Fault diagnosis based on deep transfer learning for marine turbocharger

Abstract

The high cost, the inherent risk associated with fault simulation testing, and the complexity of diagnosing variable-speed operational dynamics present significant challenges for marine turbochargers. To address these issues, a fault diagnosis methodology that integrates finite element simulation with deep transfer learning algorithms is proposed for marine turbocharger in the paper. A vibration response model of the turbocharger system is established using the numerical simulation method. The accuracy of the numerical simulated model is validated through comparison with experimental data. Rotor imbalance and bearing wear faults are simulated, and variations in vibration response under different rotational speeds and fault severities are systematically analyzed. Subsequently, a fault diagnosis model based on deep transfer learning is developed to identify rotor imbalance and bearing wear faults. Feature transfer across network layers between source and target domains is achieved using the multi-kernel maximum mean discrepancy criterion. The results demonstrate that the numerical vibration response model achieves high accuracy, with a relative error of less than 5 % compared to the experimental data. Furthermore, the proposed deep transfer learning model effectively classifies rotor imbalance and bearing wear under variable operating conditions, achieving a classification accuracy of 99.76 %.