Background: Rotating machinery is critical in various industries, including power
plants, and plays a significant role in maintaining the continuous operation of
industrial processes. The reliability of these machines is highly dependent on their
proper functioning. Therefore, fault detection in rotating machinery, such as
generators, pumps, motors, and compressors, is crucial to prevent catastrophic
failures and minimize unplanned downtime. Any proposed fault detection method
for these machines must be computationally efficient, reliable, and easy to
implement. In this study, we propose an online fault detection method for rotating
machinery based on semi-supervised artificial intelligence, utilizing auxiliary
signals from the equipment to identify and predict faults.
Aim: Specifically, we aim to utilize a 1D CNN architecture for feature extraction of
the side signals which reflect the working faults of the rotating machinery and to
develop a state-of-the-art semi supervised neural network structure by using DFNN
which can accurately classify and detect the faults while considering the normal
model drift of the machine. The model will be tried to be simple enough to be
implementable in common and economical industrial computing boards.
Methodology: In this research, we use a classified dataset of rotating machinery in
semi-supervised learning and label all the data. Then, for the purpose of feature
extraction, we input the data into a CNN. Finally, after feature extraction, we use
the DFNN method for data classification.
Conclusions: The main findings of this research are a semi-supervised learning
CNN structure, which is classified by DFNN for rotating machinery fault detection.
In this research, we obtained the initial model using labeled data through a semisupervised neural network. Subsequently, we pseudo-labeled the unlabeled data
using this model. Finally, we implemented the final model with an accuracy of
99.88% for rotating machinery fault detection using b