A Stochastic Distributed Control Allocation Method Using Probability Collectives
Elliott, Christopher Michael
MetadataShow full item record
The configuration of the aerospace vehicle of tomorrow marches forward in complexity in step with technological advances in computation, materials, propulsion, and beyond. One of many attributes resulting from this evolution is the multiple effector concept, replacing the traditional aircraft design approach of a dedicated four channel mixer where pitch-roll-yaw-speed is controlled by elevator-aileron-rudder-engine, respectively. Instead, for example on the propulsion front, a distributed suite of smaller engines can substitute the larger power plant and can be used in concert for multiple axis control with both simultaneous, collective thrust for force generation and differential cyclic thrust for moment generation. The effector surface is also a candidate for distribution. Rather than one large moving panel, the notion of a smart wing with many small effectors has been explored due to potential benefits from structural mode suppression, aerodynamic optimization via active flow control, improved radar and observability signature, adaptive geometrical morphing to the flight condition, and more. Beyond these gains, the distributed approach also offers an element of resiliency in tomorrow's platform by reducing the critical impact due to the loss of a single dedicated effector. Additionally, the science of multi-agent systems is rapidly advancing and there is a growing interest in individual vehicles working together as a collective to synergistically accomplish a mission. In this work, the applicable collective mission is to jointly stabilize and control an aerospace platform, specifically a generic hovercraft where the multi-agent system is a distributed effector suite of small, electric engines. The method of solving for the required individual effector positions in order to achieve a set of required accelerations on a vehicle is referred to as the control allocation problem (CAP) and increases in complexity as a function of the number of effectors for allocation. With the trend towards a future generation of complex distributed effector suites on the horizon, traditional centralized flight control architectures of current day may become computationally intractable, and advanced methodologies for solving the CAP are warranted. This research focuses specifically on this problem and considers the underdetermined aerospace vehicle configuration where the CAP needs to solve for a number of effectors greater than the number of desired acceleration channels. The main contribution of this dissertation is the formulation of a novel distributed control allocation method for an aerospace vehicle using a modified formulation of Wolpert Probability Collectives stochastic optimization. The method is presented in detail and applied as a distributed flight control architecture for solving the CAP, where the multi-agent system applied is a collective of local effector controllers. Each individual agent is dedicated to a physical engine controller and is responsible for modulating the local thrust required to maintain overall vehicle stability with respect to translational and angular axes. Communication between agents shares the expected value of the current strategy in order to allow local objective evaluation and optimization. This work highlights the benefits of a distributed allocation approach from the advantage of parallel processing; to failed effector robustness; and the ability to evaluate nonlinear control effectivity for L2 optimization. A MATLAB Simulink toolbox is constructed to enable future incorporation into aerospace modeling and control environments. Finally, the method is applied to a hovercraft vehicle demonstrating the viability of distributed control allocation approach for inner loop stability augmentation.