Taking into account recent progress in sensing and information technologies, many researchers prefer to adopt automatic data acquisition techniques to collect complex data information. The condition is suitable not only for the conventional manufacturing aspects, but also for some new trends such as the service operation management, healthcare, and business operation management. Consequently, a large amount of quality-related data of certain processes has become available. In industrial engineering aspects, the development of machine vision system (MVS) makes collecting image-based data in rich-data environment and dealing with complex data structure possible. Because it has the ability to quickly provide information on product geometry, surface defects, surface finish, and other produc...[ Read more ]

Taking into account recent progress in sensing and information technologies, many researchers prefer to adopt automatic data acquisition techniques to collect complex data information. The condition is suitable not only for the conventional manufacturing aspects, but also for some new trends such as the service operation management, healthcare, and business operation management. Consequently, a large amount of quality-related data of certain processes has become available. In industrial engineering aspects, the development of machine vision system (MVS) makes collecting image-based data in rich-data environment and dealing with complex data structure possible. Because it has the ability to quickly provide information on product geometry, surface defects, surface finish, and other product and process characteristics. Then, some engineering and statistical approaches for huge data information make it available to simultaneously monitor multiple quality-related characteristics in data-rich environment.

Statistical process control (SPC) of such data-rich processes is an attractive tool for monitoring their performance. Whereas, most mature conventional SPC techniques prefer illustrating each process parameter using one-dimensional geometric measurement ( eg., length, width and diameter), it seems impractical and unreasonable in real cases that quality-related characteristics should be measured in two-dimensional planes and three-dimensional spaces. More specifically, the real responses consist of the profiles or surfaces indicating the shape or characteristics of manufactured items. The quality of a process is characterized by the relationship between a response variable and one or more explanatory variables.

This research concentrates on complicated and high-dimensional data structure, the image-based data. Objectives are firstly to reduce data complexity and dimensions, because the data information within one image is sufficiently huge, mostly containing millions of observations. The second target is then to identify faults and character variation within image and monitor an out-of-control (OC) process shift between image-to-image in online scenario, simultaneously. The technical challenges are that observations are collected from a massive data environment, and that secondly the monitoring process is targeted to apply in an online process. These assignments are never been involved in existing approaches.

Using both the spatial and the temporal properties of image data, we propose an online monitoring image-based scheme considering the autoregressive relationship of spatial image-based observations, which would confirm the contribution of regressive data in explaining some nuisance variability and smoothing unknown systematic noise. Even though it would provide more detailed diagnostic information, it is greatly defiant implementing this scheme in online condition. It need to trade-off between monitoring efficiency and identifying precision.

Therefore, the autoregressive relationship among observation is formulated by nonlinear regressive profile function, some apparent explanation variables are also involved in monitoring profile. Through estimating profile coefficients in Phase I, the residual term of profile function is our monitoring benchmark, which can significantly reflect a shift pattern. The performance of the proposed method is evaluated through computer simulations and experimental studies. The results show that our proposed spatiotemporal method is capable of quickly detecting the emergence of a fault. It also provides a good estimate of the change point and the size and location of the fault. Finally, we use a real LED inspecting example to prove our method is practicable and efficient.