With Robust Control Toolbox™, you can capture not only the typical, or nominal, behavior of your plant, but also the amount of uncertainty and variability. Plant model uncertainty can result from:
Robust Control Toolbox lets you build detailed uncertain models by combining nominal dynamics with uncertain elements, such as uncertain parameters or neglected dynamics. By quantifying the level of uncertainty in each element, you can capture the overall fidelity and variability of your plant model. You can then analyze how each uncertain element affects performance and identify worst-case combinations of uncertain element values.
Using Robust Control Toolbox, you can analyze the effect of plant model uncertainty on the closed-loop stability and performance of the control system. In particular, you can determine whether your control system will perform adequately over its entire operating range, and what source of uncertainty is most likely to jeopardize performance.
You can randomize the model uncertainty to perform Monte Carlo analysis. Alternatively, you can use more direct tools based on mu-analysis and linear matrix inequality (LMI) optimization; these tools identify worst-case scenarios without exhaustive simulation.
Robust Control Toolbox provides functions to assess worst-case values for:
These functions also provide sensitivity information to help you identify the uncertain elements that contribute most to performance degradation. With this information, you can determine whether a more accurate model, tighter manufacturing tolerances, or a more accurate sensor would most improve control system robustness.
Robust Control Toolbox lets you automatically tune centralized and decentralized MIMO control systems. The controller synthesis algorithms include H-infinity and mu-synthesis techniques, nonsmooth optimization, and LMI optimization. These algorithms are applicable to SISO and MIMO control systems. MIMO controller synthesis does not require sequential loop closure and is therefore well-suited for multiloop control systems with significant loop interaction and cross-coupling.
Most embedded control systems have a fixed architecture with simple tunable elements such as gains, PID controllers, or low-order filters. Such architectures are easier to understand, implement, schedule, and retune than complex centralized controllers. Robust Control Toolbox provides tools for modeling and tuning these decentralized control architectures for plant models with uncertain real parameters. You can:
You can use this functionality to design a controller that is robust to changes in plant dynamics due to plant parameter variations.
Robust Control Toolbox provides several algorithms for synthesizing robust MIMO controllers directly from frequency-domain specifications of the closed-loop responses. For example, you can limit the peak gain of a sensitivity function to improve stability and reduce overshoot, or limit the gain from input disturbance to measured output to improve disturbance rejection. Using mu-synthesis algorithms, you can optimize controller performance in the presence of model uncertainty, ensuring effective performance under all realistic scenarios. H-infinity and mu-synthesis techniques provide unique insight into the performance limits of your control architecture, and let you quickly develop first-cut compensator designs.
The toolbox lets you model and analyze uncertainty in Simulink models. You can:
Robust Control Toolbox lets you automatically tune decentralized controllers modeled in Simulink for plant models with real uncertainty.
Detailed first-principles or finite-element plant models often have a large number of states. Similarly, H-infinity and mu-synthesis algorithms tend to produce high-order controllers with superfluous states. Robust Control Toolbox provides algorithms that let you reduce the order (number of states) of a plant or controller model while preserving its essential dynamics. As you extract lower-order models, which are more cost-effective to implement, you can control the approximation error.
The model reduction algorithms are based on Hankel singular values of the system, which measure the energy of the states. By retaining high-energy states and ignoring low-energy states, the reduced model preserves the essential features of the original model. You can use the absolute or relative approximation error to select the order, and use frequency-dependent weights to focus the model reduction algorithms on specific frequency ranges.