Research on early fault intelligent diagnosis for oil-impregnated cage in space ball bearing

As the core for the lubrication and operation of space ball bearing, the porous non-metallic cage with oilimpregnated will cause catastrophic results in case of fault. The principal used reasons for not realizing its early fault diagnosis are: vibration, temperature, and sound signals may not carry the early fault information, and the traditional models could not work effectively in the case of insufficient data samples. Aiming at the above problems, this paper first presents a visual cage motion test rig, which realizes the accurate acquisition of cage dynamic signals without changing the bearing structure. The early fault dataset of the space ball bearing cage is then provided, composed of the data sample including the dynamic signal with early fault information of the cage. Finally, a Similarity Clustering Two-Channel Convolutional Neural Network (SC-TCCNN) model, which integrates the similarity clustering method into the Double-TCCNN framework with parameter sharing, is proposed. The experimental results show that the method proposed realizes the intelligent diagnosis of cage early fault and shows more competitiveness than the traditional model in the case of insufficient training data samples.