| Description |
Shape Memory Alloy (SMA) actuators are compact and have high force-to-weight ratios, making them strong candidates to actuate robots, exoskeletons, and prosthetics. However, these actuators are thermomechanical in nature and slow cooling rates can limit their performance. Electricity can resistively heat the SMA actuators very quickly to produce contraction. To improve the convective cooling, SMA wires have been embedded in vascular networks, allowing cold fluid to pass across the actuators and extend them faster. The vascular network can also deliver hot fluid to heat and contract the wire. To minimize the weight and size of the control hardware for the vascular and electrical networks, a scalable NxN architecture has been implemented that allows for 2N control devices to be shared amongst N2 actuators. This Network Array Architecture (NAA) allows each actuator to be controlled individually or in discrete subarrays. However, this architecture does not allow all combinations of actuators to be activated simultaneously; therefore a sequence of control commands may be required to achieve the complete desired actuation. This dissertation presents the development of an intelligent controller for large arrays of wet SMA actuators with electric and thermofluidic inputs. The controller uses graph theory to identify a sequence to control commands to optimize the performance of the actuators. By treating each actuator as binary (contracted / extended), the collected states of an actuator array can be represented as nodes of the graph and the discrete NAA control commands as the graph edges. By weighting the costs of the graph edges (actuation times, energy), graph theory algorithms can find a set of control commands to transition the array to the desired state with specific performance characteristics. NAA results in a multi-graph that has a large number of nodes (2NxN) and is highly interconnected, causing problems with scalability. The search algorithm has incorporated an expanding wavefront algorithm to construct only a small portion of the graph as needed. The computational cost to construct the graph has been minimized by using bitwise operations and the discrete nature of the array of binary actuators and the NAA control commands. The algorithm was implemented in MATLAB and it is able to identify the optimal solution for a 4x4 array with more than 14 million edges. By using an expanding wavefront, the algorithm, on average, explored less than 100 edges (<0.01%) in 0.03 seconds. A 6x6 array was optimized in 0.7 seconds, exploring just 2400 edges. |
