Research On Ultra-high Speed Collision Vibration Wave Identification And Location Algorithm Based On Machine Learning

In recent years,high-speed or ultra-high-speed collision localizations have been widely used in many fields,such as the protection of manned spacecraft,robots,vehi-cles,ships,and the detection and positioning of friction collision damage of machine tools.However,for the processing of collision surface vibration signals,traditional methods can usually only deal with low-speed collisions,and the relevant recognition and localization algorithms for high-speed collisions are still lacking.In this paper,the object selected is an aluminum plate,and the high-speed or ultra-high-speed collision surface vibration signals are collected.An effective recognition and localization algo-rithm is proposed,and the high-speed or ultra-high-speed collision of vibration source accurate identification and precise positioning intended to achieve.The main research results of this paper are:(1)Analyze the propagation characteristics of vibration signals on the collision sol-id surface,study the energy distribution of surface vibration signals of three different collision types,and the energy distribution of surface vibration signals at different dis-tances.Based on the principle that wavelet time scale graphs represent different energy distributions of signals,a multi-scale discrete wavelet transform(MDWT)extraction algorithm is proposed.The MDWT feature can be applied to the characterization of surface vibration signals of high-speed/ultra-high-speed collisions at the same time Based on MDWT features,k-Nearest Neighbor(KNN)is used as an accurate and effi-cient classifier and Kernel Extreme Learning Machine(KELM)algorithm as a regression model training algorithm.The vibration signal recognition and localization algorithm of MDWT+KNN+KELM is proposed.(2)The vibration source localization problem has been improved,and the Syn-chrosqueezing Transform(SST)algorithm and the texture color distribution(TCD)of the image have been deeply studied.The second extraction of TCD features is performed by a feature-level fusion algorithm.The fusion method is based on the correlation and similarity between the labels and each dimension feature in the regression problem.(3)Finally,it was optimized for the construction of classification and regression models.Experiments were carried out with richer machine learning algorithms.The three underlying classification and regression models were merged in Voting and Stack-ing as the final algorithm for building models.The experimental results show that the improved vibration source identification and localization algorithm(SST,TCD+Vot-ing+Stacking)has a more accurate vibration source type recognition and improved distance prediction accuracy compared with the previous one.Then,based on the dis-tribution of the sensors,the four-circle centroid localization(FCL)algorithm is used to more accurately locate the vibration source coordinate localization.