Optimum choice of signals' features used in toothed gears' diagnosis

The article proposes an algorithm to choose optimum diagnostic features used in toothed gears’ diagnosis. The test object is a single-bevel gear in the research area. From the gear in two states there were collected vibration signals and eight features were calculated. Feature and machine state correlation degree depends on the type of damage and analyzed object properties. Some features are insensitive to particular damage or may transmit the same information. Signal features choice is a crucial step which influences the final technical condition evaluation. With the algorithm that automatically verifies features’ usability there were chosen four best correlated with the technical condition of the object. Gear state classifiers were two neural networks, one formed of four features and the other of all eight. The other one was set to check features’ choice accuracy.