Product quality prediction method in small sample data environment

Abstract

The low degree of enterprise digitization and the existence of personalized customization and small batch production manufacturing modes lead to the characteristics of small samples and high dimensions of data collected by manufacturing enterprises. The product quality prediction method based on big data is prone to overfitting in the small sample and high dimensional data environment, and the model generalization performance is poor. In order to solve the above problems, this paper conducts research in two aspects of data augmentation and model optimization respectively. At the data augmentation level, a data generation model RVAE-CGAN is proposed, RVAE constrains the value space of GAN and eliminates the influence of outliers in the original data on the quality of generated samples. Using DBSCAN to cluster raw data to generate conditional vectors for guiding sample generation, and finally obtain high-quality generated samples. At the model optimization level, a product quality prediction model PPO-SVR is proposed. The PPO method is used to optimize the hyperparameters of the SVR model and improve the prediction effect of the model. RVAE-CGAN and PPO-SVR models are combined, and RVAE-CGAN is used to expand the number of samples to train PPO-SVR. The experimental results show that the product quality prediction model constructed based on RVAE-CGAN and PPO-SVR outperforms the BP neural network prediction model trained on mixed data. Therefore, the combination of data augmentation and model optimization proposed in this paper can significantly improve the prediction effect of the product quality prediction model in the small sample data environment.