Motivation In the area of Industry 4.0, handling vast amounts of data that provide insights into various processes significantly increases. Often, data from monitoring systems exhibit unique characteristics, such as impulsive behaviour. Traditional methods for identifying anomalies and detecting local damage in machines may be inadequate and produce unreliable results. Consequently, dedicated approaches tailored to the specific nature of the analysed data are developed for modern industrial data analysis. This ensures that information derived can be effectively utilized. In this course, some new achievements in the area of statistical and time series methods for the analysis of industrial data with impulsive, heavy-tailed behaviour are proposed. The main focus will be on vibration-based analysis for the local damage detection of machines, as in heavy machinery exhibiting impulsive dynamic behaviour. Presented approaches and new techniques are general tools for identifying anomalies in data with non-Gaussian behaviour. All proposed methods were already applied to working industrial test cases as well as published in some scientific journals, focusing expressively on the analysis of signal processes, especially in mechanical systems. Training goals The course aims improving the capability of students to analyse the dynamic signals retrieved from machines in operation, under machine condition monitoring, for monitoring, prognosis and diagnosis purposes as well as to validate models, in case of data driven approaches to design and digital twin modelling. Particularly it provides: ¿ Presentation of real-world examples from selected areas of industry; ¿ Discussion about the problem of non-Gaussian data in industrial applications; ¿ Description of the classical approaches for anomaly detection on the basis of the selected industrial problem; ¿ Introduction to new statistical methods for analysing and forecasting of non-Gaussian impulsive data; ¿ Presentation of recent achievements in the time series analysis for non-Gaussian data; ¿ Application of the presented methodology to data in time and time-frequency domain representations. Methodology The lectures will include a first theoretical part dealing with gaussian and non-gaussian signals analysis and approaches applied in the literature and novel ones proposed in this context, and then the application to some test cases coming from real industrial practice will be shown, according to Prof. Wylomanska’s experience as member of the Hugo Steinhaus Center for Stochastic Processes, and to her research activity within heavy-tailed distributed time series, stochastic modelling, and statistical analysis of real data, with a focus on mining industry and indoor air quality, as well as to collaborations with industrial companies such as KGHM and Nokia. Didactic material covers the whole presentations along with analyses of simulated and real data. Additional resources It is meant to provide students with additional material to help them understanding the subject matter including: ¿ Research articles discussing the problem of analysing industrial data in cases of non-Gaussian signals. ¿ Simulated data (imitating vibration signals) with a non-Gaussian distribution: simulated non-Gaussian background noise (at different levels of non-Gaussianity) and the simulated signal of interest with different amplitudes. ¿ Simulated data (imitating the health index) with a non-Gaussian distribution: Gaussian-distributed HI and non-Gaussian distributed HI (with different levels of non-Gaussianity). Assessment If required by the Institution, participants will receive a quiz to assess their understanding of the items covered by this course.

-

In this course, some new achievements in the area of statistical and time series methods for the analysis of industrial data with impulsive, heavy-tailed behaviour are proposed. The main focus will be on vibration-based analysis for the local damage detection of machines, as in heavy machinery exhibiting impulsive dynamic behaviour. Presented approaches and new techniques are general tools for identifying anomalies in data with non-Gaussian behaviour. Guest lecturer: - Prof. Ing. Agnieszka Wylomanska: Professor at the Faculty of Pure and Applied Mathematics, Wroclaw University of Science and Technology (Partner University). She received her M.Sc. degree in Financial and Insurance Mathematics from the Institute of Mathematics and Computer Science at Wroclaw University of Technology (Wroclaw Tech) and a Ph.D. degree in Mathematics from Wroclaw Tech in 2006. In 2015, she achieved a D.Sc. degree in mining and geology from the Faculty of Geoengineering, Mining, and Geology at Wroclaw Tech. Currently, she holds the position of Professor at Wroclaw Tech and is a member of the Hugo Steinhaus Center for Stochastic Processes. Her research interests include heavy-tailed distributed time series, stochastic modeling, and statistical analysis of real data, with a focus on data related to the mining industry, indoor air quality, and financial time series. She has authored over 100 research papers and collaborates with industrial companies such as KGHM and Nokia.

In presenza

Presentazione orale

P.D.2-2 - Giugno

June - July