Embedded Retraining for Conductive Textile Sensor Calibration and Compensation

In this paper, a data-driven approach is presented to calibrate and compensate for degradation effects in conductive textile sensors on microcontrollers. An experimental setup was constructed to control the elongation of textile strain sensors during prolonged use. The recorded course of resistance in combination with elongation and applied force was used to generate an initial model.

During the use of the textile sensor, degradation effects occurred, which altered the sensor characteristics. A machine learning approach on a embedded system was used to retrain the internal model without knowledge about the true or measured deformation. The performance of the initial model is compared to an online retraining algorithm for recognition and compensation without a connection to the sensors from the test rig or database. All adaptions to the internal model are calculated on a resource-limited embedded microcontroller. Furthermore, this paper highlights the current requirements and possibilities for such mobile calibration or re-training methods.