Quality inspection is a critical component of compressor manufacturing, ensuring products meet stringent performance, reliability, and safety standards. Traditional inspection methods, which rely on fixed thresholds for parameters such as pressure, temperature, and vibration, face limitations due to production line variability and measurement inaccuracies. These challenges are further compounded by the scarcity of samples, making defect detection during online quality inspection particularly difficult. To address these issues, this paper proposes a data-driven features augmentation approach leveraging acoustic signals and deep generative models. By generating augmented features for both normal and abnormal samples, we enhance the training of quality inspection models, enabling more accurate and reliable defect detection even when real-world abnormal data is scarce. This method not only improves the robustness of online quality inspection systems but also demonstrates the potential of advanced machine learning and deep learning techniques in transforming quality control processes. The proposed framework offers a scalable solution for real-time quality inspection in compressor manufacturing.