Automated Glass Bottle Inspection

This project represents an end-to-end industrial AI system designed for real-time defect detection on high-speed production lines. Over a rapid 40-day sprint, we designed and implemented an automated inspection ecosystem integrating advanced hardware and software to replace manual QC. The system successfully detects and sorts defective bottles in real time, adhering to strict proprietary standards. The client's high-volume production environment faced significant bottlenecks that manual labour could not address — particularly around detection accuracy for both visible surface flaws and internal anomalies. Existing manual inspection processes were slow, inefficient, and prone to human error or fatigue. The solution required high throughput to match conveyor lines without latency while adhering to strict confidential product handling protocols.

We delivered a robust hardware-software fusion tailored for industrial resilience. High-speed vision and sensor fusion combine 80–120 FPS grayscale cameras synchronised with proximity sensors, ultrasonic sensors for thickness analysis, and infrared sensors for thermal stress detection. The AI core is a custom Convolutional Neural Network built on OpenCV, processing images in real time on RTX 3080 infrastructure. An automated mechanical sorting system categorises products into four streams based on inference, paired with an operator dashboard that converts data into actionable metrics. The system drastically reduced manual inspection errors, delivering high detection accuracy compared to previous baselines. It increased throughput by allowing the production line to run at optimised speeds and provided operational visibility through real-time data logging. By combining edge computing, sensor fusion and mechanical automation, the project set a new standard for production efficiency.