Compensation of capacitive differential pressure sensor using multi layer perceptron neural network

Abstract

Capacitive differential pressure sensor (CPS), which converts an input differential pressure to an output current, is extremely used in different industries. Since the accuracy of CPS is limited due to ambient temperature variations and nonlinear dependency of input and output, compensation is necessary in industries that are sensitive to pressure measurement. This paper proposes a framework for designing of CPS compensation system based on Multi Layer Perceptron (MLP) neural network. Firstly, a test bench for a sample popular CPS is designed and implemented for data acquisition in a real environment. Then, the gathered data are used to train different MLPs as CPS compensation system which inputs are the output current of CPS and temperature value, and the output is compensated current or computed pressure. The experimental results for an ATP3100 smart capacitive pressure transmitter show the trained three layers MLP with Levenberg-Marquardt learning algorithm could effectively compensate the output against variation of temperature as well as nonlinear effects, and reduce the pressure measurement error to about 0.1% FS (Full Scale) , over the temperature range of 5 ~ 60 ° C.