Abstract: Although machine learning and artificial intelligence technologies have been applied to fault diagnosis in smart home appliances, they still face challenges such as data coupling between sensors, dynamic changes in data, and difficulty in obtaining fault data. To address the above issues, on intelligent famlt detection method for home appliances based on dual graph attention network and interval prediction is proposed. By simultaneously modeling the relationships between sensors and the dynamic patterns of data in time series, fault diagnosis can be performed through unsupervised learning, which does not require fault data for network training. This method achieved an accuracy of 92.3% in classifying normal and fault data on a real-world household air conditioning operation dataset. It maintained high accuracy even under noise interference, providing an effective and precise technical approach for enhancing fault diagnosis of smart home appliances. It is of great significance for promoting the development of smart home technology.

Key words: Fault diagnosis, Graph attention network, Interval prediction, Intelligent household appliances, Air conditioning, Sensor