Multi-source signals synthesized via CGAN as fused process signatures for space domain surface topography prediction

Accurate prediction of surface topography is critical due to its impact on the fatigue strength, contact stiffness, and friction properties of the part. Previous studies have focused on 2D geometric feature parameters, e.g., surface roughness prediction. However, roughness is not sufficient to accurately characterize surface topography. Therefore, a new data-driven approach based on conditional generative adversarial network (CGAN) and convolutional neural network (CNN) is proposed in this paper. This method utilizes the multi-source signals generated via CGAN as fused process signatures and applies to CNN for surface topography prediction. Specifically, CGAN generates cutting force signals and vibration signals related to different machining parameters and tool flank wear. And, the quality of CGAN-synthesized images is quantified via physical metric. The synthesized and real data as the augmented dataset for training CNN to address the lack of data in real production lines due to acquisition limitations. Further, the surface topography can be reconstructed via predicted value. The milling experiments were executed on TC4 to validate the feasibility of the proposed method. Spectral quality evaluation metric shows that CGAN can learn a unique data distribution for any combination of machining parameter and tool wear, which in turn generates higher fidelity cutting force and vibration signals. In terms of training the CNN using the augmented dataset to predict the surface height, the fused process signatures perform better with a lower RMSE of 0.00903 μm. The model correlation coefficient (R²) was 0.99383. These results demonstrate the effectiveness of CGAN in synthesizing process signatures concerning cutting force and vibration. And, using fused signatures for topography prediction can reduce error and improve accuracy, given that surface generation is associated with multiple data sources rather than relying on just one.