Industrial Sensing / Flow Measurement
Ultrasonic Flow Monitoring
CLIENT — Industrial flow-monitoring product manufacturer
A three-year partnership developing non-intrusive ultrasonic flow-rate monitoring, where data analysis showed state-of-the-art accuracy was achievable by fusing ultrasonic and accelerometer data with neural networks.
The challenge
The client's non-intrusive ultrasonic technology aimed to match the accuracy of in-flow measurement devices. Following organisational change, they needed a partner who could pick up the existing codebase and extend the work across both engineering and data science.
Our approach
We began with product-strategy workshops and a code review, then expanded into full-stack development and data-science research: real-time MQTT data logging with a web UI, ultrasonic flow-rate research, and specialist analysis of acoustic signals using machine learning. Research ran alongside the National Engineering Laboratory (now part of TÜV SÜD), with extensive on-site testing to iteratively design and schedule experiments.
Results
Our analysis demonstrated that state-of-the-art measurement accuracy could be achieved by combining the ultrasonic module data with accelerometer readings — neural-network models trained on accelerometer data were especially promising. The work later extended into distributed acoustic sensing analytics.
Technologies
- Docker
- Django
- MQTT
- Neural networks
- Machine learning
- Signal processing
Fusing ultrasonic and accelerometer data with neural networks to reach in-flow accuracy from outside the pipe.