Lime Kiln Fault Detection and Diagnosis by Neural Networks

Artificial neural networks have recently been used successfully for fault detection and diagnosis in chemical processes. In this paper, we present a study on fault detection and diagnosis of an industrial lime kiln which is a complex highly nonlinear process within the pulp and paper industry. We show the capability of neural networks to learn faults which can occur during steady state kiln operation, their adaptation to different input distributions involving nonlinear mappings and their capability to spontaneously generalize. We compare the performance of two architectures, nameley BPNN (Back Propagation Neural Network) and RBFNN (Radial Basis Function Neural Network), and investigate several topologies. Through this study, it can be concluded that the RBFNN arquitecture learns faster, with less error and performs better at classifying kiln malfunctions.