An artificial neural linearizer for capacitive humidity sensor

The overwhelming use of computer based data acquisition systems in the industries and also in the laboratories, has resulted in a spurt in the development of methodologies for software linearization of sensor characteristics. With the widespread application of artificial neural network (ANN) as a soft-computing technique, it is hence quite natural that the same would also be used for numerical linearization of sensor characteristics. Although some work has been reported in this area, there is still a huge scope for much more sophistications. The present paper describes the development of an artificial neural network for linearizing the static characteristic of a capacitive thin-film humidity sensor. Computations have been carried out with the manufacturers´ data for an industrial capacitive humidity sensor. The effects of variation of temperature and operating frequency on the capacitance of the sensor have also been taken into consideration for designing the ANN. The dependence of sensor capacitance on more than only physical variable, justifies the use of ANN as a linearizing tool. The ANN employs radial basis function (RBF) network in the hidden layer, followed by an ADALINE network, for automatic adjustments of weights and biases to arrive within a desired maximum allowable error. This adaptive network successfully trained, has been used as a Pleural linearizer for the sensor under consideration. The primary variable (humidity) values generated by the neural linearizer from the capacitance values of the sensor have been found to be in reasonably close agreement with those provided by the manufacturer.