<![CDATA[Accurate error detection in mobile robots is crucial for reliable operation and prevention of mechanical or electrical failures. Mechanical defects on the wheels of mobile robot make real path deviate from the desired path and trajectory. From the kinematics equations, error in the angular velocity of wheel affects the position, velocity, and acceleration. Each of these signals is fed to the Symelet discrete wavelet transform (SDWT) for the purpose of error's feature detection and extraction. The SDWT with 5-level for each component of the signal produce 10 inputs for the Bayesian Neural network (BNN). The BNN with single layer of 18 neurons classifies the inputs either no error case or specify the wheel(s) at which error had been happened. Straight line and circular paths were tested in the presence of errors in single wheel or both wheels. Two different path's time durations are tested to investigate robustness of the proposed methodology. The simulation’s results of two wheels mobile robot showed that acceleration's signal for a straight-line path has accuracy of 100%, MSE 3.05×10-23 and 9.81×10-17, training iterations are 15 and 23 for 4- and 2-seconds durations; respectively. While for a circular path, displacement's signal gave high accuracy 100%, MSE 8.86·10-16 and 3.76×10-18, training iteration 17 and 13 for 4- and 2-seconds durations; respectively. Acceleration signal can be used for detecting errors in real time by using accelerometer. Limitations such as amount of data besides to the sensor noise affects the proposed methodology.]]>
Copyrights © 2024
