Convolutional Neural Network (CNN), which is specialized in finding the principal components of data, has been widely used to extract features and to recognize patterns in different fields. When it comes to the machine diagnostics, however, the detection of abnormal patterns in signals requires the sophisticated analysis of data. A defected bearing, for example, generates vibration signals with the characteristic frequencies of defects in low frequencies and their harmonics in high frequencies. However, the characteristic components of the signals could often be obscured by the principal components of system noise. Therefore, even though the CNN model is well-trained and predict with high accuracy, it may only detect primary patterns of data. The primary patterns could be formed by system noise, which is, in fact, highly changeable especially in maintenance, reassembly or variation of operating condition. In other words, CNN models could misdirect even with excellent performances.

In this study, a novel method that combines the data preprocessing part with the classification part using CNN model is proposed for bearing fault diagnosis. The proposed method is demonstrated by the following steps. First, training data are preprocessed so that noise of the system condition and the components of bearing faults are separated. Then, CNN models are developed and optimized to learn significant features containing information of defects. Lastly, experiments on defective bearings in a centrifugal pump have been performed and the CNN models are examined to validate whether they learn and extract the meaningful features or not. The proposed method is applied to diagnose bearing conditions and confirmed to be successful.


Convolutional Neural Network Considering the Effects of Noise for Bearing Fault Diagnosis

Division / Sponsors: 
Power Division & Nuclear Engineering Division (NED)

Power Conference
Paper Proceedings: https://asmedigitalcollection.asme.org/POWER/POWER2020/volume/83747

International Conference on Nuclear Engineering
Paper Proceedings: https://asmedigitalcollection.asme.org/ICONE/ICONE2020/volume/83761

POWER-ICONE - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.