Ship Slamming Loads Detection Algorithm Based On ELMD And KNN Classifier

With the rapid development of shipbuilding industry, the importance of the ship structural safety is also increasingly outstanding. Ship structural monitoring(SSM) system can monitor real-time stress data of each part of the hull, to access the ship structural safety. The signals collected by the SSM system contain the slamming loads components which are produced by hull sailing in the ocean. High strength slamming loads will affect hull structure;it will threaten the safety of the ship. The effect of low intensity slamming loads on the hull has cumulative damage. Although low intensity slamming loads don't like high strength slamming loads to destroy the hull structure in a short time, it will cause cumulative damage to the hull structure and may threaten the safety. By detecting the identification slamming loads' signals, and combining with the current sea conditions to evaluate the safety of the hull structure, and according to the security level to monitor,can guarantee personal and property safety when the ship is sailing. Therefore, the study of slamming loads detection and recognition is useful and meaningful.Slamming signals are non-stationary and nonlinear complex signals. Considering the character of slamming load signals,a slamming detection method based on the Ensemble Local Mean Decomposition(ELMD) and K Nearest Neighbor(KNN) is proposed in the thesis.Firstly, Compared Local Mean Decomposition(LMD) with the traditional time-frequency analysis and Hilbert-Huang Transform(HHT), LMD has a good time-frequency analysis performance and doesn't have interference term. Therefore the time-frequency analysis based on LMD is more suitable for the analysis of slamming signals. Secondly, using the ELMD,which is an improved algorithm of LMD, can decompose the collected data and extract sample entropy, energy entropy and marginal spectrum amplitude ratio of slamming signals as slamming feature to constitute the feature vector. The feature vector and KNN classifier are combined to achieve detection and identification slamming loads. The experiment results show that the method presented in this thesis has a small calculation and a high accuracy rate.The feature vector and KNN classifier combined to achieve detection and identification slamming loads. The method proposed can be used in the ship structural monitoring system to detect the slamming signal and keep the hull structure safety.In the thesis, the research content mainly includes the following aspects:First of all, the principle of LMD is introduced in detail. Hilbert-Huang transform and traditional time-frequency analysis method such as short-time Fourier transform is also introduced briefly. The advantages and drawback of each method are analyzed through Matlab simulation. By the comparison, it shows that LMD method has better effective than others.Secondly the mathematical model and classification of slamming signals are proposed.An improved method of LMD which is named ELMD is described in detail. By using LMD,EEMD and ELMD, the multi-component signals which contain intermittent signals are decomposed. By the experimental result, ELMD has a good effective in the process of signal composition.Finally, the feature vector based on sample entropy, energy entropy and marginal spectrum amplitude radio of slamming signals is combined with KNN classifier to achieve the detection and identification of slamming loads. And the algorithm is applied to detect the real slamming signals. Compared with conventional threshold classification method, the method based on KNN has a high accuracy.