Hierarchical Fault Diagnosis and Health Monitoring of Ammonia-Fueled Hybrid Power Generation Systems by an Ensemble Deep Learning Method

The hierarchical fault diagnosis and health monitoring is implemented by an ensemble deep leaning method, that is, by the ssGAN-CNN model for the fatal, major, and recoverable faults in the hybrid power generation system, The hybrid power generation system is composed of the ammonia-based hydrogen source (AHS), the proton exchange membrane fuel cell (PEMFC) and ammonia-hydrogen fueled internal combustion engine (AHICE). The health monitoring result shows that when the various faults appear in the AHS-PEMFC subsystem, the AHS-AHICE subsystem, and the hybrid power generation system. The hydrogen production and the power output are both significantly reduced by these faults such as the under-heating, catalyst deactivation, membrane degradation, and so forth, such that it is essential to identify these faults. The hierarchical fault diagnosis shows that when the fatal, major, and recoverable faults appear in different systems, the average diagnostic accuracy is 97.06% for the AHS-PEMFC subsystem, 96.02% for the AHS-AHICE subsystem, and 95.18% for the AHS-PEMFC-AHICE hybrid power generation system. Consequently the trained ssGAN-CNN model effectively identifies the complex fault scenarios and addresses the challenge that supervised models face with limited labeled data.