Research Of Online Terrain Classification Methods Based On Vibration Signals

With mobile robots being applied more and more widely, there is a demand that the ability of robots to perceive the environment being higher. The complicated ground is usually externalized by the geometric parameters and the physical parameters. Recently, the majority research of unknown environment is focused on the geometric parameters which represent environmental structural parameters like ground obstacles. However, with the using of the robots more widely, the physical parameters’online-identification and real-time modeling which affect robots’continuous survival and self-sustaining is becoming hot spot of the research. Targeted the vibration signal’s characteristics of tracked robots’running process, in this article, the online classification which using the vibration signal of the terrain is established.For the vibration signal’s complex of tracked robot, analyze the resource of vibration and the main factors that affect the local oscillator through the simulation of robot’s ADAMS model. Firstly, establish the simulation model of tracked robot. Then, based on the result of simulation, analyze the affect of tracked robot’s speed and different terrain types. In the end, the major source of track vibration and vibration’s characteristics are analyzed, this process provide theoretical reference and validation standards for the real extracted of robot’s vibration.With the results of signal’s characteristics and influencing factors of the result, the acquisition of vibration signal platform is established, the signal’s processing procedures and the collection interface are designed, and scheme acquisition of vibration signal is proposed, the preprocessing of vibration signal and the extracted methods of time and frequency domain. Based on this, the impact of feature extraction is analyzed the time domain and frequency domain characteristics are compared, the extraction of the four terrain samples are completed, characteristics of the vibration signal analysis is completed.Based on the extracted of vibration characteristics, there are a variety of feature classification methods and feature vector extraction that being compared experiment. Based on PCA and SVM, the online vibration signal and identification method are established, and then, according to the vibration signal’s characteristics, the vibration signal feature vectors is proposed, and SVM offline teach and parameter selection are conducted, the offline teach classification model is got. Based on this model, online recognition system is established, and the classification results is obtained.Based on tracked mobile robot, the hardware and software systems which are used to conducted vibration signal extracted is established. Using the model which is established with classification, the experiment which doing on the composite land that contain the asphalt and grassland, and the classification accuracy and the classification speed of response indicators are obtained. In order to overcome the shortcomings that vibration sensors cannot predict the terrain, vibration identification which is based on forward tentacles round is established, the online classification which is used to sort the pro-run terrain is achieved.