A novel framework for spatiotemporal monitoring and post-signal diagnosis of processes with image data

Advances in digital equipment and organizations' interest in having comprehensive and real-time information about products have increased the use of machine vision systems in organizations. In this paper, to monitor a sensory quality characteristic of a product based on images, the residual matrix of the intensity values of the nominal and captured images is divided into specific regions; then, the equality of the means of the regions is tested based on one-way ANOVA. To do so, a P-value-based control chart is applied to detect the out-of-control state as soon as possible. If an out-of-control alarm is received, Dunnett's test is used to identify region(s) with significant differences in the means of residuals (defective location[s] in the image) compared with other regions. After the locations of defective regions are identified, the change point of the process is estimated by using the maximum likelihood estimation approach. The performance of the proposed procedure is compared with some of the previous approaches in the literature. Then, the proposed procedure is implemented in a real-world case. The simulation study demonstrates the merits of the proposed procedure: It is not limited to specific geometric types of faults; it has the ability to detect multiple faults in the image; it takes less computational time; and it performs well in estimating the real time of change, as well as the location(s) and dimension(s) of the fault(s).