طراحي شبكه موجك فازي پيش‌بين براي حركت مفصل زانو به منظور كاهش مصرف انرژي متابوليكي در استفاده از اسكلت خارجي نيم‌تنه پاييني

Active lower-limb exoskeletons have been developed to enhance human physical capability an‎d reduce the metabolic cost of walking. However, inherent time delays arising from sensory processing, algorithmic computations, an‎d electromechanical actuator response introduce a phase lag between the userʹs motor intention an‎d the robotʹs assistive torque, thereby degrading system assistance efficiency. The objective of this research is to develop a lightweight, adaptive predictive framework to compensate for control loop delay an‎d to quantitatively investigate its effect on metabolic energy consumption. To this end, in the first phase, a short-term predictive model based on a Fuzzy Wavelet Neural Network combined with an Auto-Regressive linear pathway (FWNN-AR) was designed to forecast the knee joint angle across a 50-ms prediction horizon using historical joint angle an‎d angular velocity data. To enable real-time adaptation to the userʹs biomechanical variations an‎d reduce computational burden, an online learning mechanism based on the "winner rule" strategy with 32 fuzzy rules was employed. Accuracy eva‎luation on the Camargo biomechanical dataset using RMSE an‎d MAE metrics demonstrated that in a multi-subject scenario under mixed locomotion modes, the joint angle prediction error decreased to 1.32°, with an execution time of 2.06 ms, which falls well within the real-time execution time budget. In the second phase, a predictive proportional-derivative (PD) controller was implemented in OpenSim on the Gait2392 musculoskeletal model. The effect of applying assistive torques in the presence of time delays ranging from 0 to 100 ms was simulated using the Computed Muscle Control (CMC) tool. The variation in metabolic power consumption across 26 key lower-limb muscles was eva‎luated using the Umberger metabolic probe, accompanied by qualitative validation via reserve actuator torques. Simulation results for 10 subjects indicated that the exoskeleton achieves an average metabolic power reduction of 14.11% in delay-free conditions. By increasing the delay to 50 ms (matching the prediction horizon), the system compensates for the delay an‎d maintains an 11.20% reduction in metabolic power consumption, whereas at a 100-ms delay, system efficacy degrades, leading to a 0.45% increase in metabolic power consumption. These findings confirm that employing an online predictive model provides an effective solution for preserving biomechanical synchronization an‎d improving the metabolic power efficiency of exoskeletons under operational delay conditions.

مريم ذكري

مرضيه مجدراصيل

مرضيه كمالي , حامد جلالي بيدگلي