Boosting(机器学习)
奇异值分解
脉冲(物理)
残余物
降噪
算法
噪音(视频)
计算机科学
模式识别(心理学)
数学
人工智能
物理
量子力学
图像(数学)
出处
期刊:IEEE Transactions on Industrial Electronics
[Institute of Electrical and Electronics Engineers]
日期:2019-02-15
卷期号:67 (2): 1282-1292
被引量:39
标识
DOI:10.1109/tie.2019.2898583
摘要
The key to successful detection of a localized bearing fault lies in extracting fault-induced impulse responses. Unfortunately, these responses are often contaminated by background noise. The popular K-singular value decomposition (K-SVD) denoising algorithm can be used to extract impulse responses from noise, but it may obtain weak impulse responses against reliable fault detection. An SOSO boosting technique is proposed to improve the performance of the K-SVD denoising algorithm. Given an initial denoised signal, this paper proposes iteratively repeating the following SOSO boosting procedure: 1) Strengthen the underlying signal by adding the previous denoised signal to the noisy input signal; 2) Operate the K-SVD denoising algorithm on the strengthened signal; 3) Subtract the previous denoised signal from the restored signal-strengthened outcome; and 4) Operate the modified K-SVD denoising algorithm again on the resulting signal after the subtraction. The last step, i.e., the secondary denoising step, plays a crucial role in preventing the enhancement of the residual noise that originates from the first denoised result. The results of numerical and experimental studies show that the SOSO-based K-SVD denoising algorithm not only enhances weak impulse responses significantly but also reduces potential residual noise effectively compared to the SOS-based and the original K-SVD denoising algorithms.
科研通智能强力驱动
Strongly Powered by AbleSci AI