A robot paint film quality prediction method based on adversarial learning

Abstract

This invention discloses a method for predicting the quality of paint film in robotic spraying based on adversarial learning, belonging to the field of intelligent manufacturing and industrial robot spraying technology. The invention first obtains spraying process parameters and paint film quality indicators, completes data preprocessing and dataset partitioning, and then constructs a conditional adversarial regression network containing a generator and a discriminator. A two-stage training strategy is used to train the network: the first stage pre-trains the generator using supervised regression loss, and the second stage introduces a discriminator to conduct adversarial joint training. Finally, the trained network outputs the paint film quality indicators corresponding to the process parameters to be predicted. This method can achieve feedforward prediction of paint film quality before spraying operations, effectively balancing model prediction accuracy and generalization performance, improving the stability and reproducibility of the training process, providing quantitative support for optimizing spraying process parameters, and adapting to the quality feedforward control requirements of automated spraying production lines.

Images