The idea of a digital twin is appealing and we should certainly develop it further, but in doing so we should also be aware of the practical challenges and the limitations of such models; this became apparent during the third 2019 meeting of the Hydroinformatics Platform. Experiences at drinking water utilities in building digital twins, data mining, and insight into the added value of working with several models all provided valuable material for discussion.
During its third meeting of 2019, held on 19 September, the Hydroinformatics Platform extensively discussed the practical aspects of building a digital twin: a digital replica of a physical entity, such as a distribution network or a treatment installation.
Real-time pipe network modelling at drinking water utilities
Joeri Legierse talked about the experience of Evides in developing a digital twin, which incorporates actual measurement data into a hydraulic model. Evides’s objectives with its digital twin included more water loss detection, tracing the sources of water-quality anomalies, detection of wrong valve positions, and any possible future forecasts; the water utility assessed which of these objectives should be achieved automatically and which manually. Evides also identified the relevant data streams, and listed the technical possibilities of the hydraulic software it uses. The experience showed that building the intended digital twin would require a considerable investment. Evides therefore shifted its focus to methods to calibrate the hydraulic models faster and better.
Waternet has been busy for some time building digital twins, said Paul Stroet. He talked about recent developments and insights. Waternet, too, has noted the utmost importance of organising the data appropriately before investing in a digital twin. Waternet is now giving priority to making the existing data usable, and creating a structure in which the data can be smoothly exchanged internally.
Use of several models
Igor Nikolic of TU Delft gave a presentation on dealing with uncertainty in various forms of modelling. He emphasised that modelling is a social process and, under the motto: ‘all models are wrong, some are useful’, warned against attributing too much importance to models. He is not charmed by the term ‘digital twin’, and thinks that the best approach consists of the parallel application of several models, so-called ‘multi-model ecology’.
Henk-Jan van Alphen (KWR) presented the results of three data mining pilots at the Oasen, PWN and Waterbedrijf Groningen drinking water utilities. He focused on the success/failure factors in the process.
The success factors include onsite implementation, a clear mission, joint interpretation of interim results and, when necessary, refinement of the research plan, rapid response to questions, and an engaged client.
The failure factors include insufficient data quality or data that do not dovetail well with the research needs, and data findability and accessibility. Van Alphen made the following recommendations for data mining projects:
- Improve data management;
- Develop a vision on data;
- Work onsite;
- Create multidisciplinary teams;
- Differentiate between research and implementation projects.
Challenges and limitations
The three presentations provided ample material for extensive discussions during the closing network lunch. The idea of a digital twin is appealing and we should certainly develop it further, but in doing so we should also be aware of the practical challenges and the limitations of such models.