Sensorless operation based on CGAN aviation permanent magnet starter generator inter-turn short circuit fault sample data missing interpolation strategy

In the field of aviation health management, deep learning-based fault diagnosis technology is regarded as a key means for enhancing aircraft operational safety and reliability. Consequently, a high-quality and complete fault sample dataset is considered an essential foundation for accurate fault prediction and health status assessment. Under varying operating conditions and fault severity levels, the initial fault stage, defined by a 1% to 5% short-circuit turn ratio, is not completely covered. This results in a lack of corresponding fault samples. A method for generating missing samples based on Conditional Generative Adversarial Networks (CGAN) is proposed to address this issue. In the CGAN model, conditional coefficients are set, and current data from both the normal state and partial fault degrees (such as 2% and 5%) are utilized to effectively generate high-quality samples of missing specific fault degrees (such as 1%, 3%, and 4%). Experimental results show that the generated samples effectively compensate for the gaps in the original data set. After being combined with the measured data, a complete sample set is constructed, which provides a reliable data foundation for the training of subsequent high-precision fault diagnosis models.