Neural network development for automatic identification of the endpoint of drying barley in bulk

A thesis was proved that it is possible an automati c endpoint determination of drying barley in bulk, 1.2 meter’s deep, based on a neural network, using a continuous on-line mea surement of atmospheric air temperature and relativ e humidity, plenum air temperature and grain temperature in selected l ocations inside the bed in situations in which dr ying air temperature and relative humidity change stochastically. The us f lness of individual input variables characterisi ng the process as well as their influence on the quality of the obtained m odel were analysed. Several different topologies of the developed models were compared and the RBF type networks were select ed as the best ones. The developed networks are cha acterised by a high, ranging from 93.3 to 99.6%, correctness of case assignment to the recognised classes in the co urse f the identification process and a high capability to generalise th analysed data.