This study aims to enhance the geometric accuracy of layer external contours of 3D-printed parts by implementing a local shape-aware compensation approach.
Systematic experiments analyze the positional deviations of points along layer contours after printing, revealing inherent uncertainty, and reliance on local geometric shape. A novel local shape descriptor integrating Fourier descriptors and Hu moments measures the surrounding shape features at each contour point. Using a training data set, a Gaussian process regression model predicts shape-dependent point deviations and uncertainties. Based on the deviation predictor, a stochastic chance-constrained programming model is solved with K-means optimizer and Monte Carlo sampling to determine high-confidence compensation values for each point.
Experimental validation demonstrates the effectiveness of the proposed approach in mitigating deviation patterns arising from shape-dependent and uncertain printing behavior. Tested across various layer models, it reduces the average point deviation by over 60%. Comparative studies indicate that it outperforms state-of-the-art compensation approaches, especially for arbitrarily shaped contours.
This study introduces a point-wise compensation framework that incorporates local geometric shape in predicting deviation and compensation. Using a discriminative shape descriptor, uncertainty-aware machine learning and stochastic optimization, it offers a scalable, adaptive strategy to improve the dimensional accuracy of complex contours in additive manufacturing.
