​ Abstract: As a first step toward creating a simple evaluation system for table sliding and peg movements in upper limb rehabilitation, experiments were conducted using STMicroelectronics' micro electro mechanical system (MEMS) board, its evaluation kit, and SensorTile.box. The experiment is conducted from two perspectives, including horizontal movement on the desk corresponding to the table slide and three-dimensional movement away from the table corresponding to the rehabilitation peg. UNICO was employed to assess the operation software from STMicroelectronics was used to classify idle states, left-right motion states, and forward/backward motion utilizing decision trees. A sensor on a cloth detected features of the table slide during left-right motion, with three different speeds set: fast, slow, and stopped. To address the limitations of the USB cable connection imposed by the geometry and experimental setup, we also utilized SensorTile.box, which connects via Bluetooth and stores data for analysis on the built-in micro SD card. An accelerometer and a gyro sensor were used for the measurements. The results of the confusion matrix differed depending on whether both accelerometers and gyros were employed. A comparison of the accuracy of feature detection by machine learning cores created solely from various accelerometers revealed that the peak-to-peak core was more accurate. Sensor trajectories were obtained from the calculations. It was observed that once stationary, movements back to the origin did not calculate accurately at slow speeds due to errors in trajectory calculations, although repetitive movements were confirmed. However, repetitive movements were confirmed.

Keywords: MEMS sensor, Upper limb rehabilitation, UNICO, Table slide exercise, Machine learning, Decision tree, Imufusion.

_________________________________________________________________________________________