Machine learning assisted hybrid transduction nanocomposite based flexible pressure sensor matrix for human gait analysis

Human gait analysis strongly correlates with critical health metrics and provides significant information about physiological well-being. Therefore, accurate, fast, and cost-effective gait monitoring is required for intelligent healthcare systems. This paper reports the development of a flexible hybrid transduction Barium Titanate (BTO)/SU-8 nanocomposite-based, individually addressable pressure sensor matrix. The proposed sensor is highly suitable for wearables compared to the conventional pressure sensors due to its speedy and cost-effective design flow and ease of operation. The hybrid (piezoelectric/triboelectric), photo-patternable active layer enables strain and contact electrification-based sensing that convolves into a highly sensitive, lower cross talk and large area pressure sensing. The reported sensor is incorporated with a solder-free modular data acquisition setup for a straightforward design integration. A pressure sensitivity of 34 mV kPa-1 for the deep linear region and 2.7 mV kPa-1 for the linear region over a pressure range of 0–170 kPa is reported. The sensor shows excellent reliability and negligible hysteresis with an average deviation of 2.7 %. Furthermore, the 36 pressure cells with hybrid transduction deliver rich feature extraction to machine learning algorithms compared to single transducer-based systems for an accurate gait and grip strength monitoring. The developed convolution neural network (CNN)-2D model gives a model accuracy of 98.5 % and 98.3 % for two different gait characterizations, while delivering a model accuracy of 93.75 % for grip strength assessment. The combination of hybrid sensor design, development, and use of machine learning offers a novel approach to tackle the issues associated with sensors that are incompatible with rapidly developing smart healthcare technology.