Visual and acoustic methods are commonly used to identify faulty or failing idler bearings but these methods can become tedious and time consuming in practice. While vibration monitoring might look like an obvious choice to explore, the instrumentation of individual idler bearings would be prohibitively expensive. The potential for using an accelerometer that moves with the belt while tracking the condition of all bearings encountered along the way is therefore potentially interesting. This possibility is explored in this work on a laboratory scale test rig. Wavelet package decomposition is used to extract the bearing features and present it to an artificial neural network and support vector machine to identify and classify faulty idler bearings. The system could not only identify faulty bearings but also classify the faults accurately.