Statistical Monitoring of Multivariate Quality Profiles Using Correlated Gaussian Processes

Project: Research

View graph of relations



Due to the rapid development of sensing and information technologies, the manufacturingindustry has become a data-rich environment. A large amount of quality related process datahas become available. Traditionally, these data are in the form of simple random variables. Inthe recent decades, sophisticated inspection and sensory systems have started to generatequality profiles describing the functional relationship between response variable(s) andexplanatory variable(s). In the area of statistical process control (SPC), extensive research hasbeen done on monitoring the variations of these profiles. However, most of these works focuson univariate profile data, while in fact multivariate quality profiles are becomingincreasingly common. There are some significant difficulties in modeling and monitoring themultivariate quality profiles due to presence of both within-profile autocorrelation and inter-profiledependency, as well as the large-scale and high-dimensional nature of such data. Lackof effective tools for handling multivariate quality profiles seriously hinders its application inSPC, leaving the rich process information unexploited. In this project, we aim to address thisissue by developing a general and efficient modeling framework for multivariate qualityprofiles based on multivariate (correlated) Gaussian processes (MGP). We advocate thatMGP-based models are ideal for handling such data and can overcome the above difficulties.In this project, we intend to investigate some statistical properties of the proposed model.Some practical issues associated with implementation of the model will also be investigated.The developed methodologies will be applied to several real world applications.


Project number9042364
Grant typeGRF
Effective start/end date1/07/1622/12/20

    Research areas

  • Statistical Process Control , Manufacturing Quality Control , Profile Monitoring , ,