CALIBRATION OF A LOAD CELL USING A NEURAL NETWORK

Abstract

A neural network is used to calibrate a load cell that was built using strain gages. The inputs to the neural network
are the reference voltage applied to the Wheatstone bridge formed by the strain gages, the amplification value applied
to the Wheatstone bridge's output voltage, and the 8-bit digitized voltage value acquired by a microprocessor. The
output of the network is the estimated value of the weight being applied to the load cell. The network's main objective
was to learn an accurate input-output relationship of the variables in the load cell system. The backpropagation
Levenberg-Marquardt algorithm was used to train the network, and satisfactory results were obtained with a 5-3-1
neural network. This project could be used as an example to design similar neural networks for other applications.