With the development of digital information technology, nuclear energy systems are developing in the direction of intelligence and unmanned, which requires a higher demand for its safety, such as autonomous fault diagnosis. At present, the network structure model used in fault diagnosis usually needs professional design, which is time-consuming and labor-intensive, and the efficiency is low. To solve these problems, this paper proposes a data-driven adaptive fault diagnosis approach NSGAII-CNN. Firstly, the time-series data are mapped into two-dimensional images by Markov Transition Field, which preserves the time characteristics of the data and improves the fault diagnosis accuracy. Then, the NSGAII-CNN algorithm is proposed to realize the self-adaptive search of the network structure, which improves the construction speed of the fault diagnosis network structure model, thereby improving the diagnosis accuracy and efficiency. Finally, compared with the current three classical CNN architecture models designed by professionals, the methodology proposed in this paper has significant advantages in fault diagnosis and model structure construction. The proposed diagnosis method will provide operators with useful information and enhance the nuclear energy systems’ self-diagnostic capabilities.

随着数字信息技术的发展，核能系统正朝着智能化和无人化的方向发展，这对诸如自动故障诊断之类的安全性提出了更高的要求。目前，用于故障诊断的网络结构模型通常需要专业的设计，既费时又费力，效率低下。为了解决这些问题，本文提出了一种数据驱动的自适应故障诊断方法NSGAII-CNN。首先，通过马尔可夫转换场将时间序列数据映射为二维图像，从而保留了数据的时间特性，提高了故障诊断的准确性。然后，提出了NSGAII-CNN算法来实现网络结构的自适应搜索，从而提高了故障诊断网络结构模型的构建速度，从而提高了诊断的准确性和效率。最后，与目前由专业人员设计的三种经典的CNN体​​系结构模型相比，本文提出的方法在故障诊断和模型结构构建方面具有明显的优势。提出的诊断方法将为运营商提供有用的信息，并增强核能系统的自诊断能力。