AI-driven Defect Detection System for a Leading Paper Mill
ThirdEye Data successfully developed and deployed an AI-driven surface inspection system for a leading paper mill to automate defect detection in paper rolls during high-speed production. The solution leverages computer vision, edge computing, and deep learning to detect wrinkles, tears, and texture issues in real-time—improving production quality, reducing customer claims, and cutting financial losses due to undetected defects.
THE CUSTOMER
BUSINESS GOALS OR CHALLENGES
Business Goals
- Automate the surface inspection process using AI to minimize manual errors.
- Detect surface defects before jumbo roll formation.
- Reduce compensation payouts from undetected defects.
- Improve overall product quality and customer satisfaction.
- Avoid the high cost of replacing legacy systems like PASTEX, ABB, and VALMET.
Understanding the Challenges:
- Manual inspection allows only a 1-hour window before defects are irreversible.
- Inconsistent detection due to human error.
- High compensation costs due to defective deliveries.
- Existing inspection systems are expensive and inflexible.
- Limited insights into defect patterns and root causes.
Prerequisites and Preconditions:
The system had to integrate with the high-speed production line, operate in industrial environments, detect a wide variety of defects in real time, and be scalable for future plant-wide deployment.
- Sample video data of production line defects.
- Access to the paper production environment for camera installation and testing.
- Agreement on defect types and classification priorities.
- Availability of production-specific training data.
- Hardware deployment feasibility checks (camera placement, lighting, network).
THE SOLUTION
ThirdEye developed a custom AI-based computer vision solution using high-speed industrial cameras, deep learning algorithms, and edge computing devices. The system detects and classifies defects in real-time and generates actionable alerts for operators, minimizing the chances of defective rolls reaching the end customer. We proposed and agreed on a 3-phase implementation—starting with a PoC, followed by pilot deployment and then full-scale rollout.
High-level Solution Approach
-
Camera & Sensor Setup
Installed high-resolution line-scan cameras with synchronized LED lighting and IR sensors to continuously capture clear images of fast-moving paper rolls. -
Edge Computing for Real-Time Detection
Deployed NVIDIA Jetson-powered edge devices on-site to process images locally and detect surface defects in real time with minimal latency. -
AI Model Development
Trained deep learning models (CNNs) using thousands of labeled defect images to recognize various paper surface defects like streaks, wrinkles, patches, and more. -
Automated Alerts & Dashboard
Built a real-time monitoring dashboard that displays defect type, location, and severity. Integrated alert system notifies operators instantly for quick intervention. -
System Integration
Integrated defect data with customer’s existing MES and quality control systems to maintain traceability and streamline reporting. -
Scalable Architecture
Designed the system to be modular and scalable, allowing easy expansion to other production lines or addition of new defect types in the future.
Technology Stack:
-
Hardware: Line-scan cameras, LED lighting, infrared sensors, edge GPUs
-
Software: OpenCV, PyTorch/TensorFlow, anomaly detection models, real-time dashboards
-
Platform: Edge computing via NVIDIA Jetson, integrated with existing factory systems
VALUE CREATED
The system has been deployed 3 months back, we are sill in the phase of estimating the final ROI, Here are a few ROI calculations received from the customer based on the last 3 months performance:
- Around 93% accuracy in detecting multiple surface defect types.
- Over 36% reduction in product rejections due to early detection.
- Approximately 25% improvement in overall product quality metrics.
- 25% reduction in customer returns and rework associated with defective alloy wheels.
- As per the initial performance, we are expecting ₹1.8 crore annual savings in compensation costs within 7 months of the full-scale deployment
Transforming Enterprises with
Data & AI Services & Solutions.
ThirdEye delivers Data and AI services & solutions for enterprises worldwide by
leveraging state-of-the-art Data & AI technologies.
