Data‐Driven Monitoring and Early Warning of Valve Plate Loss in Compressor Valves

ABSTRACT

This article addresses weak signals of valve plate loss in reciprocating compressor valves and limitations of traditional methods. A data‐driven early‐warning approach combining multidimensional distance‐metric‐based dimensionality reduction and a transformer model is proposed. Multi‐sensor signals are used to construct a four‐dimensional metric to evaluate feature importance, reducing 74 features to 10. Results show temperature rise in the faulty valve and an inverse relation with the opposite valve and validate vibration energy transmission. The method improves transformer performance under small samples, alleviates overfitting, and achieves accurate fault identification and reliable early warning.