Early fault diagnosis of petrochemical rotor imbalance based on multi-feature under asymmetry

To address the challenges of diagnosing early imbalance faults in petrochemical rotors, due to limited fault samples and subtle fault signals, a multi-feature-based method is proposed. This approach analyzes the characteristics of early imbalance data and leverages the sensitivity of multiple features to capture subtle differences, addressing hidden fault features. A fault diagnosis model based on a Long Short-Term Memory Network (LSTM) with residual connections and channel attention mechanism is developed. By enhancing feature learning and dynamically adjusting weights, the model effectively mitigates issues caused by sample imbalance, improving model generalisation and fault diagnosis performance. Validation with field data shows that, even under sample imbalance, the method achieves over 99% fault diagnosis accuracy, providing strong theoretical support for early imbalance fault diagnosis in petrochemical rotors.