Model Predictive Control and the Monetisation of Control Performance

One way of monetising control performance is to evaluate the profitability or OPEX (Operating expenses) of the industrial unit with the control system activated, then deactivate the system and evaluate it again, and then subtract the latter figure from the former. The difference observed, i.e, the increase in profit (or reduction of expenses), can be explained by the reduced variability of the operating conditions that arise from the actions of the control system when the system successfully controlled.

But how does reduced variability leads to higher revenue? It allows the operating team to drive the process closer to restrictions, enabling the reduction of the product quality giveaway. The giveaway gap is the difference between the quality of the product and the quality specification. It means that the manufacturer produces products of better quality than needed, which has an effect on the cost (lower yield, more energy, higher temperature, more reflux, etc.). Restrictions related to safety also restrict profitability, e.g., the maximum temperature and pressure admissible by a chemical reactor nay restrict conversion. Hence a well-engineered control system is able to maximise operating revenue of the plant through the expansion of the range of feasible and stable operating points (OP), which can be sustained without producing off-spec products or compromising safety.

31 Any configuration of the control system is able defines the sustainable OP range up to a certain degree, but some types of controllers yield broader ranges than others and for the same control system configuration, the tuning parameters being used also affect the OP range. Most of the methodologies developed for the IPDCF of chemical processes have feedback controllers such as PID (proportional–integral–derivative) embedded in the analysis. PIDs are SISO (Single-Input Single-Output) feedback controllers, which normally operate with a single setpoint (SP), and are standard in the chemical industry as well as numerous other applications. Ideally, SPs are set at each variable’s economic optimal OP, as defined by the project team.

In practice, for multivariable problems, this approach introduces a well-known conflict between control goals. The issue arises as each PID tries to keep its controlled variable (CV) at the required SP without regard to disturbing other variables. Since most systems do not possess enough degrees of freedom, either due to the lack of manipulated variables (MV) or saturation of control elements, meeting all control goals simultaneously is frequently impossible (González and Odloak, 2009). Advanced “decoupling”

techniques can help lessen the problem, but nevertheless, interactions between controllers inevitably impose limits to feasible OPs. In short, one should look at the bigger picture and consider the problem of interactions between SISO controllers to define optimal SPs.

Model Predictive Control (MPC) is a more powerful kind of control structure that has matured for almost four decades of development in which it has been widely implemented and recognised. MPC control schemes are popular solutions to meet the control requirements of complex chemical processes due to their capacity for handling multivariable systems with the inverse response, as well as time delayed and highly nonlinear systems. MPC is a far more suitable for use with MIMO (Multiple Inputs Multiple Outputs) systems with strong interactions since it controls simultaneously all variables and minimises the global error according to control goal prioritisation (Morari and Lee, 1999). More information concerning MPC is provided in Section 2.2.

The MPC packages may replace PID controllers entirely in some processes, but they are more commonly encountered operating in conjunction with them. In this arrangement, MPCs normally occupy a higher position in the control hierarchy and provide SPs to PIDs (Scattolini, 2009). Moreover, by introducing economic goals in their objective functions or integrating with a Real-Time Optimisation (RTO) layer, some MPC schemes are designed to drive the plant as close as possible to the economically optimal operating point (Limon et al., 2013). Fig. 1 presents the standard hierarchy of

32 control systems found in many industrial operations (Zanin et al., 2002), which includes process instrumentation (sensor and valves), the regulatory control system, i.e., basic control loops (PIDs), and the advanced control systems such as MPC and RTO.

Fig. 1 – The typical hierarchy of control systems.

In IPDCF methodology presented in this work, the structure of control systems presented in Fig. 1 is assumed to be present in the future industrial unit, which means that the control system being engineered should be able to drive the plant to the optimal OP.

To this end, it is especially important in this scheme the presence of the MPC layer, which may have by itself economic optimisation capabilities or work in conjunction with an RTO system - it does not matter the exact layout. Fig. 2 presents a simplified scheme to illustrate the economic benefits adding such an MPC to the usual regulatory control. With only the regulatory control system activated, the operating team must set the operating conditions at a safe distance from key restrictions, according to the observed process variability. When the MPC is activated in conjunction with the regulatory control, variability is further reduced and thus it becomes safe to drive the process closer to restrictions.

33 Fig. 2 – MPC can be used to minimise the quality giveaway.

Hence MPC increases control performance monetisation by increasing plant revenue. Because of such interesting characteristics, incorporating MPC in the IPDCF is a desirable development which has sought after by a number of researchers, whose results are discussed in Section 2.3. Some of these works share with this Thesis the goal of optimising flowsheets jointly for both MPC and feedback controllers. This motivation is explored further in Section 1.4.