Sander Bohté (Machine Learning) and Kees Oosterlee (Scientific Computing) have been awarded with funding from NWO’s Indo-Dutch joint research programme for ICT, together with their research partners Indian Institute of Science Bangalore and private partner Shell. Their research project called 'Physics based ICT: the digital twin in pipelines' will focus on using modern AI methods to develop a Digital Twin for Pipeline Transport Networks.
Pipeline monitoring
It is extremely complicated to carefully check tens of kilometers of pipeline, usually located at difficult locations, for example remotely in deep water at a depth of 3 kilometers. Because of that, a modern-day pipeline network typically relies on an industrial Internet of Things (IoT) for monitoring its operations. This research project is inspired by the need for understanding different flow phenomena in the transport of fluids (like liquid and gas) through pipelines, despite the difficult circumstances. In order to clarify this process, a so called ‘digital twin’ will be developed. A digital twin is basically a living model of a physical asset which is continuously calibrated using continuously data collected from the sensors.
Developing a real-time model
In this project, the researchers will develop an accurate and efficient real-time model for detecting leakage in pipelines of the water network in the city of Bangalore and for an industrial environment with multi-phase flow pipeline transport. The intended digital twin is meant to increase the reliability of the pipeline transport network because of its ability to continuously monitor and mitigate degradation and anomalous events. Furthermore, it will assist in making knowledgeable decisions regarding the consequences of possible changes to the network. Particularly the focus is on how the digital twin can be used for the early detection of a leakage from the pipeline with only a very small risk of a false alarm.
Sander Bohté: “I look forward to applying our expertise in deep learning in a whole new research setting. Effective generative neural networks are a recent development in machine learning, and until now these have mainly been applied to relatively static computer vision tasks. The innovation here is to extend these generative models to the temporal dimension using state-of-the-art temporal pattern recognition neural networks.”
It is expected that two PhD students will work on this project, one at the Dutch and one at the Indian side. The collaboration with India will be intensive as the consortium will also work with the sensor data of Bangalore's water network.
