粒子群优化
方位(导航)
振动
阈值
超参数
滚动轴承
小波
降噪
聚类分析
工程类
主成分分析
计算机科学
噪音(视频)
人工智能
有限元法
还原(数学)
点(几何)
模式识别(心理学)
缩小
小波变换
断层(地质)
Morlet小波
算法
随机振动
特征(语言学)
融合
频域
特征提取
时域
降级(电信)
遗传算法
作者
Junning Li,Y T Wang,Mengsha Bai,Jinpeng Wen
标识
DOI:10.1177/01423312251411462
摘要
Addressing the challenges of dividing the degradation states of rolling bearings in Remaining Useful Life (RUL) prediction, as well as the issues of prediction lag and low prediction accuracy resulting from manual experience judgments, a method of RUL prediction of high-speed rolling bearings based on multi-signal fusion is proposed in this manuscript. First, wavelet thresholding denoising is used to denoise the original vibration signals. Feature extraction is performed on denoised vibration signals and degradation features are selected which effectively characterize the degradation progression throughout the entire life cycle of rolling bearings. Second, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) cluster analysis is conducted on the degradation features. The purpose of this analysis is to derive the initial degradation point and failure point based on vibration signals. Subsequently, Finite Element Thermal Analysis (FETA) is performed on the temperature signals, aiming to identify the initial degradation point and fault point based on temperature information. The two sets of results are compared and analyzed to verify the effectiveness of the method, and then these results are integrated based on principal component analysis (PCA) to obtain the final stage division result. Finally, the Gray Wolf Optimization (GWO), Sparrow Search Algorithm (SSA), and Osprey Optimization Algorithm (OOA) are compared through test functions to select the optimal optimization algorithm SSA. Then, the hyperparameters of long short-term memory (LSTM) are optimized based on SSA, and the RUL prediction for bearings is conducted using the SSA-LSTM model. The result demonstrates that this method achieves optimal prediction accuracy, with RMSE and MAE reduced by 77.59% and 78.03%, respectively, compared to the unoptimized LSTM model.
科研通智能强力驱动
Strongly Powered by AbleSci AI