SBIR-STTR Award

The Gearbox Fault Detector is an artificial neural network based defect detection system for the helicopter gearbox. This system will detect defects in the helicopter tail rotor gearbox and also classify the type of fault. Several (four) sensors will be installed on various locations of the gearbox. The data will be continuously fed into a pretrained artificial neural network system (ANS) in real time without pre-processing. The ANS will then process that raw data and indicate if a fault exists and, if so, what type of fault exists inside the gearbox. The unique significance of this system is the use of a new neural network training algorithm, called the "Delta-Activity" algorithm to train the neural network. This algorithm can accelerate the neural network learning process, overcome learning instabilities, and optimize the net configuration. It is now possible to solve much more difficult problems on a small computer, such as an IBM-PC, within a reasonable time.

The AFT Transmission fault detector for the H-46 helicopter is an artificial neural network detection system. This system will detect and classify known faults, and identify the existence of unknown faults (novelty detection). Data is from an a H-46 AFT-MAIN gearbox and eight channels of vibration data will be continuously fed into a pre-trained artificial neural network system (ANS) in a real-time sequence without pre-processing. The significance of this approach is the use of a new neural network training algorithm, called the "optimized entropy". This algorithm is unique and has demonstrated the ability to accelerate the learning process, overcome learning instabilities, and optimize the net configuration in a high noise environment. The optimized entropy algorithm solves difficult problems on a desk top computer in (near) real time. The optimized entropy algorithm was successfully applied to the phase I project "Helicopter Gearbox Fault Detector".