Classification methods play a significant role in the fault diagnosis field. However, they cannot effectively recognize new types of fault data and improve their classification performance timely through learning the testing data, for they lack continual learning ability. Inspired by the continual learning mechanism of the biological immune system, we propose a continual learning classification method (CLCM) and apply it to equipment fault diagnosis. During the testing stage, it continually cultivates new memory cells and new types of memory cells through learning the testing data to improve its classification performance. It classifies the known types of data and clusters the new types of data. Experimental evidence on six well-known datasets from the UCI repository and ball bearing test data verified its effectiveness and superiority. Results show that it has better classification performance when the testing data include all types of data, and it outperforms the other methods when the testing data include new types of data, especially new types of labeled data. The fewer types of training data, the more advantages it has.

分类方法在故障诊断领域中起着重要作用。但是，由于缺乏持续的学习能力，他们无法有效地识别出新的故障数据类型，无法通过学习测试数据及时地提高分类性能。受生物免疫系统持续学习机制的启发，我们提出了一种持续学习分类方法（CLCM），并将其应用于设备故障诊断。在测试阶段，它通过学习测试数据来不断培养新的存储单元和新的存储单元类型，以提高其分类性能。它对已知数据类型进行分类，并对新数据类型进行聚类。来自UCI资料库的六个著名数据集的实验证据和滚珠轴承测试数据证明了其有效性和优越性。结果表明，当测试数据包括所有类型的数据时，它具有更好的分类性能；当测试数据包括新类型的数据，尤其是新类型的标记数据时，其性能优于其他方法。训练数据的类型越少，其优势就越大。