Modeling and Optimal Adaptive Control of an Actuator Using Machine Learning Techniques with Long Short-Term Memory Integration

Author(s):

Richard McKee, D. Johann Djanal-Mann, Sun Yi, and Muthuvel Murugan

Abstract:

Adaptive control has been around for the past few years. Recent advancements in ML techniques have allowed these new techniques to be paired with traditional adaptive control to better track the controller output. A long short-term memory (LSTM) neural network has the unique ability to handle thousands of previous time steps, which makes it an ideal choice when using time-series data. There are several considerations for the design of a controller for an actuator, including nonlinearities such as hysteresis. The hysteresis is time dependent; therefore, the LSTM can capture the behaviors of the actuator with relative ease. By using experimental data, a neural network is trained and used as a controller to achieve robust control of an actuator.