a Case on the Assistive Bed

An assistive bed was developed to address the challenges faced by long-term paralyzed patients who are unable to turn themselves. The system comprises three components: lying position data acquisition, turning intention model prediction, and assistive bed turning motion control. Appropriate distributed pressure sensors were selected based on an analysis of human lying positions to acquire data on body pressure distribution. Subsequently, K-means++ was employed to reduce the dimensionality of the high-dimensional data. A turning intention recognition model was established using a convolutional neural network-long short-term memory network (CNN-LSTM), with the whale optimization algorithm (WOA) applied for hyper-parameter optimization. A closed-loop control system was constructed utilizing an STM32 chip, a DC motor driver, and a nine-axis digital inclinometer to facilitate precise turning of the assistive bed. Design experiments involved male and female subjects performing assisted turning, during which the recognition of the lying position body pressure distribution data was evaluated. The experimental results indicated that the accuracy of the turning intention recognition model's predictions reached 98.94%. Additionally, to assess the model's generalization capability, a multi-dimensional open-source dataset was utilized for verification, yielding an accuracy of 94.83% for the turning intention recognition model predictions, which aligns with the anticipated outcomes.  By using the accumulated data of body pressure time series after the classification of lying position as the basis for assistive bed turning control, and changing the lying position in time in order to reduce the local pressure, the blood circulation of long-term paralyzed patients can be improved and pressure sores can be prevented.