A Novel Time-Domain Fault Diagnosis Method With ELM for Aviation Intermediate Frequency Inverter

The fault diagnosis for the intermediate frequency (IF) inverter in the aviation industry has always been a difficult problem. However, most of the existing fault diagnosis methods are not directly applicable to it due to changes in system characteristics. In order to solve this problem, a new fault diagnosis model is proposed. Firstly, preliminary parameters are extracted from the measured current sequence. Secondly, two types of features values are obtained directly using the proposed current amplitude-deviation feature complement (ADC) method, which can effectively distinguish usual open-circuit (OC) fault modes of the inverter. Thirdly, the dataset consisting of the above feature values is acquired under different load condition and fault modes. Finally, based on low-dimensional feature values data, the extreme learning machine (ELM) is used to generate a fault diagnosis model rapidly through matrix operations. By experiments, the method proposed can accurately realize the fault diagnosis of aviation IF inverter. Compared with most existing methods, the proposed feature extraction method has advantages in fault discrimination ability and implementation difficulty. ELM speeds up the training process and facilitates debugging.