基于提升卷积神经网络的航空发动机高速轴承智能故障诊断

Aero-engine bearings usually operate for long hours under harsh conditions of high speed and heavy loads, inevitably leading to performance deterioration and even causing various faults, and automatic and accurate fault diagnosis methods for high-speed aero-engine bearings can help to improve operation safety and maintenance economy. The original vibration signals collected from aero-engine high-speed bearings have strong instability and the number of faulty samples is much smaller than that of healthy ones. The traditional intelligent diagnosis method tends to skew to large samples, thereby inducing degradation in diagnosis performance. To solve the above problem, we propose an enhanced convolutional neural network model based on adaptive weight and multi-scale convolution. A multi-scale convolution network is first constructed to extract multi-scale features of fault samples and mine useful identifying information. An adaptive weight unit is then designed to fuse the multi-scale features to increase the contribution of the important features while reducing the influence of unrelated features. Focal Loss is finally used as the loss function to enable the model to consider the small faulty samples and easily confused samples more. The test and analysis of aero-engine high-speed bearing vibration data confirm the feasibility of the proposed method in fault diagnosis tasks with unbalanced data.