1.7 When CCD May Not Be Worth the Additional Effort
Integration has a cost¶
CCD requires cross-disciplinary models, shared objectives, optimization expertise, data exchange, and organizational coordination. The integrated problem may be larger, more nonlinear, less transparent, and harder to verify than separate discipline models. Using CCD everywhere is neither efficient nor credible.
Conditions favoring a simpler process¶
Weak coupling¶
Controller retuning preserves the ranking of candidate plants, cross-sensitivities are small, and active constraints do not migrate. A modular sequential workflow may then deliver nearly the same system result.
Frozen or standardized hardware¶
If regulation, certification, supply chains, or platform reuse fix the plant, optimizing unavailable variables adds no value. The appropriate task is robust controller design for the fixed plant.
Mature interfaces with ample margin¶
Well-characterized subsystems may interact through interfaces designed with comfortable authority and bandwidth margins. Standard actuators and controllers can be cheaper and less risky than a custom integrated design.
Model uncertainty dominates the predicted gain¶
An optimizer may exploit details that the model cannot predict. If the expected CCD improvement is smaller than model-form error or manufacturing variability, the apparent optimum may be numerical fiction.
Implementation or certification cannot accept the result¶
An ideal trajectory may require unavailable preview, computation, sensing, or actuator rates. A plant optimized around that trajectory may perform poorly with the realizable controller.
The decision is low consequence¶
If the component is inexpensive, easily replaced, and weakly related to safety or lifecycle cost, the engineering effort may exceed the value of integration.
A minimum evidence rule¶
Before committing to full CCD, require evidence from at least one of these tests:
plant rankings change after controller retuning;
a sequential baseline is infeasible but joint changes recover feasibility;
an active constraint depends on both plant and control decisions;
system-level sensitivity shows a meaningful interaction;
architecture or information choices dominate performance; or
preliminary co-design gains exceed uncertainty and implementation margins.
Staged alternatives¶
Declining full CCD does not mean ignoring control. Lower-cost alternatives include:
control-aware plant requirements;
parameter sweeps with controller retuning;
alternating design reviews;
response surfaces of optimized closed-loop performance;
robust control over a family of plant designs; and
integrated optimization of only the few strongly coupled variables.