A hybrid just-in-time soft sensor for carbon efficiency of iron ore sintering process based on feature extraction of cross-sectional frames at discharge end. (June 2017)

• Record Type: Journal Article
• Authors: Chen, Xiaoxia; Chen, Xin; She, Jinhua; Wu, Min
• Is Part Of: Journal of process control. Volume 54(2017)
• Page Start: 14
• Page End: 24
• Publication Date: 2017-06
• Subjects: Carbon efficiency -- Iron ore sintering process -- Multitask learning -- Just-in-time learning -- Back-propagation neural network (BPNN) -- Fuzzy c-means (FCM) clustering; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2017.01.006
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract : Highlights: The CCR was proposed as a metric for estimating the carbon efficiency. An image selection method was devised to select the KFs from the video taken at the discharge end of the sintering machine with an accuracy of 95%. An HJITL-SS was devised to estimate the CCR online. The simulation result of HJITL-SS, offline, and JITL-based methods using actual run data were thoroughly compared. A soft sensing system was devised, and actual run results show the validity in estimating the CCR. Abstract: Iron ore sintering is the second-most energy-consuming process in steelmaking. The main source of energy for it is the combustion of carbon. In order to reduce energy consumptions and improve industrial competitiveness, it is important to improve carbon efficiency. Reliable online prediction of the carbon efficiency would be extremely beneficial for making timely adjustments to the process to improve it. In this study, the comprehensive carbon ratio (CCR) was taken to be a measure of the carbon efficiency; and a soft sensing system was built to make an online estimation of the CCR. First, the sintering process was analyzed, and the key characteristics of the process parameters were extracted. Then, the configuration of the soft sensing system was devised based on the characteristics of the process. The system consists of three parts: an image selection, an image segmentation, and a hybrid just-in-time learning soft sensor (HJITL-SS). First, an image selection method was devised to automatically select the key frames (KFs) from the video taken at the discharge end of the sintering machine. Then, a genetic-algorithm-based fuzzy c-means clustering method was devised to extract feature parameters from the KFs. Finally, an HJITL-SS, which consists of online and offline submodels, was devised to estimate the CCR using the extracted feature parameters as inputs. Actual run data were used to verify the validity of our system. Accuracy, overfitness, and error distribution of the HJITL-SS, offline, and JITL-based soft sensing methods were compared, which show the validity of the HJITL-SS. The actual run results also show the validity of the soft sensing system with 97% of the actual runs are in an acceptable range.