Current Ph.D. dissertations in progress:

    • Agnieszka Mikołajczyk, M.Sc. - PhD topic: Data augmentation and explainability for bias discovery and mitigation in deep learning.
      The purpose of the research conducted within the framework of the thesis is to conduct studies on detecting and minimizing the impact of data bias on the effectiveness of deep neural network models. For this purpose, known and proposed new techniques for augmentation of learning data and methods of explainable artificial intelligence (XAI) are used. (z ang. XAI – explainable artificial intelligence).
    • Arkadiusz Kwasigroch, M.Sc. - PhD topic: Improving the classification quality of deep neural networks by optimizing their structure and two-stage learning process.
      The aim of the research conducted within the framework of the thesis is to develop an effective and efficient method to optimize the structure of the neural network that will ensure high performance and generalization ability of the network, taking into account the characteristics and constraints arising from the analyzed problem. Another goal is to develop a two-stage learning of the developed network (self-supervised + supervised) ensuring effective learning under conditions of deficit of labeled training data. 
    • Mateusz Czyzniewski, M.Sc., Title: Synthesis of state observers for non-linear uncertain systems in application to biological  wastewater treatment plant.

Main goal: The dissertation aims to develop theoretical foundations and applicable methods for state reconstruction (estimation) for non-linear uncertain dynamical systems under unstructured and parametric uncertainty. As an application, one of the critical infrastructure systems, i.e. wastewater treatment plant is considered.

    • Krzysztof Laddach, M.Sc., Neural process modelling and state estimation of dynamic systems.

Main goal: The doctoral dissertation aims to develop the tools that allow obtaining neural (black-box) model with the optimal architecture of selected dynamic systems. The derived neural model will also provide the basis for the developed neural state observer.

    • Maria Ferlin, M.Sc., Title: Synthesis of a reliable deep neural-based system for detection and analysis of cerebral lesions.
      The purpose of the research carried out within the PhD is to develop a neural-based decision support system to diagnose Small Vessel Disease in the brain. The developed system should be transparent, reliable and trustworthy (XAI AI).  The work is being carried out in close cooperation with the Medical University of Gdansk.
    • Zuzanna Klawikowska, M.Sc., Title: Development of methods to synthesise a trustworthy, intelligently supervised control system, robust to process anomalies and faults of a selected class.

The main research goal is to develop methods to synthesize a trustworthy, intelligently supervised control system, robust to process anomalies and faults of a selected class. Realisation of the goal includes the development and final synthesis of  optimization methods, control methods, methods for increasing trust in intelligent systems, methods for detecting anomalies, or methods for predictive maintenance. These issues cover the latest challenges in the field of advanced control systems, machine learning and decision support.

    • Tomasz ujazdowski, M.Sc., Title:  Optimisation of biological processes and management of multiple reactors in a batch wastewater treatment plant
      The research goal of the thesis is designing an operator decision support system of biological processes in one SBR, by solution single-objective and multi-objective optimisation of biological processes in a batch-type WWTP. Another objective is designing a system for efficient and optimal managing the operation of several SBRs, by solving the Task Scheduling issue, for a specific time horizon.

Topics of current research:

Research domain: Systems state and parameters estimation

Key words: estimation, system state reconstruction, soft-sensors, system parameters identification, observers, interval observers, sliding mode observers,  wastewater treatment systems, drinking water distribution systems, power systems.

The research area focuses on developing theoretical foundations and applicable methods within the estimation (system state reconstruction, soft-sensors) of the process variables for control and monitoring purposes in large scale complex systems. Critical infrastructure systems such as wastewater treatment systems, drinking water distribution systems, and electric power systems (including nuclear power plants components) are the primary focus. The methods developed to enable the estimation of the state of the systems under unstructured and parametric uncertainty. The interval observers, sliding mode observers and inverse model parameters identification techniques, also using computational intelligence methods, are a core methodology.

Research domain: Computational intelligence

Keywords: computational intelligence, deep learning, decision support systems, faults detection and diagnosis, explainable artificial intelligence, neural architecture search, medical image analysis, recurrent neural networks.

The research focuses on developing and application of methods of artificial intelligence with a special attention on deep neural networks for decision support, classification, diagnostic and process modelling. 

