In simple the model will be superior to other models for the same faults/problems.

1. A robot which employs MPC to walk at a home, is like a blind person who can only rely on the model of home it has already made in its brain. The MPC robot has no sense about its environment, except the pre-defined model it has in its CPU about that environment.

2. In control of a dynamic system, there is a trade-off between simplicity of the model and its accuracy. If the model is simple, the available control strategies (particularly for nonlinear systems), are simply in access and can be implemented with no excessive algebra. But for accurate modeling of a dynamic system, the state-space model becomes a bit complex and it requires lengthy algebra for controller design. Moreover, almost any mechanical dynamic system is built based on Newton's second-law of motion, which is itself an approximate model of reality. Newtonian model of motion is itself partly inaccurate, and when you ignore some of the parameters in your model, this inaccuracy would be exacerbated. Hence, the controller should be robust against model uncertainty (model inaccuracy), to be able to control the dynamic system with a good performance. If the controller is not robust, then Newtonian inaccuracy (approximated model), along with neglecting some other parameters that are ignored for model reduction, give way to a control process which is totally unsatisfying, because the model is not suitably known. This is why some control theorists are trying to develop model-free control strategies, it means control strategies which are not highly sensitive to the inaccuracy of the dynamic model available.

3. Controller design using transfer function in Laplace domain is somehow a model-free control strategy. Although we usually derive the transfer function of a second-order mass-sprashpot system by taking the Laplace transform of the ODE, but transfer function is usually obtained using experiments in frequency domain as Bode plot, while through the experiments we only characterize the relation between the input and output of the system in frequency domain (using Laplace variable). Hence, transfer function does not say anything about the physical structure of the plant which is to be controlled. Such input-output models are actually used for model-free control strategies, and are not based on the evolution model of the system (plant) described as an ODE, so they are not sensitive to model uncertainty (model inaccuracy).

Dear;

I agree with Siamak Heidarzadeh ;there is no such thing as a robust model. But the term "model" refers to the mathematical description of a physical system. Usually, there exists no guaranty a system is adequately modeled.

Regards

i am interested

There is no accurate model because a model is a mathematical representation of a physical system and there is always the consideration of mistakes in the model as a result of unmodelled dynamics or neglected parameters etc. Hence, a model cannot also be robust.

Although model-based control elements have shown promising outcomes in simulation and in theory, these generally worsened or had no effect in various applications. Generally, it is expected that the better performance could have been obtained with a more accurate estimate of model parameters. However, many exploratory tests have identified that the best performances were obtained by driving model-based contributions to zero.

Hence, the model-based controller design need not involve an accurate and or a robust model.