New and improved on-line analytical sensors have been developed for monitoring of important polymer properties, namely an FTIR optical fiber probe, a coupled density-ultrasound velocity analyzer, a reaction calorimeter and a Micro Motion on-line sensor for conversion monitoring, an ultrasound sensor for the determination of conversion and solid content. Vibrational Viscometers for viscosity measurements and a NIR sensor for monitoring the end groups as well as particle size and PSD. A number of comprehensive unified mathematical models have been developed for the prediction of molecular and morphological properties in emulsion multicomponent batch, semi-batch and continuous polymerization reactors, including a model calculating the full particle size distribution in addition to the basic kinetic considerations, a model calculating the full molecular weight distribution of the polymer, a global model featuring a multi-component, multi-purpose structure and a model focused on product quality optimization in terms of composition and glass transition temperatures. The models have been successfully validated by both academic and industrial partners. An additional model was also developed for simulating special aspects of the vinyl chloride emulsion polymerization. Stacked Generalization and Bootstrap Aggregated Regression approaches have been implemented for improving prediction accuracy by combining several models. Neural network based techniques have been developed for estimating the amount of reactive impurity and reactor fouling for batch polymerization reactors. A range of linear and non-linear methodologies have been used for data analysis. The generation of robust models using bagging regression and data augmentation based bagging regression has been examined. Two estimation methods, namely the Extended Kalman Filtering (EKF) and the Receding Horizon State Estimator (RHSE) have been applied to batch and continuous solution polymerization reactors. Both linear and long range non-linear model predictive control methods have been tested, including DMC, IMC, MPC, TDNN, a GRNN-MPC plus PI control strategy and a GRNN with feed-forward action, for improving set point tracking and disturbance rejection performances. Selected predictive control techniques have been applied to industrial polymerization reactors of vinyl chloride and fluorinated polymers.
Intelligent manufacturing of polymers has provided methodologies and algorithms which can be successfully applied to various European Industrial Materials production and processing sectors, such as : a specification framework for on-line polymer quality sensors operating in an industrial environment, user-friendly software simulating multicomponent emulsion polymerisation in batch and continuous reactors, algorithms and software for reactor control using model-based and adaptive techniques, PLS methods and algorithms for the interpretation of both reactor sensor-based data and off-line analyser data. Statistical Process Control methods specifically structured for process reactor applications. Neural Network methods and software for the modelling (characterisation) and interpretation of both on-line and off-line sensor data, procedures to provide insight into process operation and potential malfunction from routinely monitored data using SPC and neural network approaches, algorithms and software for the application of reactor-based soft-sensors, inferential estimation and control, and Neural network based tools for the construction of process estimators and closed loop controllers. A particular spin-off from the programme is the range of generic process modelling, data characterisation, and control methods that may be widely applicable in the European process industries. These can provide rapid, cost effective development tools to help catalyse improvements in process monitoring, supervision and control.
In the continuous and batch chemical processing industries computer integrated manufacturing is being very actively promoted in both Japan and North America aiming at safer and more stable plant operations, productivity improvement, quality improvement, reductions in manufacturing waste and environmental impact, energy conservation and manpower reduction. All these need to be achieved in an integrated form. Integrated multivariable process control can have a significant strategic impact on polymer plant operability and economics. Polymer production facilities face increasing pressures for production cost reductions and more stringent quality requirements. The term "polymer quality" encompasses a set of structural characteristics of the final polymer product such as the molecular weight distribution (MWD), copolymer composition distribution (CCD), degree of branching distribution (DBD), particle size distribution (PSD), etc. These molecular and morphological properties determine to a great extent the end-use properties of the polymers. The main goals in operating a polymer reactor (high yield, better product quality and safe operation) are very difficult, if not impossible, to achieve without efficient and reliable polymer characterisation techniques. Although the weakest link in polymer reactor control schemes is undoubtedly the on-line instrumentation, lack of understanding of the process dynamics the highly sensitive and nonlinear behaviour of polymer reactors and the lack of well structured control strategies all contribute to the impairment of competitiveness. Appropriate process control technology and optimisation provide leaverage points for cost reductions and improvements in product uniformity by enabling processes to be operated closer to economic, plant and safety (both human and environmental) constraints.
Funding SchemeCSC - Cost-sharing contracts
20021 Bollate (Mi)
8022 AW Zwolle
92091 Paris La Defense
2726 Mem Martins Codex
5600 MB Eindhoven
NE2 4AA Newcastle Upon Tyne