This project focuses on integrating information theory with machine learning techniques to develop intelligent safety monitoring and quality assurance methods for chemical and process related manufacturing. It is a joint PhD program that will be supervised by Dr. Yuan Yao and Professor David Shan Hill Wong at National Tsing Hua University in Taiwan and Dr. Peter Green of the Institute for Risk and Uncertainty, University of Liverpool. Students with a background in chemical engineering, mechanical engineering, industrial engineering, statistics, computer science and other related fields are welcome.
Safe operation and quality assurance are two of the key challenges to sustainable development of chemical and process related industry. “Safety monitoring” refers to the detection of abnormal process behaviour, the diagnosis of the root causes of such behaviour and the assessment of potential risks. “Quality assurance” refers to the control of product quality and minimizing the risk of defects in delivered products. With the availability of Big Data and the advancement of machine learning, data-driven modelling has become an essential tool in both safety and quality assurances of chemical process operations (Qin 2012, Yan et al 2015).
However, in real-life applications, it is impossible for a fixed data-driven model to handle ever changing process behaviour. A truly “intelligent” approach must be able to distinguish whether the current scenario has been incorporated during past learning or whether additional information is embedded in new data. To handle this issue, adaptive modelling (Ma et al., 2009), local models (Pan et al., 2010), recursive learning (Ahmed et al., 2009) and just-in-time learning (Liu et al., 2012) approaches have been adopted.
Recently, information theory has been used to determine whether new data are sufficiently informative to justify additional deterministic model identification for a nonlinear dynamic system (Green et al., 2015). It is the objective of the current project to extend the above approach to the development of nonparametric, data-driven models that can be used for chemical process safety.
The proposed research will leverage the expertise in process safety monitoring of the group from the department of Chemical Engineering, National Tsing Hua University and the experiences in risk and uncertainty analysis, especially Bayesianmethods, of the group from the Institute for Risk and Uncertainty, University of Liverpool.
The Tennessee Eastman (TE) process (Downs and Vogel, 1993), created by the Eastman chemical company, is a realistic simulation of an actual chemical process, that can potentially ‘run-away’ and has a complete set of documented faults. The TE process has been commonly used by the process monitoring community for comparison studies / risk assessments of various schemes and methods. Additional examples of safety monitoring and quality assurances derived from realistic industrial processes (such as distillation columns in petroleum refineries, injection and transfer moulding in plastic industries, high-mix, multiple steps, parallel tool manufacturing lines in semiconductor and liquid crystal display industries) will also be provided.
Nonparametric models, such as but not limited to, Gaussian process models (Rasmussen and Williams, 2006) will be used as our machine learning tool. A framework will be developed which will establish, once a new set of measurements become available, whether they should be used to update (and therefore improve) the current model. This framework will be based on the concept of ‘highly informative training data’ that was introduced in (Green et al., 2015). This approach measures the information content of a new set of measurements, where informative measurements are defined as those which would make a significant change to the data-based model.
The successful applicant joining the project will receive training in chemical process engineering, safety monitoring and quality assurance, risk and uncertainty analysis as well as machine learning. The project will focus on generating new machine learning algorithms and applying these methods within the context of chemical process engineering.
For academic enquiries please contact Professor Scott Ferson [email protected]
, 0151 795 8039.
For enquiries on the application process or to find out more about the Dual programme please contact Miss Hannah Fosh [email protected]
Ahmed, F., Nazir, S., and Yeo, Y. K. (2009). A recursive PLS-based soft sensor for prediction of the melt index during grade change operations in HDPE plant. Korean Journal of Chemical Engineering 26, 14-20.
Downs, J., and Vogel, E. (1993). A plant-wide industrial process control problem. Computers & Chemical Engineering 17, 245-255.
Green, P. L., Cross, E. J., and Worden, K. (2015). Bayesian system identification of dynamical systems using highly informative training data. Mechanical Systems and Signal Processing 56–57, 109-122.
Liu, Y., Gao, Z., Li, P., and Wang, H. (2012). Just-in-time kernel learning with adaptive parameter selection for soft sensor modeling of batch processes. Industrial & Engineering Chemistry Research 51, 4313-4327.
Ma, M.-D., Ko, J.-W., Wang, S.-J., Wu, M.-F., Jang, S.-S., Shieh, S.-S., and Wong, D. S.-H. (2009). Development of adaptive soft sensor based on statistical identification of key variables. Control Engineering Practice 17, 1026-1034.
Pan, T.-H., Wong, D. S.-H., and Jang, S.-S. (2010). Development of a novel soft sensor using a local model network with an adaptive subtractive clustering approach. Industrial & Engineering Chemistry Research 49, 4738-4747.
Qin, S. J. (2012). Survey on data-driven industrial process monitoring and diagnosis. Annual Reviews in Control 36, 220-234.
Rasmussen, C. E., and Williams, C. (2006). "Gaussian processes for machine learning," The MIT Press, Cambridge, MA, USA.
Yan, Z., Huang, B.-L., and Yao, Y. (2015). Multivariate statistical process monitoring of batch-to-batch startups. AIChE Journal 61, 3719-3727.