Abstract
Ethylene cracking furnace pipe is usually manufactured by centrifugal casting process. Centrifugal casting process contains raw metal material melting and centrifugal casting. Single measurement data, including proportion of each metal element and casting temperature, can be collected for each batch, while data in time series, including surface temperatures at different locations of molding cylinder and rotational speeds, can only be obtained from centrifugal casting. Apparently, both types of data are important for final pipe quality. In this work, the span of batch process is determined according to the reading of rotation speed, therefore a batch process monitoring model is developed by multiway principle components analysis, MPCA, and principle component analysis, PCA, is employed for the feature extraction from single measurement data. Based on the cross validation by both models, a framework is proposed for the on-line centrifugal casting process monitoring. The best correct identification rate and the false alarm rate of this model are 100% and 0% respectively for studied process. The contribution plots are given to analyze forecasting results and identify root cause for failure batch, which can provide a reference for the improvement of pipe manufacturing and save the cost on penetration testing.
