Data-driven parameter optimization for laser-induced coloration on stainless steel

Laser-induced coloration (LIC) on metallic surfaces has attracted interest for various applications. Achieving precise color generation still presents significant challenges, caused by the complex correlation between laser processing parameters and induced colors. Limited to parameter sensitivity, complex coupling of several processing parameters and unevenness of the coloring, it is difficult to achieve accurate parameter prediction and color control with traditional experimental or neural network based methods. Problems such as poor color acquisition and insufficient consideration of processing parameters exist in current studies. In this regard, a high-quality and full-pixel LIC color dataset processed by the Octree algorithm is established. In addition, modified ANN models and regulation cocktails are adopted to predict processing parameters (waveform, frequency, power, scanning spacing, scanning speed), and HSV (color model describes colors in Hue, Saturation and Value channels) values, respectively. The models proposed in this paper achieve accurate parameter control and color prediction. For Model MOC, its training and test RMSE reaches 7.82 and 8.33. Besides, precise RGB artworks were processed applying the predicted parameters. For Model MLP + C, RMSE reaches 4.83 for test set. The difference in color between the predicted color and the actual color reaches ΔE⁎ ≤ 5. With well-tuned ANN models and curated LIC dataset, the correlation between processing parameters and induced colors can be revealed. Thus, the processing parameters for LIC are optimized. •Uneven color in LIC is studied by full factorial experiment and microstructural analysis.•The Octree Algorithm enhances the quality of LIC dataset.•Five main laser parameters are considered in the experiment, suitable for application.•The adoption of regulation cocktails balances the efficacy and accuracy of the model.