Quick Facts: Built an edge-based computer vision system to detect conveyor belt failures in real time, enabling 25% lower inspection costs and 10% reduction in annual maintenance fees.

About the Client

A major forestry company operating large-scale industrial facilities, where machinery uptime is critical for productivity and safety.

The Challenge

The client needed to prevent failures in the conveyor belt feeding wood logs into a chipper, a machine that processes logs at high speed. This process was previously supervised manually via video monitoring, with operators using a joystick to pause, reverse, or resume operations.

The challenge was to automate detection to reduce reliance on 24/7 human monitoring and avoid costly downtime or equipment damage.

Marvik’s Approach

We developed a proof of concept using computer vision, deployed on an NVIDIA Jetson Nano edge device, capable of analyzing live video streams from the conveyor belt in real time.

Key steps included:

• Stabilizing and filtering the video feed to ensure model accuracy.
  
  
• Applying classification and object detection tasks to detect potential failures.
  
  
• Filtering ML model outputs to reduce false positives.
  
  
• Integrating the software and hardware solution to send signals directly to the plant’s control system for immediate action.
  

The system was built with Python, Golang, React, OpenCV, TensorFlow, Keras, PyTorch, and NVIDIA Jetson Nano TX2.

The Results & Impact

• 25% reduction in inspection costs.
  
  
• 10% reduction in annual maintenance fees.
  
  
• Automated detection of systematic failures that would otherwise require continuous human supervision.
  

According to McKinsey & Company, AI-based predictive maintenance can increase equipment availability by up to 20%, benefits our client can now capture with reduced costs and increased uptime.

Why This Matters

By combining computer vision, edge computing, and direct integration with plant controls, we transformed a monotonous manual monitoring task into an automated, predictive safeguard, protecting equipment, reducing costs, and improving operational resilience.