Executive Summary:
Integrating video-based AI for leak detection is crucial for maintaining a competitive edge and operational excellence in biopharmaceutical manufacturing. The following describes a project with the objective of instantly finding leaks/drops/streams/puddles on or near bioreactors. The goal is to enable instant countermeasures so that batches can be saved. DeepEyes’ AI solution offers significant time and cost savings while ensuring the highest standards of safety and efficiency, ready for immediate deployment to safeguard production processes and enhance product quality.
There will always be blind spots in monitoring – even in the most sophisticated IIOT environments. The human factor is such a blind spot and leaking bioreactors (upstream process) is often the result of inappropriate handling/human error. Leaks in these critical systems can lead to significant issues, including contamination, production downtime, financial losses, and safety risks. To address these challenges, DeepEyes has been invited by a worldwide top 10 pharma manufacturer to develop an AI-driven video-based real-time leak detection system.
We took up this challenge three years ago and accepted the invitation to put our AI technology to the test.
PROJECT OVERVIEW: AI POWERED REAL-TIME LEAK DETECTION
The project focused on detecting colourless liquid leaks in the form of tiny drops, streams or puddles on the surface of single-use bioreactors of a specified brand. We used standard, off the shelf 4k NDI cameras. Here’ s a detailed look at the project
ASSIGNMENT - GOALS
Objective:
Detect colorless liquids on a transparent plastic or metal surface of single-use bioreactors, or on tubes and/or connectors, valves or water in the form of puddles on the floor in real-time.
Setup:
Installed four 4K, NDI cameras along with flicker-free LED illuminators under the ceiling.
Training Data:
Six hours of video recording served as the training sample.
Processing Resources:
An on-premise video server with specific resources allocated per camera (1 i) 7900X 3.3 GHz, 8 GB DDR4 RAM, 120 Pbps LAN).
Development:
Real-time AI video-based leak detection developed on pre-produced 4K video footage.
Test Phase:
Demonstrated ability on live streams.
Team:
Composed of AI developers, computer vision experts, statisticians, metrologists, electronics and sensor specialists, and backend developers.
Results:
Achieved a 100% recognition rate from 400 control events.
Challenges:
Addressed false positives caused by minute vibrations and human presence.
Privacy Compliance:
Implemented blurring to meet GDPR requirements.
Timeline:
Completed in 7 months, including camera installation, video production, algorithm development, and documentation.
Solution Readiness:
The DeepEyes leak detection solution is ready for deployment, requiring an adaptation phase of 2-4 months.
CONCLUSION
The integration of advanced technologies like video-based AI is essential for maintaining competitive advantage and operational excellence in the biopharmaceutical industry. By leveraging AI for real-time leak detection, companies can safeguard their production processes, enhance product quality, and secure a sustainable future for biopharmaceutical manufacturing. The DeepEyes leak detection solution offers reliable, non-invasive monitoring and stands ready for deployment, offering substantial time and cost savings while ensuring the highest standards of safety and efficiency.
JOIN THE CONVERSATION
How is your organization leveraging AI to enhance manufacturing processes and prevent contamination? Share your experiences and insights in the comments below. Connect with me to stay updated on the latest trends in biopharmaceutical manufacturing and AI-driven innovations.
