We engaged Viston to reduce visual defects on our brake assembly line and prevent unplanned downtime across our CNC manufacturing cells.

Business Challenge

A 2.2% defect rate on a high-volume brake assembly line and six unplanned stoppages per quarter were causing missed On-Time In-Full (OTIF) delivery targets.

Our Approach and Solution

Viston deployed an edge-based computer vision solution on NVIDIA Jetson devices, trained YOLOv8 models using our labeled defect data, and integrated the system with PLC signals for real-time inspection.

AI-Powered Quality Inspection

Implemented edge computer vision for automated defect detection on the brake assembly line, enabling continuous inspection and reducing reliance on manual quality checks.

Predictive Maintenance

Collected vibration and temperature sensor data, developed anomaly detection models managed with MLflow, and integrated predictive maintenance alerts directly into our CMMS.

AI Applications and Benefits Delivered

Computer Vision for Defect Detection

Real-time detection of manufacturing defects using AI-powered vision models with continuous model drift monitoring.

Predictive Maintenance

Early warning alerts based on machine condition monitoring to prevent unexpected equipment failures and downtime.

Business Benefits

• Reduced scrap and rework
• Improved Overall Equipment Effectiveness (OEE)
• Fewer unplanned production stoppages
• Increased manufacturing efficiency

Cost of Implementation

EUR 110,000, including edge hardware, data labeling, MLOps implementation, and operational technology (OT) security.

Time to Implement

16 weeks for deployment across two production lines and three CNC cells.

Tools and Technologies Used

• AWS SageMaker
• NVIDIA Jetson Xavier NX
• OpenCV
• YOLOv8
• Apache Kafka
• Grafana
• Prometheus
• MLflow
• OPC UA Connectors
• HashiCorp Vault
• Amazon S3 with VPC Endpoints

Quantitative Outcomes

• 47% reduction in defect rate (2.2% → 1.16%)
• 32% reduction in unplanned downtime
• 11% increase in production throughput
• Full investment payback achieved within 9 months

Key Performance Indicators (KPIs) Tracked

• Overall Equipment Effectiveness (OEE)
• False Positive (FP) and False Negative (FN) rates for computer vision
• Mean Time Between Failures (MTBF)
• Mean Time to Repair (MTTR)
• Scrap rate
• Rework hours
• Alert precision and recall

Pre- and Post-Implementation Metrics

• Overall Equipment Effectiveness (OEE): 73% → 80%
• Scrap Rate: 3.1% → 1.9%
• Unplanned Stops per Quarter: 6 → 4
• Vision Model Precision/Recall: 0.94 / 0.91 after calibration

Stakeholder Quotes

Plant Manager

“The inline vision system removed subjectivity and made quality scalable. Operators trust it because false alarms are rare.”

Regulatory and Compliance Considerations

• GDPR compliance for operator-facing cameras
• ISO 27001 security controls
• TISAX compliance for automotive information security
• Network segmentation for Operational Technology (OT) systems