This thesis investigates convergence in the framework of Voronoi-based deployment of a multi-agent system to a convex polytopic multi-dimensional environment. The deployment objective is to drive the system into a stable static configuration which exhibits optimal coverage of the target environment. To this end, the system is subjected to a collection of decentralized control laws steering each agent towards a Chebyshev center of its associated time-varying polytopic Voronoi-neighborhood. In non-degenerate cases, socalled Chebyshev configurations of the multi-agent system achieves the above objective. In these configurations, all agents are at a Chebyshev center of their Voronoi-neighborhood. Proving convergence to the set of Chebyshev configurations is an open research question. This is the most pertinent issue with regards to the framework viability. While such a property is supported by simulations, neither complete formal convergence proofs nor formal characterizations of the equilibria to be achieved exist. This thesis is oriented towards strengthening the theoretical convergence results. For the special case of deployment to one-dimensional environments, we prove convergence to an unique static Chebyshev configuration. Moreover, we highlight connections to discrete time averaging systems and show how the system converges to consensus on the Chebyshev radii. The remaining results apply in the general case of multi-dimensional environments. We introduce a novel undirected interaction graph as a theoretical tool for a deeper understanding of the multi-agent system's functioning. Exploiting this graph, we prove that the set of static configurations are Chebyshev configurations in which all subsets of agents within the same connected component of the interaction graph are in consensus on their Chebyshev radii. Finally we prove convergence to a Chebyshev configuration, with consensus on the Chebyshev radii, provided the interaction graph is connected along the trajectories of the multi-agent system. Throughout the presentation, the theoretical results are motivated and supported by simulations.

In the heavy asset industry, such as oil & gas, offshore personnel need to locate various equipment on the installation on a daily basis for inspection and maintenance purposes. However, locating equipment in such GPS denied environments is very time consuming due to the complexity of the environment and the large amount of equipment. To address this challenge we investigate an alternative approach to study the navigation problem based on visual imagery data instead of current ad-hoc methods where engineering drawings or large CAD models are used to find equipment. In particular, this paper investigates the combination of deep learning and decomposition for the image retrieval problem which is central for visual navigation. A convolutional neural network is first used to extract relevant features from the image database. The database is then decomposed into clusters of visually similar images, where several algorithms have been explored in order to make the clusters as independent as possible. The Bag-of-Words (BoW) approach is then applied on each cluster to build a vocabulary forest. During the searching process the vocabulary forest is exploited to find the most relevant images to the query image. To validate the usefulness of the proposed framework, intensive experiments have been carried out using both standard datasets and images from industrial environments. We show that the suggested approach outperforms the BoW-based image retrieval solutions, both in terms of computing time and accuracy. We also show the applicability of this approach on real industrial scenarios by applying the model on imagery data from offshore oil platforms.

Hydrocarbon production systems generate huge datasets, often with time series going back many years. However, much of the data may be obsolete due to changing reservoir conditions and modification of the asset, and there may be scant data close to optimal operating conditions due to the inadequacy of existing optimization tools. It is widely recognized that data science, artificial intelligence (AI) and machine learning can contribute significantly to the optimization of production operations, and there is a trend towards hybrid AI, which combines data science with traditional physics-based simulators to deliver added value. In our work we show how to make use of physical principles in feature engineering to improve machine learning outcomes. This squeezes additional value from a pure data-based approach while avoiding expensive, time-consuming and often inaccurate simulations. Our toolbox includes energy, mass and force balances; PVT data for production fluids; order-of-magnitude analysis; and dimensional analysis. We illustrate the value of physics guided machine learning with three examples from production optimisation: First example shows a significant improvement in separator operation to achieve environmental limits for safe disposal of produced water using a root-cause analysis to identify bad actors in the production system and recommending operator actions to mitigate oil-in water issues. By physics modeling of key physical processes, such as choke-dispersion and separator efficiency, the predictions were greatly improved. Second example is a data-based VFM using physics-based feature engineering, outperforming a VFM based purely on measured data. Last use-case is a dynamic maximum separator flow capacity calculation that safely allows flow rates above static design limits. We conclude that physics-guided machine learning can add tremendous value to digitalisation rollout across a wide range of production optimisation use cases, and speed up the decision process toward mitigation of production losses in complex industrial phenomena.