Research On Single Leg Mechanical Damage Detection Of Six Legged Robot Based On SVM

The main research objective of this paper is to construct a mechanical damage detection system for a large-sized heavy duty hydraulically driven hexapod robot with single leg.According to the research objective,the main contents of this paper that are finished are as follows:(1)Discussed and analyzed machine learning problems and SVM methods.The classification and regression of SVM is mainly studies.(2)According to the small sample of experimental data,non-linearity,high dimension and other characteristics,based on consultation of a large number of technical data at home and abroad,a set of mechanical damage detection program for a large-sized heavy duty hydraulically driven hexapod robot with single leg based on SVM was proposed.For the mutual constraint relation between RBF-SVM model parameters,the “good area” of parameter combination was obtained and the parameter optimization scope is narrowed.The influence of key parameters on pattern recognition and regression fitting accuracy was analyzed by combining the evaluation methods of the advantages and disadvantages of classification and regression model parameters.Combined with the GS methods,an improved GS variable-step length multiple search method based on “good area” was proposed.Finally,computer simulation technology and digital signal processing technology were used as the main means to verify the validity and applicability for the use of data acquisition of vibration test and the use of SVM methods for pattern recognition.(3)According to the characteristics of hexapod robot's working environment,the main mechanical damage patterns of the legs were analyzed.A three-dimensional model was established based on its typical structure and various damage patterns and the test sample data was obtained.The SVM methods and improved parameter optimization method were applied to the mechanical damage diagnosis,damage pattern recognition and damage location experiment,respectively.The effects of data preprocessing,different types of kernel functions and different parameter optimization methods on the test results were analyzed and compared with the analysis results of neural network to verify the effectiveness of this method.