Abstract: To address the demand for fault diagnosis of aero-engine bearings under extremely complex working conditions,a novel high-precision intelligent fault diagnosis method for aero-engine bearings was proposed.Initially,a dual-channel pseudo-siamese neural network was constructed,incorporating a one-dimensional adaptive batch normalization convolutional neural network module and a one-dimensional physics-guided convolutional neural network module,which could address integration of multi-source sensor data effectively,extract the key features and achieve the efficient fusion of diverse data sources.Subsequently,the gated recurrent unit optimized by nutcracker optimization algorithm was used to recognize fault state,enhancing the model's adaptability,robustness and generalization capability,especially under complex working conditions and noisy environments.Finally,the model's effectiveness was validated through case analysis,demonstrating fault diagnosis accuracy rates of 99.99%,100% and 100% respectively,with a maximum improvement of 23.97% in recognition accuracy compared to existing diagnostic methods,highlighting clear performance advantages.Additionally,generalization and noise resistance tests were conducted,and the results revealed superior noise robustness and strong generalization performance.

Key words: multi-source sensor data, fault diagnosis, pseudo-siamese neural network, nutcracker optimization algorithm, gated recurrent unit

摘要： 为应对航空发动机轴承在极端复杂工况下的故障诊断需求,提出了一种新的航空发动机轴承高精度智能故障诊断方法。首先,构建了具有一维自适应归一化卷积神经网络模块和一维物理特性指导的卷积神经网络模块的双通道伪孪生神经网络,有效应对多源传感器数据的融合问题,提取关键特征并实现多源数据的高效融合。其次,采用星雀优化算法优化门控循环单元模块进行故障状态识别,提高了模型的自适应性、鲁棒性及泛化能力,特别是在复杂工况和噪声环境下。最后,通过算例验证模型的有效性,提出方法的故障诊断准确率分别达到99.99%、100%和100%,与现有几种诊断方法对比,所提方法的识别准确率最高提高了23.97%,展现出明显的性能优势。此外,进行了泛化能力验证实验与噪声对抗实验,结果表明了其优越的抗噪声能力和强大的泛化性能。

关键词: 多源传感器数据, 故障诊断, 伪孪生神经网络, 星雀优化算法, 门控循环单元