For many simple processes, PID control works well. But, for multizone processes where the heating zones interact, a model-based multizone control may prove more effective at minimizing temperature overshoot and thermal lag.

Multizone heated systems exist in many forms, including ovens, heated platens, and plastic extruding or injection molding equipment. Typically, temperature feedback control is used to regulate or drive these systems to the required processing temperatures. Effective temperature control is a challenge in these types of systems because each zone is not insulated from the effects of neighboring zones, creating interactions from zone to zone. In these applications, simple control may not be able to regulate the heater effectively.

Consider a two-zone heated platen used to control the temperature of a load (figure 1). It has two heater zones arrayed in concentric circles and two corresponding temperature sensors. The position of the heaters and sensors creates a dynamically coupled, multi-input, multi-output (MIMO) system, where the effects of one heater zone are seen at both sensors. In practice, the platen must operate under varying temperature ranges and loads and adjust to disturbances due to load entry and exit.



Figure 1. A two-zone heated platen is an example of a dynamically couple multi-input, multi-output system requiring multizone temperature control.

Limitations Using PID Controllers

The traditional approach to temperature control for multizone systems is to use PID control independently for assigned heater/sensor pairs. In the two-zone platen example, the center heater would be paired with the center sensor and the outer heater with the outer sensor. In general, PID control is beneficial because:

  • It has a simple, fixed control structure with few design knobs that must be tuned.

  • It is readily available from many suppliers.

  • It usually will work, to some degree.

For multizone applications, however, PID control is lacking in that:

  • It cannot account for dynamically coupled MIMO systems. PID control's single-input, single-output (SISO) structure does not allow coordinated control of multiple, independent heater commands. The single-input, single-output response results in opposing, interacting command signals that reduce the achievable response speed and increase heater saturation, producing a corresponding loss of temperature control.

  • It cannot account for a process's high-order dynamic responses.

  • It allows only limited tuning capabilities due to its fixed control structure.

  • It cannot account for disturbance dynamics.

Using a PID control on a multizone system may demand some compromises in performance. Because settings for one zone will affect the others, significant deployment time is needed to select controller gains. Also, the achievable response speed must be reduced to avoid interactions with neighboring PID loops and corresponding heater saturation, and the resulting loss of control. In addition, the response speed must be reduced to account for load-mass variations. Finally, it may be necessary to design a custom feedforward control element to account for load entry, processing and exit disturbances.



Using Model-Based Controllers

Model-based control techniques offer a means of improving process performance when traditional PID techniques and discrete logic do not meet performance criteria. They utilize a dynamic model of the process and disturbances to account explicitly for the dynamic responses (and interactions) between actuated inputs and sensed process variables. Unlike PID control, which is constrained to three dynamic control parameters and a SISO system structure, a model-based control technique embeds explicit dynamic information about the process to be controlled in the control structure, making it better suited to dynamically complex MIMO systems.

A model-based design methodology can account for multiple design criteria specifically tailored to the process performance requirements -- output regulation or tracking, load disturbance rejection and stability robustness, to name a few -- and can be adapted for different system configurations and control objectives such as feedforward control for disturbance rejection. The resulting controller can outperform the PID controller because it uses explicit process- and disturbance-specific dynamic models.

Model-based approaches systematically address problematic process characteristics that can cause traditional PID approaches to fail. These characteristics include:

  • Dynamic coupling between multiple inputs and multiple outputs.

  • Measurement noise.

  • Process disturbances.

  • Process nonlinearities.

  • Process dynamic variations.

Used in many industries including semiconductor and petroleum processing, model-based controllers were not widely available off-the-shelf until recently at costs near the range of high-end multiloop PID controllers. Those that were available typically were custom designed by specialized control system engineers. Today, however, there exists a handful of suppliers that provide generic model-based controllers of varying complexity and cost.



Figure 2. The dynamic coupling and process dead time in the two-zone platen system can be seen from its open-loop bump test.

PID vs. Model-Based Control

As an example of the performance available with PID vs. model-based control, consider the two-zone heated platen displayed in figure 1. The dynamic coupling and process dead time can be seen from its open-loop bump test results (figure 2). In this test, each heater input is stepped from a static value, held for a duration, then stepped back to its original value. The dynamic coupling in the system is apparent, with each zone input having a significant effect on both sensor outputs. From figure 2, it can be seen that the first heater zone has half the effect at the second sensor than at the first; the second heater zone has more than half the effect at the first sensor than at the second.



Figure 3. Temperature response from the interior and exterior sensors of the two-zone platen system vary depending on whether PID or model-based control is used. Model-based control reduces temperature overshoot, settling time and interaction from neighboring loops.
Figure 3 shows the closed-loop response of the two-zone platen system with PID and model-based control. The PID control is hindered by the process dynamic coupling and time delay, resulting in temperature overshoot, settling times and interactions from neighboring zones. In the initial reference step, with PID control both outputs overshoot dramatically as a result of interactions in the same direction. In the second reference step, where only one reference is changed, these interactions are in opposing directions, resulting in equally degraded performance.

By contrast, the model-based response is able to account for the process interactions and time delay, resulting in responses that are uniform, damped and consistent -- even when individual references are changed.



Choosing a Control Structure Type

When do you need to use model-based control instead of PID to solve a multizone temperature control problem? In general, the type of control structure chosen depends on two conditions:

  • Scope of the control problem to be solved.

  • Controller development and deployment budget.

The scope of the control problem to be solved depends on the level of temperature system complexity (high or low) and its performance requirements (stringent or lax). System complexity relates to the input/output structure (SISO vs. MIMO) of the system to be controlled and the complexity of its dynamic behavior (level of dynamic coupling, nonlinearities and variability). Performance requirements usually relate to steady-state and transient response characteristics for reference error or load disturbance attenuation and stability robustness. Higher levels of system complexity or performance requirements demand more sophisticated control structures.

The development and deployment budget depends on component availability (off-the-shelf or custom control structures), availability of process or control system engineering, design tools, process hardware and control software integration, process availability, programmability, instrumentation, and solution timeline and budget. Low levels of component and engineering availability, short timelines and limited financial budgets require simpler control structures.