Multimodal Process Monitoring Based on Dynamic Extended K-means-VAE-RF

With the development of customization and flexibility of manufacturing industry, intelligent manufacturing process gradually presents multimodal characteristics, which are reflected in the strong coupling, nonlinear, multi working conditions and other characteristics of the process. In this paper, a multi-modal process monitoring method based on dynamic expansion is proposed. In view of the dynamic characteristics of process data, the process data is dynamically expanded through the amplification matrix, and then the multi-modal process data is modal divided through the improved K-means algorithm. The divided multi-modal process data is monitored using Variational Autoencoder, and relevant statistics and control limits are designed. Finally, the multi-modal process monitoring is constructed by introducing the method of random forest for modal identification. The validity of the algorithm is verified by real data of steel rolling process.