Microcontroller Based ANN for Pick and Place Robot Coordinate Monitoring System

Abstract

Industrial Robots have captivated the manufacturing process of a product in the present assembly lines. The pick and place robot plays a vital role in this process for handling the products. But sometimes it may deviate from its desired position due to vibrations in the motors or due to external factors such as the impact on the robotic arm by the nearby robotic arm in an assembly line, resulting in aberrant gripping of the product. The resulting product either becomes unusable or gets damaged. As a solution to this a microcontroller- based machine learning coordinate monitoring design is proposed. A Feed-Forward neural network is used to determine whether the robot can pick the product or not. Before the robot picks the product the position of the robot arm is tracked by the three-axis angle sensor. The simple design of the system makes it easier to implement. The output of the feed forward neural network in microcontroller will determine whether the robot arm can grip the product. The network is trained through an iterative process with the training data which consists of both accepted and rejected values. The performance of the network is tested by exposing the outputs of the sensor (i.e. test data) to the network. The accuracy and the performance of the network are achieved by modeling the network architecture with the required number of neurons in the hidden layers. The accuracy of the neural network designed is observed to be around 98% from the respective accuracy graphs at different training process. The simple design procedure makes this system compact and reprogrammable.