Multimodal loosening detection for threaded fasteners based on multiscale cross fuzzy entropy

In various mechanical systems, threaded fasteners are widely used to connect two or more separated components. Loosening in threaded fasteners is prone to occur due to the exposure of vibration environment for time. Regular loosening detection cannot be overemphasized. Traditional single-modal loosening detection method easily generates insufficient feature representation due to the limitation of information. Thus, the detection accuracy and reliability are decreased. This study is the first attempt to conduct multimodal loosening detection exploiting ultrasonic and audio response signals simultaneously. A novel loosening detection method is proposed making use of the complementarity of multimodal signals. In the method, the concept of multiscale cross fuzzy entropy (MCFE) is proposed, and the multimodal information is mapped into a unified feature space to construct more representative and effective loosening features. Linear discriminant analysis method is applied to remove redundant features and a random tree is used to detect loosening severities of threaded fasteners. The detection performances are both excellent in the applications of two different types of threaded fasteners (i.e., lap joint and globecone joint), which validates that our proposed multimodal loosening detection method shows great application potentials in industry. In addition, it demonstrates that our proposed method outperforms other loosening detection methods and MCFE shows great advantages in extracting representative loosening features.

Automated summary

This study is the first attempt to conduct multimodal loosening detection exploiting ultrasonic and audio response signals simultaneously. A novel loosening detection method, utilizing the complementarity of multimodal signals, is proposed and proved to have excellent detection performances in the applications of two different types of threaded fasteners. The proposed method outperforms other loosening detection methods and MCFE shows great advantages in extracting representative loosening features.