Health Monitoring of Aircraft Air Conditioning Subcomponents Based on Bayesian Optimized SE-INCEPTION-LSTM

Traditional aircraft air conditioning systems typically follow a reactive maintenance approach, where components are repaired after failure, leading to increased maintenance costs for airlines. Some airlines adopt a policy of regular replacement of consumable parts to reduce the probability of unplanned maintenance and thus reducing costs. However, due to variations in operating environments, it is difficult to determine the optimal replacement cycle for consumable parts. Addressing these issues, this paper utilizes historical operational Quick Access Recorder data from aircraft air conditioning systems combined with the principles of the air conditioning systems. By employing the channel attention mechanism SE to optimize the INCEPTION module and integrating it with Long Short-Term Memory neural networks, a health working model for air conditioning consumable parts is constructed. Bayesian Optimization is used to autonomously select LSTM parameters to enhance model accuracy. With each flight’s data-combined with the health working model, the performance of current air conditioning consumable subcomponents is monitored and analyzed, enabling timely warnings based on performance degradation and providing effective support for maintenance planning. Experimental results show that this method can timely detect performance degradation of subcomponents with high interpretability.