Optimization has received more attention in industry recently since it is a natural choice for reducing production costs and improving product quality and reproducibility without violating safety requirements and environmental regulations. The main bottleneck in applying standard optimization techniques at the industrial level is the fact that they rely heavily on an accurate mathematical model of the proces, which, however, is rarely available. The Laboratoire d’Automatique investigates various ways of using measurements in the optimization framework to combat uncertainty in the form of plant-model mismatch and process disturbances. When measurements are available, adaptive optimization, also called real-time optimization (RTO), can help adjust to process changes and disturbances.
Three different classes of RTO methodolgies can be distinguished, depending on how measurements are incorpated in the optimization framework:
- (i) model-parameter adaptation, where the measurements are used to reﬁne the process model, and the updated model is used subsequently for optimization,
- (ii) modiﬁer adaptation, where modiﬁer terms are added to the cost and the constraints of the optimization problem, and measurements are used to update these terms,
- (iii) direct input adaptation, where the inputs are adjusted via feedback control, hence not requiring on-line numerical optimization but off-line controller design.
Research at the Laboratoire d’Automatique has focused on the two latter classes. A process subjected to optimal inputs must satisfy the necessary conditions of optimality (NCO). However, due to uncertainty, the inputs computed using a process model might not meet the plant NCO. In the class labeled « Modifier adaptation », the optimization problem is modified iteratively using measurements in order to meet the plant NCO. In contrast, in the class labeled « Direct input adaptation », measurements are used to directly adapt the inputs so as to enforce the plant NCO.
Successful simulation studies have been carried out on several batch reactors, biotechnological systems, distillation columns, and polymerization processes. These studies show performance improvement in the range of 20-30% compared to the conservative strategies that would otherwise be necessary to ensure feasibility. The measurement-based methodology has also been tested on laboratory-scale setups such as a batch reactor with safety constraints, a biofilter for waste-water treatment, and a fed-batch fermenter growing Baker’s yeast. Furthermore, these strategies have been applied successfully at the indstrial level toan electro-erosion process and a batch emulsion copolymerization reactor.
Keywords : Optimization; Measurement-based Optimization; On-line Optimization; Run-to-Run Optimization; Real-Time Optimization; Modifier Adaptation;
Optimization has received more attention in industry recently since it is a natural choice for reducing production costs and improving product quality and reproducibility without violating safety requirements and environmental regulations. The main bottleneck in applying standard optimization techniques at the industrial level is the fact that they rely heavily on an accurate mathematical model of the process that, however, is seldom available. Research at the Laboratoire d’Automatique studies various ways of using measurements in the optimization framework to combat the lack of process knowledge and process variations. Successful simulation studies with this approach have been carried out on several batch reactors, biotechnological systems, distillation columns, and polymerization processes. These studies show performance improvement in the range of 20-30 % compared to a conservative strategy. The methodology has also been tested on some laboratory-scale setups such as a batch reactor with safety constraints, a biofilter for waste-water treatment, and a fed-batch fermenter growing Baker’s yeast. Also, an application of this strategy for electro-erosion at an industrial level is underway.
In an optimization problem with constraints (safety restrictions, operational limitations, etc.), the standard industrial practice is to use a conservative strategy. Process operation will be kept quite far from the constraints to reduce the danger of violating them. However, operating closer to the constraints shows enormous potential for performance improvement. In the proposed approach, this is achieved using measurements and can be summarized by the motto: squeeze and shift. Using adjustments based on measurements, the process variability is reduced (squeezed) and so the operating point can be shifted closer to the constraints without the risk of violating them. This way, the conservatism can be reduced, thereby leading to improved performance.
This research direction also addresses seeking the optimal solution that is not governed by the constraints of the problem. The use of on-line measurements towards the search of the optimum is studied.
- Measurement-based Optimization of a Batch Distillation Column
- Measurement-based Optimization of a Batch Reactor under Safety Constraints
- Measurement-based Optimization of a Fed-Batch Fermenter
- Neighbouring extremals for measurement-based optimization and predictive control
- Optimization of Solar Powered Heating Systems
- Tracking of Terminal Conditions using On-line Measurements
- Tracking the Necessary Conditions of Optimality
Run-to-run optimization methodologies exploit the repetitive nature of processes to determine the optimal operating policy in the presence of uncertainty. The goal of run-to-run optimization is to find iteratively the optimal operating conditions in the presence of uncertainty, while performing the smallest number of sub-optimal runs and preferably no unacceptable ones.
As with on-line optimization, constraints play an important role in optimization. However, the dependence of the input profile on the terminal constraints (i.e., constraints that depend only on the final condition of the run) is considerably more involved and poorly addressed in the literature. This research direction addresses various issues concerning the squeeze-and-shift approach in the context of meeting terminal constraints. Furthermore, the case when elements of the optimal operation are not determined by the constraints is also investigated.