With advancement in technology, several changes can be observed in various technological equipments. With advancement in hearing aids technology, hearing disabled individuals are benefited abundantly. In this study, we aim to improve hearing aid technology, by proposing an advanced solution to one of the major problem of wind noise disturbance. In particular, we design a three-stage multi-block U-Net model to suppress the degradation with high quality reproduction of sound. We analyzed time-frequency domain-based audio representation analysis, and trained model on realistic noise for better user experience in hearing aids. The properties of wind noise, affecting the signal quality have been discussed deeply in addition to effect of wind noise for hearing aids users. Fully-convoluted two different blocks of U-Net were used in order to generate the proposed model, which outperforms existing models when evaluated with different performance metrics. The importance of deep learning methodologies in combination with hearing aids chips, and importance of realistic data for training an balanced model is also demonstrated in this study.