Hielscher, Jürgen ; Schlemmer, Sascha ; Hessinger, Markus ; Hatzfeld, Christian ; Werthschützky, Roland ; Kupnik, Mario (2017)
Situation detection in a powered lower limb orthosis.
Annual Meeting of the German Society of Biomedical Engineering and Joint Conference in Medical Physics. Dresden, Germany (10.09.2017-13.09.2017)
Konferenzveröffentlichung, Bibliographie

Kurzbeschreibung (Abstract)

Intoduction: Movement-assistive devices are designed to to recover lost abilities after injury or illness, compensate weakness or enhance human strength, sensitivity or accuracy. A situation detection subsystem can help to achieve a safe, reliable and intuitive human machine collaboration. We present the situation detection strategies to control a powered lower limb orthosis. The orthosis recognizes the current movement situation by evaluating the user’s posture, motion and muscle activity and adapts to an adequate level of power support. Methods: Our powered lower limb orthosis is designed to support the elderly in demanding movements, such as standing up from a seated position or stair climbing by providing an external torque to the knee. Integrated sensors measure ground reaction forces, angles in knee and ankle and muscle activities. We persue two approaches to detect the seven most relevant situations in human everyday life: A fuzzy-logic based algorithm evaluates the sensor signals and calculates the probability of each situation based on a predefined set of rules. The situation with the highest probability is chosen to set the level of support. The second strategy is a machine learning approach based on an artificial neural network. By training the network with offline sensor data the system extracts features and builds a user-adapted situation detection system. Both approaches are evaluated on data sets derived from young healthy subjects performing a range of tasks in a predefined order. Results: With both approaches, the seven most relevant situations for human mobility can be distinguished. Although all transitions between subsequent situations can be recognized without error, the moment of decisionmaking is crucial. A delayed support reduces comfort and reliability of the device. Both algorithms can be adjusted, considering the tradeoff between high dynamics and low error. Conclusion: The fuzzy-logik based approach results in a simple and concise program code, that can be executed in real-time on simple processors. The manual implementation and adaption of the fuzzy rules is complex and time-consuming. The machine learning approach requires more processing power but adapts automatically to the user by learning his movement patterns.

Frage zum Eintrag

Redaktionelle Details anzeigen