Case study · Industrial equipment
Fix machines before they stop the line
Unplanned breakdowns are the most expensive kind of maintenance. This is how we used sensor data and machine learning to detect abnormal behaviour early and raise maintenance work before failure, starting with the assets where downtime costs most.
Project overview
Condition-based and predictive maintenance.
- Industry
- Industrial equipment
- Client
- Confidential
- Solution
- Predictive maintenance
- Platform
- Azure IoT + ML
- Timeline
- 12–20 weeks
- Team
- 3 people
- Signals
- Vibration, temperature, current
- Output
- Alerts and work orders
- Starting scope
- Most critical assets
Expected business outcomes
What the client set out to achieve
Results depend on sensor coverage and failure history.
-
20–40%
Fewer unplanned stoppages
Developing faults detected days or weeks ahead.
Target range for covered assets; depends on failure modes
-
Earlier
Warning of developing faults
Anomalies flagged when behaviour drifts from normal.
Lead time measured per failure mode
-
Tracked
False alarm rate
Alerts reviewed with engineers to keep false alarms low.
Threshold agreed with maintenance
Business challenges
Where the client started
Maintenance is either scheduled or reactive, rarely based on condition.
Maintenance
Time-based schedules and repairs after failure.
Sensors
Some sensors installed, data rarely analysed.
History
Failure records in the maintenance system, often incomplete.
Engineers
Experienced engineers hear problems before systems do.
Cost
A single unplanned stop can cost more than the whole project.
Proposed solution
What the build covers
Simple, explainable models first, more advanced ones where data allows.
-
Sensor ingestion
Sensor data collected and stored with asset context.
-
Anomaly detection
Models learn normal behaviour per asset and flag deviations.
-
Failure prediction
Where failure history exists, models estimate time to failure.
-
Engineer dashboard
Asset health with the signals behind each alert.
-
Work orders
Confirmed alerts create maintenance work orders.
-
Feedback loop
Engineers label alerts, improving the models.
Project requirements
What the client asked us for
The request was to predict failures. We start with anomaly detection on the most critical assets, because it needs little failure history and gives engineers early warnings quickly.
Engineers review every alert at first; their feedback is what makes the models useful.
What the project had to deliver
- 01Early warning on critical assets
- 02Explainable alerts
- 03Work orders from confirmed alerts
- 04Engineers in the loop
Key challenges & considerations
What made it harder, and what we did
-
Few recorded failures to learn from
What we did
The project started with anomaly detection, which needs no failure history.
-
Sensors missing on older equipment
What we did
Sensors were added only where the pilot showed value.
-
Data connectivity in plants
What we did
Data was buffered at the edge, so gaps in plant connectivity lost nothing.
-
Alert fatigue if thresholds are wrong
What we did
Thresholds were tuned with the maintenance engineers to keep false alarms low.
What we watched for
- Start with anomaly detection
- Keep engineers reviewing alerts
- Tune for low false alarms
- Expand only after proving value
Development process
Critical assets first
Value is proven on a small set of assets.
-
Asset selection
Critical assets chosen by downtime cost.
-
Data
Sensor data connected and cleaned.
-
Models
Anomaly models built and tuned with engineers.
-
Alerts and work orders
Alerts connected to the maintenance system.
-
Expand
More assets and failure prediction where history allows.
Technology stack
IoT and ML on Azure
Edge collection, cloud models, maintenance system integration.
IoT
- Azure IoT Hub
- Edge gateways
Machine learning
- Azure Machine Learning
- Anomaly detection models
Data
- Time-series storage
- Fabric
Integration
- Maintenance system / D365 Asset Management
Quality assurance & deployment
Measured against a test set before launch
AI quality was measured, not judged from a demo. A test set of real examples with agreed answers decided whether the system was ready, and the same set was re-run after every change.
- A test set of real examples with agreed answers
- Accuracy thresholds agreed with the business before launch
- Safety checks: refusals, permissions and prompt-injection tests
- A person reviewing low-confidence output
- Cost and response time measured at real volumes
Deployment
- 01Alerts and work ordersAlerts connected to the maintenance system.
- 02ExpandMore assets and failure prediction where history allows.
- 03Pilot, then staged rolloutA small group of real users first, with results reviewed weekly before widening access.
- 04Monitoring and warrantyAccuracy, drift, usage and cost tracked after launch, and six months of free bug fixing.
Final deliverables
Maintenance guided by condition
Engineers see early warnings on critical assets with the signals behind them, and confirmed alerts become planned work before failure.
Every reviewed alert improves the models.
- Asset health monitoring
- Early fault warnings
- Work orders from alerts
- Continuous improvement
Handed over to the client
- 01Production AI serviceDeployed in the client's own cloud tenant
- 02Source code and modelsIn a repository the client can access
- 03Test set and resultsThe evidence behind the launch decision
- 04MonitoringAccuracy, usage and cost
- 05Six-month warrantyFree bug fixing after delivery
Success metrics
How success is measured
| Measure | Outcome | How it is measured |
|---|---|---|
| Fewer unplanned stoppages | 20–40% | Target range for covered assets; depends on failure modes |
| Warning of developing faults | Earlier | Lead time measured per failure mode |
| False alarm rate | Tracked | Threshold agreed with maintenance |
Verified reviews
What clients say about our custom software work
Verified reviews from our clients on the same kind of work, published on Clutch.
-
Custom Software Development
Web-Based Coupon App Development
The work produced by the team was of exceptional quality.
Jon Thies
CTO, Model Rocket Feb 2022 – Mar 2023
-
Custom Software Development
Mobile & Web Platform Development & Design
Silver Scintilla was responsive, respectful, and enjoyable to work with, quickly addressing any questions or requests.
DeShawn Brown
CEO & Founder, Lithios Jun 2021 – Nov 2022
-
Custom Software Development
Web-Based HRM System Development
Silver Scintilla provided the best solution with scalable functionality, personalized features, and robust security.
Suheb Khan
Director, Collaborative Insight Technologies Jul 2020 – Jun 2022
Free discovery call
Have an idea? Let's turn it into AI-powered software.
Book a free discovery call with our experts. Share your idea and we will help you shape the scope, timeline and budget, under NDA.
- Free consultation
- NDA before we talk
- Transparent estimate