
Investigation of defects in roll contacts of machine elements with Acoustic Emission and Unsupervised Machine Learning
Author(s) -
J. Hillenbrand,
J Detroy,
Jürgen Fleischer
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1193/1/012085
Subject(s) - downtime , artificial intelligence , acoustic emission , computer science , unsupervised learning , machine learning , supervised learning , filter (signal processing) , classifier (uml) , pattern recognition (psychology) , artificial neural network , acoustics , computer vision , physics , operating system
In the age of Industry 4.0 and IIoT machines are becoming increasingly connected enabling continuous monitoring. A variety of information from machines and installed sensors is used to develop condition monitoring solutions. These systems are used to prevent premature failures and the follow-up costs due to machine downtime associated with them. Recent research in this area applies supervised machine learning, extracting features from captured signals and training classifiers. Supervised learning approaches require large amounts of labeled data, whose generation is time consuming and requires domain knowledge. For this reason, an unsupervised learning approach is being used in this work to distinguish between different defect and operation states of axial ball bearings. Within the scope of this work, acoustic emission (AE) measurements in the ultrasonic range are recorded and evaluated. Artificial defects are seeded in the rolling contact of axial bearings. From the AE signals a selection of state-of-the-art features is extracted. Then, the Laplacian Score, an unsupervised filter algorithm, is used to select the most significant features. Subsequently, the DBSCAN clustering algorithm is used to draw conclusions about the existing damage.