Product Quality Mornitoring Based on Data from Industrial Furnace Tube Centrifugal Casting Process
|School||Beijing University of Chemical Technology|
|Course||Chemical Engineering and Technology|
|Keywords||industrial furnace tubes centrifugal casting processmonitoring contribution plot|
Industrial furnace tubes are the core part of ethylene cracking unit, which include radiation section and convection section. Cracking reactions are carried out in the tubes of radiation section, and these tubes always suffer high temperature and low pressure conditions, which easily leads to carburization, corrosion, fracture etc. Hence, the quality of radiation tube plays an important role in petrochemical production process. Tubes in radiation section are usually manufactured via centrifugal casting process, by which compact structure, crystal integrity, less metal impurity and good mechanical properties can be obtained.At present, most manufacturing operation in domestic centrifugal casting companies usually depends upon individual experience of engineers or single variable control. But because of the complexity of the production process and the correlations among those existed influence factors, workers are not able to keep the products quality stable and find out key influence factors of quality failures.Production information used in this paper is collected from centrifugal casting process. Material data, centrifugal casting data and quality inspection data are included. Based on these data, centrifugal casting process is monitored by multi-statistical model, so tubes with defects are identified by its casting condition. Based on principal component analysis method, contribution plot is used to find out the key factors which contribute to the failure. Then, workers can adjust the production conditions in time to avoid quality failures in later casting. The enterprise can also improve product quality and reduce production cost. This paper mainly includes:(1) Raw material smelting process and centrifugal casting process are analyzed. Combining literature information, production experience with the feasibility of data collection, the type, method and frequency of the measurements are determined for establishing the model.(2) The production data are preconditioned in order to remove the outliers. Cross validation method is used to determine the number of principal components in quality monitoring model and the established model is used to analyze the production data.(3) In order to improve the model performance, the number of modeling data and different variable set are considered. With the application of contribution plot for SPE, the causes leading to quality failure can be identified. To the quality of tubs, with the comparison of the result of model detection and factory actual testing, the rate of correct identification of unqualified tubes by proposed model is86.67%, and the rate of false alarm by model is2.86%. This conclusion provides certain reference for practical production.