Abstract
Inkjet-printing as an on-the-go, inexpensive, and green method of creating instant flexible sensors and circuits will not proliferate until reliable device fabrication is possible outside the research environment. Shortfalls exist due to non-uniform fabrication/curing, environmental humidity/temperature influence, and uncontrollable deposition conditions, particularly in low-production setups. Electrical non-uniformity and variations from low-quality prints made by a minimally produced inkjet-printed sensor may be overcome by training a machine learning model to interpret the variabilities and output a high-confidence prediction of the signal. In this report, an inkjet-printed tactile sensor is modeled to simulate generate a rich data-set for training and testing an echo state network. The end goal of the reported work is to attach the echo state network to the imperfect, on-the-go, inkjet-printed sensor as an edge computing device, transforming the unreliable data into a more stable readout. In this way, the sensor design may be printed using any suitable inkjet-printer with minimal production effort and still extract reliable data. This enables inkjet-printers to be used at home by those in isolated/restrictive settings, poor communities, resource starved environments, or by enthusiasts. Applications include biometric, environmental, electro-chemical and -mechanical sensing, and the concept may be extended to inkjet-printed circuits for signal stabilization.