Process Monitoring and Quality Control in Hot Rolling Processes Using Image Sensing Data.

Quality control in hot rolling processes has been a critical concern. The objective of this study is to thoroughly utilize an emerging new image sensing system for the online monitoring and diagnosis of hot rolling processes in order to improve the quality of hot rolled products. Three major research tasks have been conducted in the research for this dissertation. The first task is to classify different individual surface defects. The proposed new hierarchical weighted Support Vector Machine (SVM) classifier is designed to classify the defects in the image data that have the following challenges: low signal-to-noise ratio due to background noise and material textures mingled with defects' images; nonlinear separability between different classes of defects; unbalanced sample sizes in the training dataset; and high similarities among some defects classes. The GA-based feature selection method is used to select the efficient feature subset for improving the SVM classification's performance. The second task is to detect a repeating surface defect pattern that is induced by roll failure. An automatic detection approach is developed by integrating Canny edge detection, Hough transform, auto-correlation monitoring, and a spectral clustering method that is not sensitive to the noisy background or random defects. Furthermore, two robust estimators of Least Median of Squares (LMS) and Least Trimmed Squares (LTS) are applied for estimating the periodicity of the defects, based on which, the proposed diagnostic model can be applied to automatically identify cracked rolls. The third task is to fully utilize the available image data for monitoring the vibration of the rolling bar. The proposed contactless vibration monitoring system and signal extraction method are more suitable for the hot rolling process than the traditional vibration sensing and monitoring systems. The proposed vibration extraction model is based on four cameras, and it has a denoising and auto-calibration capability. The proposed model provides an innovative vibration sensing technology for hot rolling process monitoring. The methodologies developed here have been tested via either experimental tests or via actual rolling processes.Ph.D.Applied SciencesIndustrial engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/127206/2/3519651.pd