László Móricz, Zsolt János Viharos
Vibration Based Cutting Tool Path Differentiation by Feature Selection

During machining along a complex tool path, it is difficult to connect the change in the amplitude and in the frequency of the recorded vibration signals to the actual state of tool wear. In this article, a method is presented in which the tool wear processes that occur during the machining of a complex geometry (cooling pocket formed on the ceramic coating of the turbine blade) can be evaluated based on the vibration data series recorded during the machining differentiating of some elements of the complex path movement performed by the tool.

Monica Egusquiza; Alex Presas; David Valentin; Carme Valero; Beibei Xu; Diyi Chen; Eduard Egusquiza
Diagnostics of a hydraulic turbine failure

This paper presents the failure analysis and diagnostic of a hydraulic turbine. Shortly after a maintenance revision the thrust bearing of the turbine was destroyed when the turbine was put into operation. The damaged bearing was examined and the possible causes discussed. To identify the problem that led to the failure of the thrust bearing, a comprehensive on-site measurement campaign was done. Vibrations, pressures, temperatures and operating parameters were acquired at different operating conditions of the turbine. The analysis of the data allowed to determine the source of the problem and to implement a solution.

Iacopo Marri, Emil Petkovski,Loredana Cristaldi, Marco Faifer
Lithium-ion batteries soh estimation, based on support-vector regression and a feature-based approach

Lithium-Ion batteries, have become enormously used in many systems and applications, and are the most widespread energy storage system. Optimizing the usage of batterie is therefore very important to increase the safety of systems like electric vehicles or portable devices, to reduce economic loss in industrial environments, and to increase their availability. An accurate State of Health (SoH) estimation is important since it allows us to know battery conditions and make an appropriate use of it, and it improves the accuracy of other diagnostic measures, like State of Charge (SoC). In this paper, an approach for SoH estimation is proposed, based on Support Vector Regression machine learning algorithm and a smart feature extraction process, finding a good trade-off between applicability, light computation effort, and accuracy of results. Features selection and parameters tuning are discussed, and performances are measured on a dataset from the Prognostics Center of Excellence at NASA, considering 3 batteries of the dataset.

Yu Jun, Olena Hordiichuk-Bublivska, Yan Lingyu, Marian Kyryk, Mykola Beshley, Hu Jiwei
Big Data Аnalysis in Smart Grid Systems

The problem of Big Data processing in large industrial systems requires the use of machine learning methods. A smart grid system is an example of improving the efficiency of traditional data processing systems, which allows much more efficient and flexible distribution of electricity to end-users. However, for a smart grid to work properly, it needs to constantly monitor data from sensors and meters. The Singular Value Decomposition (SVD) algorithm is used to improve the efficiency of Big Data processing and reduce its dimensionality. The paper proposes the use of advanced SVD, which can work in distributed industrial systems and ensure the reliability and speed of data processing.

Krzysztof Dowalla, Piotr Bilski, Ryszard Kowalik
Series arc fault detection and line selection based on Non-Intrusive Load Monitoring Method

The series arc faults are a common cause of household fires. Low fault current amplitude is the reason for the difficulties in implementing effective arc detection systems. The paper presents a novel arc detection and line selection method. It can be easily used in the low-voltage Alternate Current (AC) network to enhance the functionality of the Non-Intrusive Load Monitoring (NILM) system with arc detection for the whole household. Unlike existing methods, the proposed approach exploits not only current signal but also voltage signal time domain analysis. In the case of arc fault detection, line selection is based on the mean values of changes in the consecutive current signal periods during the arc and comparing them with current waveforms for each appliance in non-arc conditions. Tests have been conducted with up to 6 devices operating simultaneously in the same circuit. Single period arc detection accuracy was 98.47%, with recall at 97.5% and F-score of 0.983. The arc detection accuracy in terms of the IEC 62606:2013 standard was 99.33%, F-score of 99.33. Line selection accuracy was 91.57%.