The examples of research concern developing the methods of explainable AI (XAI) to help to make models more interpretable for system designers and more trustworthy for the end-users, to avoid the effects of bias in the dataset on predictions, to extract new knowledge from datasets.

The research team has experience in employing deep learning methods to analyse 2D and 3D medical images such as skin lesions images, the fundus of the eye or 3D brain scans.

Moreover, the team works on methods on neural architecture search, to optimize the deep neural structures to fit best the neural structure to features of the given process or dataset.

Another important research issue is the development and use of recurrent neural networks for modelling dynamical plants and processes and for early fault and anomalies detection and diagnosis.

Reinforcement learning I zuzy rzeczy

Research domain: Fractional order (FO) modelling and control systems

Keywords: FO modelling, FO identification, FO systems, FO controllers, FOPID  controllers.

Within the research domain, the methods are being developed to exploit and implement the fractional order operators for modelling, identification, and control purposes. Current interests include modelling and identification of plants that are characterized by fractional order dynamics, application of fractional order operators in the synthesis of linear and non-linear control systems with fractional order controllers (FO PID controllers), identification of approximation methods for systems of fractional order which are characterized by a high degree of fidelity, and their effective implementation in various industrial control platforms i.e. PLC and PAC controllers.

Research domain: Modelling and control of biological processes at wastewater treatment systems

Keywords: aeration system, control of biological processes, fractional order calculus, modelling of biological processes, mixed integer nonlinear optimisation problem, optimisation of biological processes;

The research focuses on developing of modelling and controlling of industrial processes, including biological wastewater treatment plant (WWTP).

In particular, scientific research includes:

  1. Modelling biological processes at WWTP.
  2. Modelling of aeration system (blowers, pipes, valves, diffusers)
  3. Control of biological processes at WWTP.
  4. Optimization of biological processes at WWTP.
  5. Fractional order calculus in modelling and control of biological processes at WWTP.

Research domain: Advanced control systems: structures, implementation, simulation and verification with Hardware In the Loop (HIL) technology

Keywords: decentralized and distributed control systems, advanced control algorithms, computer control systems, process fieldbus and Ethernet networks, real-time process automation, HIL technology

The research domain focuses on developing and verification of advanced control systems for large scale and complex industrial plants with the use of the HIL simulation technique. The examples of research focus on the synthesis and effective implementation of the advanced control algorithms working in the multilayer hierarchical control structures based on the typical industrial components i.e. PLC and PAC controllers, process fieldbus, SCADA systems, and infrastructure of the Distributed Control System (DCS). The critical infrastructure systems such as wastewater treatment systems, drinking water distribution systems, and power systems (including nuclear power plants components) are the primary applications. The main considered control algorithms are MPC algorithms with adaptation of model for prediction purposes, steady-state set-point optimization techniques, multiregional fuzzy controllers with switched local controllers, fractional order controllers and issues related to the interaction or cooperation between control algorithms within hierarchical control structure.

Research domain:  Detection, classification and protection of ferromagnetic vessels based on the magnetic fields analysis

Keywords: magnetic fields, magnetic signatures, inverse modelling, parameters identification, optimization methods, fitting and cross-validation phases 

The research in this area concerns the analysis of the magnetic fields. The naval objects’ own magnetic field, identified by its complex magnetic signature which can be treated as disrupting of natural Earth's magnetic field in its own surrounding. The most reliable magnetic data comes from physical measurements of a real object. When the real measuring range is not available, the numerical simulation may be used as a synthetic data source for further analyses, i.e. sophisticated FEM method.

The research concerns the development of the relevant mathematical model (e.g. multi-dipole model, prolate spheroidal harmonic model) based on partial knowledge that allows reconstruction of the magnetic signature in any direction and depth underwater. That approach has practical significance for applications focused on object detection and/or classification and the safety of naval transport by reducing ship’s risk of being detected by naval mines.

The main research areas are mainly related to the developing magnetic model's structures, their parameters identification techniques, specifying a minimum sufficient data set for the fitting phase, and selecting representative cross-validation methods. The classical and computational intelligence methods are under investigation for those purposes.