A Hybrid AI–Biomechanics Digital Twin to Predict Skin Damage in Prosthetic Sockets

Skin irritation and soft-tissue overload are major issues for transtibial prosthesis users, affecting 30 - 70 % of amputees and often resulting from dynamic changes in pressure, shear, temperature, and muscle fatigue. Existing monitoring systems typically rely on single sensing modalities and cannot anticipate skin-damage risk in real time. This paper presents a multimodal digital twin that integrates surface EMG, distributed socket-pressure mapping, temperature sensing, and IMU-derived gait features with a viscoelastic tissue model and learning-based predictor. A hybrid CNN-LSTM network fuses spatio-temporal sensor data to estimate both global and spatial skin-risk levels, while a rule-based module enables real-time prosthesis adaptation. A pilot evaluation on five transtibial users demonstrates that the proposed digital twin reduces high-pressure exposure (>55 kPa) by approximately 40 % and decreases early redness occurrences by 55 % compared with a pressure-only baseline. The results support the feasibility of real-time, multimodal digital twins for improving interface safety and comfort in lower-limb prostheses.