In an indoor pulse-jet bagfilter plant, erroneous activation of individual valves is to be detected. The present approach uses data of a microphone array to localize the source of sound originated by the valve. Since the valves are all installed in a horizontal layer, a two-dimensional (2D) microphone array can be used. Multipath (constructive and destructive interference) and reverberation effects due to the presence of considerable obstructions make the use of timedelay-based approaches impossible. Thus, the problem was encountered by an energy-based concept. With additional use of a kernel classifier, proper feature selection, and adaptive training, the density of microphones can be reduced, keeping a high probability of correct detection.