A simulated parameter optimization method–based manifold learning for a production process

Summary A production process parameter optimization method based on feature extraction for manifold learning is proposed to achieve precise optimization of steel anomaly data of different grades in the same series and to improve the quality of industrial products. First, the appropriate neighboring samples are found in the state of the sample point and the next state to form the neighborhood matrix. Then, the manifold hidden inside the data is extracted, ie, the evolution trend of the process parameters between different brands. At the same time, a monitoring model is built with the training data based on the support vector data description (SVDD). If an outlier is detected, it will be projected onto the manifold to obtain the adjustment values. Thus, the outlier can return to the normal state. The Swiss roll and actual production data of interstitial‐free (IF) steels are employed to verify the effectiveness of the proposed method. The results show that the new method considers the continuity of process parameters of different product grades in the production process and uses data to extract the potential manifold, ie, using the evolution trend of process parameters among different product grades to achieve the optimization of the process parameter. The proposed method provides a new process parameter optimization method for the actual production process.