Keynote Speakers

University of Lorraine, France Benefits of Petri nets for systems modeling and probabilistic assessment in reliability engineering


Assessment of systems dependability belongs to important tasks in many engineering fields. Such an assessment can be done using various mathematical methodologies depending on the mathematical representation of the system. In this lecture, we will focus on Petri nets and especially in Stochastic Petri nets which are one of the most common representations of functional and dysfunctional behaviour of system and its components. They are able to take into account stochastic processes of failures and maintenance, reconfiguration of dynamic systems due to failures, redundancy. We will present the dynamic behavior of Stochastic Petri nets, their performance measures from which dependability measures can be obtained. To assess these measures, two approaches can be considered: an analytical approach based on Markov chain theory, or an approach based on MonteCarlo simulation. Both approaches are discussed and compared. Some applications to real industrial systems will also be presented. 


VSB—Technical University of Ostrava, Czech Republic Maintenance optimization of complex multicomponent systems


A complex multicomponent system consists of a finite number of nonidentical components that can be realized as maintained components with different maintenance modes, for example nonrepairable components, repairable components with corrective maintenance, repairable components with latent failures that are identified by means of preventive maintenance, component with preventive maintenance policy in which the component is restored (either repaired or renewed), etc. Arbitrary components are considered without any restrictions on the form of the probability distribution assigned to time to failure and repair duration, i.e. ageing components are allowed. 


University of Ljubljana, Faculty of Electrical Engineering, Ljubljana, Slovenia Selfsufficient electric energy supply at home


Electric energy supply at home is an issue, which offers more solutions than years ago. The objective is to present an evaluation of selfsufficient electric energy supply at home. The method bases on the solar power plant as the primary source of power generation. The storage of electric energy with direct current electric battery is considered as the secondary power source, knowing that night hours are without the primary power generation. The real case time dependent home consumption is determined and the real case of solar power plant generation is considered. The size of solar power plant and the size of the battery are optimised based on the consumption and based on 100 % selfsufficient power system. One parameter optimisation bases on minimisation of costs for such a system. The model includes realistic yearly time dependent home consumption curve, realistic yearly time dependent solar power plant generation curve and realistic yearly time dependent curve of state of charge of battery, which is a function of solar power and home consumption. The time resolution of this model can vary based on density of data points of solar power generation and density of points of determined power consumption. Results include comparison of costs related with different size of solar power plant and with different size of battery. The most important result is a combination of the size of solar power plant and the size of battery, which are both related with the smallest overall costs. Consideration of different years gives different results due to different sun irradiation through the hours of the year due to weather changes. Consideration of different locations gives different results due to the same reason. Results show relatively large cost of all cases, which exceeds the cost of buying electrical energy from actual provider at the current conditions. 


Masaryk University, Faculty of Medicine, Czech Republic Datadriven decision making in practice: Experiences in academia and government


Datadriven decisionmaking is nowadays one of the domains that have huge potential. This is mainly because data, information systems and user interactions on the Internet are constantly increasing.
However, it is crucial to handle this phenomenon correctly. The use of proven methodologies, the choice of effective and secure technologies and the appropriate involvement of the human factor seem to be essential. The workshop will not only provide the theoretical background of the complex data mining process and its application in real life. Still it will also present selected domains from medical education and Czech healthcare, where data plays a key role, using reallife examples.



Nazarbayev University, Kazakhstan MultiDiagnosis Cough Classification Evaluation


The sound classification is an open problem when it comes to classification. In particular and with the recent outbreak of COVID19 a large amount of research has been invested in cough classification as a method of early detection and subsequent prevention. However usually the methodologies available are considering specific approaches such as very large datasets, data augmentation or even combination of breathing with coughs in order to increase the classification accuracy. In this work we study the classification of coughs into several diagnostic categories as a function of volume of the dataset and size of the data samples. For this purpose we use a dataset collected using our developed mobile application, prepare several datasets and evaluate different classifiers. First we assume that we do not have enough data for an endtoend deep learning approach. Second we also consider that the variety might be low. Finally we also assume that the data is unbalanced. In order to deal with these problems we propose a study on using fast and shallow classifiers, data manipulation such as sample adaptive length and sample overlapping. As a result we determine that while it is overall the most accurate to process the sounds as whole, sampling them into samples with overlapping segments allows to recover most of the information from the whole samples and obtains similar accuracy.

