Abstract: The reducer is a key power component in new energy vehicles. The built-in rolling bearings are prone to various failures, which can lead to major traffic accidents and seriously threaten driving safety. To address issues in diagnosing reducer bearing faults under strong background noise–such as insufficient anti-interference capability of deep models, limited diagnostic accuracy, and difficulty in tuning deep model parameters. By leveraging the synergistic effect of the strong noise-robust feature extraction capability of stacked denoising autoencoder (SDAE) and the temporal modeling advantages of Bidirectional gated recurrent unit (BiGRU), combined with bayesian optimization (BO) for model parameter tuning to enhance diagnostic performance, this paper proposes a novel fault diagnosis method for new energy vehicle reducer bearings based on the BOBiGRU-SDAE model. Experiments were conducted on common fault states of new energy vehicle reducer bearings. The results showed that the proposed method achieved a diagnostic accuracy as high as 98.75%, significantly higher than other mainstream neural network models. The study confirms that the proposed method demonstrated excellent effectiveness and accuracy. The study can provide an efficient and feasible technical solution for diagnosing new energy vehicle reducer bearing faults.

Key words: new energy vehicle; reducer; stacked denoising autoencoder; gated neural network

摘要： 减速器是新能源汽车重要动力主要部件，而内置滚动轴承易发生各种故障，进而导致 重大交通事故发生，严重威胁行车安全。针对强背景噪声环境下减速器轴承故障诊断中深度 模型抗干扰能力不足、诊断精度受限、深度模型调参难度大等问题，文章通过堆叠降噪自编 码器（SDAE）的强抗噪特征提取能力与双向门控循环单元（BiGRU）的时序建模优势协同作 用，结合贝叶斯优化（BO）算法优化模型参数，提升诊断性能，提出了一种基于 BO-BiGRUSDAE 模型的新能源汽车减速器轴承故障诊断新方法。以新能源汽车减速器轴承的常见的故 障状态为研究对象开展实验验证，实验结果表明，所提方法的诊断准确率高达 98.75%，显著 高于其他主流神经网络模型。研究证实所提方法具有优秀的有效性和准确性，可为新能源汽 车减速器轴承故障诊断提供高效可行的技术方案。

关键词: 新能源汽车；减速器；堆叠降噪自编码网络；门控神经网络