Application of neural networks for sensor performance improvement

Sensor technology has developed in parallel with advances in the fields of electronics and computing. Beyond obtaining a suitable sensing element, stringent demands on accuracy has led to continued developments in the improvement of compensation and calibration techniques. Typically, signal conditioning would attempt to minimise the effects of zero offsets and nonlinear temperature and pressure effects. Conventional analogue compensation methods have been phased out in favour of digital methods which provide a lower cost solution due to the reduction in test and calibration time. However, digital methods currently employed have been deemed to be insufficiently accurate or highly memory intensive, thus there is a need for an alternative approach that provides a compromise between the above. The use of neural networks may offer this compromise, with the added advantage of possessing certain characteristics that could contribute to the development of a smart transducer.<>