AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Neural Network PID Control of a Variable-Stiffness Shoulder Exoskeleton with Load Prediction

Author

Term

4. semester

Education

Publication year

2025

Pages

81

Abstract

This thesis addresses how user‑intent‑aware control can be implemented to manage stiffness and support in a variable‑stiffness shoulder exoskeleton for overhead and static tasks. It develops and evaluates a dual‑layer architecture: a low‑level neural network‑scheduled PID (NNPID) controller that compensates torque and adapts to changing loads and joint stiffness by adjusting gains in real time, and a high‑level controller that uses an FMG armband with an LSTM network to interpret biosignals for load estimation and intention‑aware assistance. The exoskeleton and controllers were modeled and simulated in Simulink and demonstrated on embedded hardware (Arduino Due with a Maxon motor system). Experiments indicate improved adaptability under nonlinear dynamics compared to a conventional PID baseline, supporting the feasibility and benefits of intelligent control strategies in industrial wearable robotics.

Denne afhandling adresserer, hvordan intention‑bevidst styring kan implementeres til at håndtere stivhed og støtte i et skulder‑eksoskelet med variabel stivhed til over‑hovedet‑arbejde og statiske opgaver. Der udvikles og evalueres en dobbeltlaget arkitektur: en lavniveaustyring med neuralt netværk‑planlagt PID (NNPID), som kompenserer moment og tilpasser sig skiftende laster og ledstivhed via realtidsjustering af gevinster, og en højniveaustyring, der bruger et FMG‑armbånd, hvor biosignaler tolkes af et LSTM‑netværk til belastningsestimering og dermed intention‑afhængig assistance. Eksoskelet og styring blev modelleret og simuleret i Simulink og demonstreret på indlejret hardware (Arduino Due og Maxon‑motorsystem). Eksperimenter viser forbedret tilpasningsevne under ikke‑lineær dynamik sammenlignet med konventionel PID, hvilket understøtter gennemførligheden og fordelene ved intelligente styringsstrategier i industrielle bærbare robotter.

[This apstract has been generated with the help of AI directly from the project full text]