Description in which company/unit the thesis will be placed:
The thesis will be carried out at the Department of the Civil and Environmental Engineering at NTNU (Hydraulic engineering group). The AI and ML are becoming more and more popular therefore our group is actively following these trends and tries to find a way how to implement in the different applications related with data processing from hydraulic or hydrological measurements.
However, the data will be obtained from several other institutes and universities that are in collaboration with NTNU: the German Federal Institute of Hydrology (BfG), Statkraft, University of Bologna, University of Ottawa, the Norwegian geotechnical institute (NGI), etc.
Problem Description:
The sediment transport plays a very important rule in the highly exploited rivers. The intensive climate change and the increased possibility of extreme events in combination with inefficient management practices can lead to hazardous events and large financial loss. More specifically, the distribution of the bedload through alluvial streams contributes to shaping of the river morphology; thus, it is one of the key components in many branches of river management. Therefore, the sediment transport data is a fundamental requirement for proper management of engineering practices in complex river systems.
However, the conventional sediment transport measurements are seldom available and notoriously hard to perform.
An alternative to acquire a better data is the application of modern, indirect, non-intrusive methods, such as the hydro-acoustic instruments- The acoustic Doppler current profilers (ADCP) are the most widely used hydro acoustic instruments and commercially available. They are not disturbing the riverbed and can continuously measure various hydraulic and sediment variables. With a proper application, they can efectivly estimate the water velocity, water discharge, water depth, but also the suspended sediment concentration, and the bedload transport rate; all at the same time. The ADCPs measure indirect variables (e.g., the Doppler effect and the backscattering strength of the reflected signal), and as such require special pre-processing of these variables and development of models that translate the acoustic variables in hydraulic or/and sediment data.
Recently several institues (e.g., NTNU, BfG, etc) are deploying ADCPs to improve and advance the sediment measurments. Many campaigns have been performed in the field and the laboratory resulting developing several semi-smpirical and analytical models to estiamte the sediemnt transport. However, the complexity and the high nonlinearity of the sediment transport, often lead to highly uncertain results. This is even more exaggerated with the bedload transport due to a larger span of particle sizes, and mobility in different modes (e.g., jumping, rolling, vibrating). Therefore, these models tend to be instrument-related and site-specific, which complicates their usage and applicability.
The recent advancements in the machine learning and artificial intelligence offers an opportunity to resolve above-mentioned issues. Using the datasets obtained both, by the conventional techniques (i.e., bedload samplers) and the ADCP at the same time and place, the ML techniques can help to develop a smarter, unique, and robust model for estimating the bedload transport. This model should incorporate datasets from different rivers and bedload transport conditions, in addition to different types of ADCPs. The model should give an accurate estimate of the bedload transport in similar hydraulic and sediment conditions independently of the location. This approach should decrease the uncertainty due to the spatial and temporal variability of the bedload and facilitate the utilization of the ADCP collected data for bedload transport measurements.
Thesis Description:
Protentional thesis title: Bedload transport estimation using ADCP data by deploying machine learning operators.
The master thesis will consist of developing a smart model for estimating the bedload transport data based on the ADCP measured variables, as described in the previous section. The Smart ADCP bedload Model should be a based-on machine learning algorithm and AI principles, bult up on the study/work developed by NTNU and BfG as part of the ADCPx project for developing new modern bedload monitoring methods. The master thesis shall consist of the following steps:
- Literature review and understanding of the problem.
- Reconnaissance of the problem, get to know the working principle of the ADCPs.
- Collecting and classification of the datasets (most of this area already available at the NTNU storage).
- Selection of the framework and identifying possible ML/AI algorithms.
There are several options for approaching and constructing the framework:
For example, the ML regression opperators (Neural Networks, Neuro Fuzzy Networks, Gaussian Process Regressions, etc.) are good candidates for developing a supervised ML model.
Data Description:
The inputs of the AI/ML model shall be derived from the ADCP measured data and the targets should be taken form the physically (i.e., conventual) measured samples. The spatial, temporal and physics-based features should be derived mainly from the ADCP data but also from the eventual axuilary emasurments (e.g., water discharge, GPS, etc).
The ADCP data sets consits of more than 200 000 meassured ensambles dividedd in approximetly 200 sets form 20 different meassurment points. Each ensamble consists of backscattering strength, apparent velocity water velocity, depth and other variables that the ADCP meassures instantaniously. The targets are given as 3 measurments per each set of ensambles, giving aproximetly 600 measurments.
Pleae not that this only the availnle set at NTNU. The student and the supervisors should doulble it by obtaitning tther sets of data from the institutes and universities that has been collaborating with NTNU on similar sediment – related projects.
The data at NTNU is already preprocessed, but the freshly obtained data must be processed following the same steps. The preprocessing of the data will be done in collaboration with the supervisors or other students working on similar assignments.
The German Federal Institute of Hydrology (BfG), Statkraft, University of Bologna, University of Ottawa are some of the institutes that own such data and are willing to collaborate. This data is usually publicly available thus no need of extra security. However, If necessary, written agreements could be made between NTNU and some of the companies /institutes (e.g., Statkraft) for further use of their data.
Challenges (business and/or research):
As mentioned before the sediment data is rare and unavailable, but more and more required for an efficient operation management of the hydropower plants, flood management, ship-navigation, etc. Thus, this method/application would be very attractive, mostly for governmental organizations but also private companies working in the hydropower or river transport sector. Few examples of agencies dealing with river sedimentation are listed here but many others exist :
- USGS - United State Geological Survey (hydroacoustics section)
https://hydroacoustics.usgs.gov/
- NVE - The Norwegian Water Resources and Energy Directorate
- BfG - Federal Institute for Water Science (Germany)
https://www.bafg.de/EN/Home/homepage_en_node.html
- WSV - Waterways and Shipping Administration (Germany)
- AIPO - Po River Interregional Agency (Italy)
https://www.agenziapo.it/content/english-presentation
- AdBPO - Po River Basin Authority (Italy)
Moreover, the same approach could be adopted for other measurements (e.g., suspended sediments) with the ADCP. Most likely the same framework could be reapplied for similar measurement techniques and other hydraulic or sediment related variables. It is especially relevant for many engineering methods that are still relying on the traditional approaches and accept a large uncertainty. This study should be a springboard and inspiration of a possible improvement of the data and reducing the uncertainty in the hydraulic, hydrological or sediment measurements.