Sensor Modeling and Linearization Using Artificial Neural Network Technique

Abstract

Many Sensors show a nonlinear relationship between their input and output. Sometimes the reason for nonlinearity is inherent and sometimes it is due to the changes in the environmental parameters like temperature and humidity. Ageing is also responsible for the nonlinearity of sensors. Due to the presence of nonlinearity, it becomes very difficult to directly read the sensor over its whole sensing range. The accuracy of the device is affected if it is used in its full input range. Hence it is very much necessary to study the problem of nonlinearity present in sensors and to solve it. Thermistor and thermocouple are the temperature sensors that exhibit nonlinear characteristics. Thermistor is the most nonlinear device but thermocouple is linear if operated in a specific operating temperatures. Thermocouple shows nonlinearity if operated in its entire operating range. The nonlinearity of a sensor can be compensated by designing an inverse model of the sensor and connecting it in series with the sensor. This enables the digital readout of the output of the sensor. So the inverse models of these temperature sensors are designed and connected in series with them, so that the associated nonlinearity can be compensated and the output can be read digitally. The neural network technique seems to be an ideal technique for designing the inverse model of such sensors. Also, a direct model of such sensors is also designed which can be used for calibrating inputs and for fault detection. A technique for linearizing the output of the sensor without using inverse modeling is also discussed.