Abstract: Traditional fault diagnosis methods for helium leak detection equipment provide limited accuracy due to the inability to intuitively represent fault characteristics and heavy reliance on experts. A time series image generation scheme combined with the convolutional neural networks (CNN) is proposed for the sensor data of helium leak detection equipment, which performs data-driven analysis and fault classification on the fault data. Thereby, the faults of helium leak detection equipment can be diagnosed more accurately. Firstly, a more intuitive grayscale transformation was proposed to convert one-dimensional time series signals into grayscale images for visual representation, in order to adapt to different helium leak detection equipment fault diagnosis scenarios. Then, the LeNet-5 CNN model was optimized, and LeNet-5s CNN model with faster computing speed was designed to maximize recognition accuracy, and a dual model detection method was proposed for complex detection tasks. Finally, the effectiveness of the algorithms was verified through experiments. The results show that the improved model has a certain improvement in accuracy compared to the pre-improved one, and the dual model detection method has good recognition performance for all three types of faults in this complex detection task.

Key words: helium leak detection equipment, time series data, data processing, fault diagnosis, classification, deep learning