Neural network application in a carbon monoxide measurement system

Reports an application of neural networks to the measurement of carbon monoxide in air. The measurement system uses a tin film sensor. To increase system performances and capabilities, a microcontroller and complex processing are added to the sensor. Several artificial neural network architectures are designed, tested and implemented in the microcontroller to perform the CO digital readout task with compensation of errors due to the most important influence quantities. Special attention is paid to the implementation of multilayer perceptrons and radial basis functions and to the selection of the smallest neural network topology for an imposed modeling error.