Journal of Quality Engineering and Management

Abstract Nowadays, in line with the rapid growth of image and video-based inspection technologies such as machine vision systems, applications of image-based statistical process control are found in a wide variety of industries and processes. Considering the spatio-temporal variation of the observations in an image-driven process monitoring, the purpose of this study is to use multivariate statistical process control methods in order to evaluate and analyze the quality of samples from a dairy production process. In this research, the color content of each pixel is identified in the standard RGB format, and then the color conversion is performed to the new three-dimesnional L*a*b* color space. After estimation of the spatio-temporal autoregressive (STAR) model, multivariate control charts are employed to monitor both mean and vrainace of the estimates. Change point analysis using likelihood ratio statistic and decomposion of control statistic have improved the interpretability of the out-of-control signals on the control chart. The results of a case study related to the dairy industry reveals the capability of the propsed method in recognition of out-of-control conditions using image processing and analysis of the product surface color. 

Abstract In image monitoring, the information contained in an image is evaluated using control charts. A spatial control chart is a type of control chart in which the horizontal axis represents the position within the image. These charts are used to detect abnormal points in an image. Thermovision is a branch of machine vision that deals with the analysis of infrared images. Although infrared cameras have long been used in preventive maintenance to identify faulty equipment, overloads, and loose connections, the images captured by these cameras are usually analyzed only through empirical methods, and the few quantitative studies conducted in this area have not utilized control charts. In identifying defects in power distribution panels, several challenges must be considered, including the variety of equipment used in electrical panels, the lack of sufficient data to train pattern recognition models, autocorrelation, and the complex behavior of heat transfer by radiation, convection, and conduction. The insufficient data for training pattern recognition models such as neural networks makes spatial control charts relatively more advantageous than these methods. In this study, a combination of spatial control charts and robust regression is employed to detect defects in power distribution panels, and the detection capabilities of various control charts for identifying these defects are compared.