Intelligent Hybrid Scheme for Health Monitoring of Degrading Rotary Machines: An Adaptive Fuzzy c-Means Coupled With 1-D CNN

The development of a health monitoring (HM) scheme for complex rotary machines using unlabeled and imbalanced multivariate data is a well-recognized challenging problem. The challenges and complexity further increase when the health of the machine degrades over time. To utilize unlabeled and imbalanced multivariate degradation data, this article develops an intelligent HM scheme for rotary machines based on the proposed adaptive fuzzy
$c$
-means clustering algorithm followed by ensemble learning, mean-variance-based samples generation method, and 1-D convolutional neural network. The adaptive fuzzy
$c$
-means clustering algorithm uses nondominated sorting genetic algorithm II to automatically determine the different health state (HS) labels from unlabeled degradation data. The principle of ensemble learning is used to increase the robustness of the obtained HS labels. Further, based on obtained HSs, a new solution is presented to construct the true remaining useful life (RUL) adaptively. The proposed framework presents a novel method for generating samples based on mean-variance to deal with imbalanced HSs. Furthermore, an architecture of 1-D convolutional neural networks for multivariate data is designed to predict the HSs and RUL of rotary machines. The proposed approach has been validated through aero-engine and XJTU-SY bearing datasets. The HS prediction accuracy is as high as 99.97% and 100% for aero-engine and XJTU-SY datasets, respectively, using the proposed scheme. The results and comparative studies show the effectiveness of the proposed scheme.