Bolt Loosening Detection of Rocket Connection Structure Based on Variational Modal Decomposition and Support Vector Machines

Abstract

This paper designed a bolt-loosening Support Vector Machines’ conduct detection method with feature vectors comprising eigenvalue decomposition based on Variational Modal Decomposition (VMD) and Singular Value Decomposition (SVD), combined with permutation entropy. Particle Swarm Optimization-Support Vector Machines (PSO-SVMs) are used for small-sample machine learning and can effectively identify and judge the state of bolt preload. The effectiveness of the proposed method is verified in a typical example of a connection structure under random-amplitude impulse loads and Gaussian white noise with different signal-to-noise ratios. The effect of other bolt numbers being arranged is also discussed in the results. This method’s bolt-loosening identification rate is close to 90% under both equal-amplitude and variable-amplitude loads. Following the interference, with a signal-to-noise ratio of 20 dB, the method also has a recognition rate higher than 70% under various working conditions and bolt equipment schemes. The effectiveness of the method was verified by experiments.

Keywords

• bolted flange
• bolt-loosening detection
• variational modal decomposition
• support vector machines
• permutation entropy
• bolt preload