A Reliability Modeling Method Based on Logical Verification Generative Adversarial Networks

The balance of training samples is a critical factor affecting the generalization performance of reliability models. To address the scarcity of abnormal samples in civil aircraft operational data, this paper proposes a reliability modeling method based on logical verification generative adversarial networks(LV-GAN). The LV-GAN approach utilizes generative adversarial networks to generate abnormal failure data and establishes a mathematical relationship between monitored parameters and failures through fault logic diagrams, enabling logical verification of the generated samples. To demonstrate the effectiveness of the proposed method, LV-GAN, Kriging, BP Neural Network(BP-NN), Response Surface Methodology(RSM), and Support Vector Machine(SVM) are applied to reliability modeling and analysis of electric pump failure. Results indicate that the proposed LV-GAN method achieves the best overall performance in both modeling accuracy and efficiency, showing promising potential for engineering applications.