当前制冷系统故障诊断算法先通过数据增强对原始数据进行扩充，再利用数据筛选算法过滤低质量样本，最后构建分类模型对故障类型进行预测，从而缓解由故障数据样本量稀少带来的诊断性能低下的问题。但是，这类方法需要人为设定阈值进行增强数据筛选，无法适用于自动化的工业生产中。而且，方法依赖增强后的数据训练诊断模型，忽略学习样本间的关联信息，导致模型过拟合。因此，本文提出一种基于数据增强和类特征聚合模型的故障诊断方法，将数据增强和故障诊断模型集成在一起，在减少模型构建过程中所需人工干预的同时保证诊断精度。同时，一个类特征聚合模块被嵌入到模型中以保证学习到同类样本特征彼此接近，从而提高故障预测准确率。实验结果表明提出的方法能够很好的应用于制冷设备的故障诊断。

It is difficult to obtain enough samples for fault diagnosis of refrigeration equipment, which limits the performance of fault diagnosis models. To alleviate this problem, existing fault diagnosis algorithms for refrigeration systems first augment data with the data enhancement technology, then utilize the data filtering algorithm to filter low-quality samples and finally predict the fault types,. During the entire process, it is necessary to manually set appropriate thresholds for filtering samples, which is not applicable to automated industrial production. Moreover, existing methods only learn class information from augmented data, which neglect learning the relationship between samples of the same class. To alleviate the manual intervention problem while preserve the high accuracy of the fault diagnosis model, a novel fault diagnosis approach based on data augmentation and class feature aggregation model is proposed, which simultaneously implements data enhancement and fault diagnosis. In addition, a class feature aggregation module is embedded into the fault diagnosis model to ensure that sample features of the same class are close to each other, thereby improving the accuracy of fault prediction. The experimental results show that this method can be well applied to fault diagnosis of refrigeration equipment.

大连市揭榜挂帅科技攻项目(2021JB12GX019)