting & Escalation

Automated alerts to the right maintenance team as soon as a machine shows early warning signs.

Goals & Objectives

What Success Looked Like

Predict Failures Before They Happen

Move from calendar-based maintenance to data-driven, predictive scheduling.

Centralize Machine Health Data

Bring sensor data from all 3 factories into one central platform.

Reduce Unplanned Downtime

Give maintenance teams early warning instead of discovering failures after the fact.

Standardize Maintenance Across Factories

Replace inconsistent paper logs with one digital system across all sites.

Features Developed

What We Built

Predictive Failure Alerts

Early warning days or weeks before a machine would historically fail.

Central Machine Health Dashboard

Real-time visibility into every instrumented machine across 3 factories.

Digital Work Orders

Auto-generated maintenance work orders from predictive alerts.

Cross-Factory Reporting

Pattern detection across sites to catch recurring failure types.

Sensor Retrofit Kit

Non-invasive sensor installation on existing machines.

GDPR-Compliant Data Pipeline

Employee-linked data anonymized wherever identity isn't required.

Technology Stack

Built With the Right Tools

We selected every technology based on this project's real requirements: compliance obligations, scalability needs, and long-term maintainability. No trend-chasing, only battle-tested solutions.

Data Ingestion

MQTT Azure IoT Hub

ML & Analytics

Python TensorFlow InfluxDB

Frontend

React D3.js

Infrastructure

Azure (EU region) on-prem edge gateways

Development Process

How We Delivered It

Agile delivery with regular demos and continuous deployment. Full transparency at every stage.

Total Timeline

12 months

Started → Ongoing

1

01

Factory Floor Assessment

Walked each production line with maintenance staff to identify the highest-value machines for sensor retrofitting.

2

02

Sensor Retrofit Planning

Fitted vibration, temperature, and power-draw sensors to priority machines without halting production.

3

03

Data Pipeline Build

Built the MQTT-to-cloud ingestion pipeline feeding a central time-series database.

4

04

Predictive Model Training

Trained the failure-prediction model on Nordholt's own historical maintenance and failure records.

5

05

Pilot on One Production Line

Validated model predictions against real outcomes on a single line for several weeks before expanding.

6

06

Rollout Across 3 Factories

Expanded sensor coverage and the predictive model to all production lines at all 3 sites.

Security & Compliance

Built for the Strictest Standards

GDPR-Compliant Data Handling

Any employee-linked shift data is anonymized or aggregated wherever individual identity isn't required for the analysis.

EU Data Residency

All machine and operational data is stored on Azure infrastructure within the EU.

Industrial Data Security

Sensor and network architecture aligned with IEC 62443 industrial security principles.

Results / KPIs

Measurable Impact

Numbers measured at 6 months post-launch, independently verified by the client's operations team.

-47%

Unplanned Downtime

-22%

Maintenance Cost

-35%

Mean Time to Repair

600+ machines

Sensor Coverage

Before vs. After

Before	After
Fixed-interval maintenance regardless of wear	Predictive, data-driven maintenance scheduling
No cross-factory machine health visibility	Central dashboard covering all 3 sites
Paper-based maintenance logs	Digital work orders with full history
Failures discovered after the line stopped	Early-warning alerts days or weeks ahead

"

We'd been maintaining machines on the calendar for decades because that's simply how it's always been done in manufacturing. Seeing real predictive alerts catch a bearing failure two weeks before it would have stopped a line — that changed how our maintenance team thinks about their job. Downtime is down nearly half, and just as importantly, our technicians trust the alerts now.

KL

Klaus Reinholt

Head of Plant Operations, Nordholt Manufacturing GmbH

★★★★★

Key Achievements

Why This Project Matters

Beyond the numbers: what this project changed day-to-day for Nordholt Manufacturing GmbH and the people who rely on what we built.

600+ Machines Instrumented

Sensor coverage now spans priority machines across all 3 factories.

Unplanned Downtime Nearly Halved

Predictive alerts cut unplanned downtime by 47% within the first full year.

Maintenance Costs Down Without Added Headcount

More targeted, data-driven maintenance reduced costs while using the existing maintenance team.

FAQ

Common Questions

Have more questions? Book a call with our team.

Did you need to replace existing machines to add sensors?

No — the sensor retrofit kit was designed to attach to existing machines non-invasively, so no equipment replacement was needed.

How is GDPR handled for shift and machine-operator data?

Wherever the maintenance or predictive analysis doesn't actually require knowing which individual employee was involved, that data is anonymized or aggregated at the point of collection rather than stored in identifiable form.

How accurate are the predictive maintenance alerts?

We validated the model against real outcomes on a pilot production line for several weeks before wider rollout, and continue to tune it against Nordholt's own ongoing maintenance data rather than relying on a static, generic model.

Did the sensor installation disrupt production?

Installation was scheduled around existing planned maintenance windows specifically to avoid any additional production disruption.

Related Case Studies

Predictive Maintenance: How Nordholt Manufacturing Cut Unplanned Downtime by 47% with Industrial IoT