Open AccessApplication of spectral analysis methods for data pre-processing of anomaly detection problem of vibration diagnostics in non-destructive testing
Author(s) -
N N Trufanov,
Д. В. Чуриков,
O. V. Kravchenko
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2127/1/012028
Subject(s) - python (programming language) , computer science , vibration , signal processing , fourier transform , anomaly detection , process (computing) , frequency analysis , data processing , machine tool , algorithm , artificial intelligence , machine learning , digital signal processing , data mining , mechanical engineering , engineering , acoustics , mathematics , mathematical analysis , physics , computer hardware , operating system
The paper is devoted to the problem of primary data processing obtained in the vibration measurements during the processing of the workpiece on a milling machine with computer numerical control. An experimental setup is described and an algorithm for analysing vibration diagnostics signals using a mathematical machine learning tool is proposed. Special attention is paid to the study of the rigidity characteristics of the machine at different relative positions of its components. The analysis of the equipment design and factors affecting the ongoing process is carried out, as a result of which the received signal is processed and its characteristic fragments in the time and frequency domains are identified. The data is prepared for further use in solving the problem of detecting anomalies of the technological process, which implies predicting the progress of the technological process based on a mathematical model constructed using machine learning methods, and identifying deviations of the real technological process from the forecast. Preliminary preparation is carried out using the windowed Fourier transform. Various variants of windows in the transformation are considered, including those constructed using atomic functions. Calculations are performed using the Python 3.9 language, the main results are supported by graphs. The development of training methods for the considered models of neural networks is the subject of further research.