A transfer learning based deep neural network adaptive controller for the Furuta pendulum subject to uncertain disturbance signals

• Published in: Scientific reports Vol. 15; no. 1; pp. 24012 - 12
• Main Authors: Yilmaz, Firat Can, Onler, Recep, Tatlicioglu, Enver, Zergeroglu, Erkan
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 05.07.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 639/166; 639/166/987; 639/166/988; Accuracy; Adaptive control; Closed loop systems; Control algorithms; Controllers; Deep neural networks; Humanities and Social Sciences; Lyapunov stability analysis; Mathematical models; Methods; multidisciplinary; Neural networks; Performance evaluation; Robust control; Robust disturbance rejection; Science; Science (multidisciplinary); Stability analysis; Transfer learning; Underactuated mechanical system; Robust disturbance rejection; Deep neural networks; Lyapunov stability analysis; Adaptive control; Underactuated mechanical system
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-10021-1

In this study, a control algorithm that integrates a Deep Neural Network (DNN) adaptive control architecture with a robust disturbance rejection mechanism for the underactuated mechanical system referred to as the Futura Pendulum. The study has a hybrid learning structure that combines offline supervised pretraining of the DNN’s inner layers with an online adaptation law that updates the output-layer weights in real time. The adaptive mechanism is supported by a Lyapunov-based stability analysis, which guarantees the convergence of the tracking error and the boundedness of all closed-loop signals. The impact of DNN layer size on performance was investigated using standard indices (ISE, IAE, ITSE, and ITAE). The results show that the controller achieves stable tracking with minimal chattering, even under random disturbances. Numerical simulations validate the robustness, adaptability, and improved efficiency of the proposed control method, demonstrating its potential for real-time implementation in uncertain and dynamic environments.