Fault diagnosis of a power transmission system using artificial neural networks

Mechanical power transmission systems are essential part of industrial process. The most complicated equipment is gears in power transmission systems. Worm gearboxes where needed high transmission ratios in one reduction stage are used in many various applications. Worm wheel, which is the driven member of the worm drive, may defect easily because it is produced softer material in comparison with the worm. Whenever a defect occurs in a worm wheel (e.g. pitting, abrasive wear), the performance of the gear decreases. And then, inevitably transmission of motion and power cannot be transferred as demanded. As a result, happening of serious faults becomes unavoidable. The research work presented in this thesis focuses upon the detection of localized pitting damages in a worm gearbox using artificial neural networks (ANN) and vibration analysis. Worm gear vibrations are acquired from an experimental rig utilizing a fifteenth worm gearbox. Simulated surface, which looks like pitting, was introduced on some of the worm gear wheel teeth using a drilling machine. The accelerometers are mounted permanently, radial and axial direction on the input shaft bearing housings. Statistical parameters of vibration signals in the time and frequency domains are used as an input to classifier ANN for multi-class recognition. Conclusions are drawn about diagnosis and classification of faults in worm gearboxes using ANN. The directions of accelerometers are compared with each other in respect to efficiency of classification of faults.