A computational approach for decentralized control of turbine engines

We propose a heuristic approach to approximately compute the optimal decentralized control for linear systems. The method exploits the notion of quadratic invariance, which characterizes a class of convex problems in decentralized control design, and extends the application to general unstructured models. The plant model is approximated such that the decentralized information structure is quadratically invariant under the approximate plant. Then the optimal design is efficiently found via convex optimization, and it is applied back to the original full plant. A simple convex condition to prove the closed loop stability in this setup is presented. The method finds a satisfactory decentralized control design efficiently, and furthermore, the resulting design can be used as a good initial point for local optimization algorithms. A numerical example on a simplified turbine engine model is presented for demonstration.