TRANSFORMER FRAMEWORK FOR THE FAULT DETECTION OF A RECIPROCATING COMPRESSOR

Reciprocating compressors are crucial in the industrial sector, playing a key role in energy, manufacturing, and chemical engineering production processes. Effective health monitoring technologies can prevent failure risks, reduce unplanned downtime, and ensure production continuity. Although machine fault diagnosis technologies have developed rapidly, they still need improving with data collection, insufficient feature extraction, and weak generalization ability in practical applications. Traditional fault detection methods make it challenging to train models, especially when only healthy data and no faulty data are available. This study compares three unsupervised fault detection methods: autoencoders (AE), contrast abnormality detection (C-AD), and contrast one-class abnormality detection (COCA), focusing on which model is more compatible with the Transformer framework. The experimental results show that AE combined with the Transformer framework is more suitable for fault detection.