طراحي سيستم‌كنترل موشك به روش وارون ديناميك و شبكه‌ي عصبي

DESIGN OF A MISSILE CONTROL SYSTEM USING DYNAMIC INVERSION AND NEURAL NETWORK

در اين نوشتار طراحي سيستم كنترل به روش‌هاي وارون ديناميك و شبكه‌ي عصبي براي يك موشك پدافند هوايي برد متوسط مورد بررسي قرار گرفته است. براي اين منظور، ابتدا با استفاده از روش وارون ديناميك با يك و دو حلقه‌ي پسخور، كنترل‌كننده‌يي طراحي ‌شده است. سپس براي مقاوم‌سازي سيستم كنترل نسبت به نامعيني و ديناميك هاي مدل‌نشده، يك شبكه‌ي عصبي با لايه‌ي مخفي ــ براساس نظريه‌ي پايداري لياپانف ــ به كنترل‌كننده‌ي وارون ديناميك طراحي شده با يك حلقه پسخور اضافه شده است. در ادامه، عملكرد سه سيستم كنترل طراحي شده با در نظر گرفتن عدم قطعيت‌ها و تغييرات ضرايب آيروديناميكي بررسي و مقايسه ‌شده است. از شبكه‌هاي عصبي براي تبديل معكوس غيرخطي، كه در خطي‌سازي‌ پسخور كاربرد دارد، استفاده مي‌شود. قانون سازگاري وزن‌هاي پايدار براي شبكه‌ي عصبي هم‌زمان به‌دست مي‌آيد. نتايج شبيه‌سازي شش درجه آزادي پرنده، بيان‌گر برتري سيستم كنترل عصبي نسبت دو كنترل‌كننده ديگر است. در بين سه كنترل‌كننده، عملكرد سيستم كنترل وارون ديناميك با يك حلقه‌ي پسخور در پايين‌ترين جايگاه قرار مي‌گيرد.

Advances in fighter aircraft technology have created new challenges for designers of anti air weapons systems and the need for agile missiles. The dynamics of agile missile flying are intrinsically nonlinear and may vary rapidly with time. Furthermore, these dynamics are highly uncertain, since aerodynamic data for vehicles operating under such conditions are difficult to obtain or simulate, and may in fact be a poor approximation of the actual dynamics. These and other worries have prompted researchers to look beyond classical methods, which have historically dominated the field of missile autopilot design, to robust, nonlinear, and “intelligent” control techinques. Most nonlinear control techniques are based on linearizing the equations of motion by the application of nonlinear feedback in trim points. Known variously as feedback linearization, dynamic inversion or gain scheduling methods, they depend heavily on knowledge of plan dynamics. More recently, neural networks have appeared as means of clearly accounting for uncertainties in plant dynamics. This paper describes a hybrid approach to the problem of STT missile control in front of unknown nonlinearities and unmodeled dynamics. The mentioned controller has been designed with the help of neural networks and dynamic inversion methods. The dynamic inversion is a derivate of the feedback linearization technique, which has been one of the most popular methods during recent decades. The number of feedbacks in the Dynamic Inversion method is determined by the order of inputs. First, the controller is designed based on the dynamic inversion method with a single feedback loop and a double one. Then, the neural network with a single hidden layer is added to the dynamic inversion controller with a single feedback loop. Finally, it is applied to STT missile air defense via two stages. The six degree of freedom (6DOF) simulation shows that neural networks adaptively can cancel the linearization errors of the approximate dynamic inversion controller by a simple weight update rule derived from the Lyapunov theory. Using the Lyapunov theory guarantees the stability of the closed-loop system. In addition, results compare the above controllers and indicate that the performance and tracking error of the neural-adaptive nonlinear autopilot is better than other designed dynamic inversion controllers; and dynamic inversion with a double feedback loop is better than a single one.