Discriminative Angle Feature Learning for Open-Set Deep Fault Classification

One of the most important parts of rotating machinery is the gearbox, which is also essential for fault diagnosis. This work overcomes the limitations of using limited features to diagnose gearbox faults by introducing a unique model of gearbox fault diagnosis that makes use of multi-model feature fusion. This is especially important in situations where different gearbox speeds mix complex, unimportant noise components with the vital signals in the vibration data. Therefore, before performing any analysis, it is crucial to successfully separate the vibration signal’s relevant components from its noise. This work introduces a novel totally automatedfault classification techniqe for gearboxes that does not require specialized knowledge of the gearbox’s operational load or design. Our technique separates the predicted error from the vibration data using an adaptive filter. In order to precisely categorize the gearbox’s state, the standard deviation of these prediction errors is further examined using a support vector machine (SVM).