基于迁移QCNN的孪生网络轴承故障诊断方法

轴承故障诊断对于降低旋转机械的损坏风险，进一步提高经济效益具有重要意义。深度学习在轴承故障诊断中应用广泛，但是深度学习模型在训练与测试时容易受到噪声的干扰导致性能下降。并且轴承的工况变化频繁，不同工况下的数据采集困难。对此，提出了一种基于迁移QCNN的孪生网络轴承故障诊断方法，先预训练QCNN获取具有较强判别性的模型参数，将预训练的参数迁移到QCNN作为子网络的孪生网络中，然后正常训练孪生网络获取模型，最后将测试数据与故障数据组成数据对输入模型，即可得到测试数据的故障类型。该方法将QCNN与孪生网络相结合，QCNN中的Quadratic神经元具有强大的特征提取能力，孪生网络共享权重和相对关系的训练方式，使得模型可以缓解噪声和工况数据不平衡问题的影响。实验结果显示，相较与传统机器学习模型和QCNN等模型，所提出方法在面对噪声和工况数据不平衡问题表现更好。

Twin Network-based Bearing Fault Diagnosis Method with Transfer QCNN

Bearing fault diagnosis is of great significance for reducing the risk of damage to rotating machinery and further improving economic benefits. Deep learning has been widely used in bearing fault diagnosis, but deep learning models are prone to performance degradation due to noise interference during training and testing. Moreover, the operating conditions of bearings change frequently, making it difficult to collect data under different conditions. To address this issue, this paper proposes a bearing fault diagnosis method based on transfer QCNN (Quadratic Convolutional Neural Network) and Siamese network. The QCNN is first pre-trained to obtain model parameters with strong discriminative power. Then, the pre-trained parameters are transferred to the QCNN used as a sub-network in the Siamese network. The Siamese network is then trained to obtain the model. Finally, the test data and fault data are combined to form data pairs input to the model, and the fault type of the test data can be obtained. This method combines QCNN with Siamese network, where the Quadratic neurons in QCNN have powerful feature extraction capabilities, and the Siamese network is trained with shared weights and relative relationships, which helps alleviate the impact of noise and imbalanced operating condition data. Experimental results show that compared to traditional machine learning models and QCNN, the proposed method performs better in dealing with noise and imbalanced operating condition data.