Real-time classification performance and failure mode analysis of a physical/chemical sensor array and a probabilistic neural network

The U.S. Navy program Damage Control-Automation for Reduced Manning is focused on enhancing automation of ship functions and damage control systems. A key element of this objective is the improvement of current fire-detection systems. An early warning fire-detection system is being developed by properly processing the output from sensors that measure different physical and chemical parameters of a developing fire or from analyzing multiple aspects of a given sensor output (e.g., rate of change as well as absolute value). The classification and speed of the probabilistic neural network (PNN), deployed in real-time, have been evaluated during a recent field test aboard the ex-USS SHADWELL, the Advanced Damage Control Fire Research Platform of the Naval Research Laboratory. The real-time performance is documented and as a result of optimization efforts, improvements in performance have been recognized. Early fire detection, while maintaining nuisance source immunity, has been demonstrated. A detailed examination of the PNN during fire testing has been undertaken. Using real and simulated data, a variety of scenarios (taken from recent field experiences) have been used or recreated for the purpose of understanding potential failure modes of the PNN in this application. © 2001 John Wiley & Sons, Inc. Field Analyt Chem Technol 5: 244–258, 2001