The force and torque sensor is a tool widely used in industrial processes and research centers where data need to be obtained with high accuracy. To use the sensor is needed a specific data acquisition board, as well as appropriate software. In addition, a specific calibration for different types of plates is required. This paper aims to calibrate a force and torque sensor, using neural networks in a universal manner, where the reproduction of the neural network for calibration of any sensor in other data acquisition boards is possible, apart from using a different software recommended by manufacturer. To perform the experiments, we used a sensor force and torque, Gamma model 3805 of ATI Industrial Automation, a data acquisition board (National Instruments myDAQ ©) and a pulley system for the distribution of weights between the axes of the sensor. To acquire the data we used the Labview software. For the modeling of neural network, we used a specific Matlab software toolbox. The experiment was conducted by placing loads with different values in known positions, so that the forces and torques along the three axes, have had predefined values. After this procedure was collected data voltage to the weights defined. These measures served to feed the neural network developed in Matlab. The partial results obtained were satisfactory and the neural network, tested with experimental data presented minimum values of mean and degree of accuracy close to one. Work is in progress and we intend to conduct a new experiment using the measurements of the sensor in the control loop of a robotic manipulator.
Read full abstract