Research

Modeling is a vital activity in our research for the understanding of dynamics, the effect of external  factors and interconnections. In our lab, we focus on developing predictive rigorous or hybrid models using fundamental laws and informative data. The level of detail we consider depends on the end use of the model in a model based operation support technology. Extensive efforts are dedicated towards tailoring rigorous /hybrid models for online model based technology while keeping the physical knowledge intact.  By developing technology for model maintenance, model update/learning, we aim to address problems hindering the online use of rigorous/hybrid models in daily operation of processes. To this end, we consider the global identifiability problem,  reparameterization of unidentifiable parameters and their effect on control strategies and process monitoring in practice.

Moment Based Model Predictive Control :

Several distinct robust model predictive control techniques are established over the last decades([1,2]), which can be classified according to the modeling and treatment of uncertainty realizations. All of these mentioned approaches are yet to be adopted into the industrial applications. The essential reason for this gap between industry and academia is the huge deterioration of the closed-loop performance and the required computational power to solve the robust optimization problem. In order to surpass these problems, we have introduced moment based MPC formulations ([3,4]) which consider the linear combinations of the (centralized) moments of the uncertain (stochastic) predictions within the MPC problem. In this approach, the ability to discard rare-and-severe realizations is done by tuning parameters, namely weightings of the moments of the cost function. Hence these parameters balance the suppression of uncertain effects versus the on-average closed-loop behavior.

Figure 1. The performance of moment based MPC formulations (Mean, Mean-Variance, Mean-variance-Skew MPC) on a double integrator example

Figure 2. Schematic representaton of the MPC tuning approach.

Automating Model Based Controllers:

The performance of model based controllers and maintaining this performance require successful completion of several steps. These steps include experiment design, modeling, control structure selection as well as the MPC tuning. It is common that the performance of these controllers detetoriate over time.  For example, the closed- loop performance could be affected by a change in the plant dynamics or disturbance characteristics. Bringing the current model-based operation support technology to a higher level of autonomy requires developments in experiment design, performance monitoring and diagnosis, model maintenance, autotuning so that it can optimize plant performance under varying operational conditions and adapting to changing circumstances. In addition to model maintenance mentioned above, we also focus on developing systematic techniques in selecting the weighting matrices in the formulation of the MPC. Our tuning philosophy is expressed in the following figure.

Integration of Scheduling and Control in Real Time:

The decision making in large scale complex manufacturing systems such as chemical plants is distributed over several layers of automation.

In this hierarchical structure, the information flow is generally top to bottom (from scheduling to control). On the other hand, the information from the control level to the scheduling level is limited and irregular. This information channel is triggered only if it becomes clear at the factory floor level that it is infeasible to follow the desired trajectories/tasks. Industrial reality asks for more interaction (smart interface between control and scheduling).

We collaborate closely with Dr. ir. Ton van den Boom  from TUDelft and aim to increase information exchange between supervisory control and scheduling layers.

Figure 3. Isllustartion of Hierarchical Operation