Machine Learning Assisted Design of MXene Aerogels for Personal Thermal Management

Thumbnail Image


Publication or External Link





Personal thermal management is necessary in maintaining body temperature in humans through the use of building insulation, personal garments, and heating or cooling units. Electrically conductive aerogels can be used as a multifunctional material, where the aerogel structure is intrinsically thermally insulating, and the incorporation of electrically conductive components allows for Joule heating of these materials for wearable heaters. Ti3C2Tx (MXene) has been incorporated in materials for Joule heating due to its excellent electrical conductivity. Cellulose and gelatin based aerogels have been used as bio-based materials with good structural properties in aerogels. Due to the large range of possibilities in parameters for aerogel formation, from percentage of components in each sample to sample concentration and presence or absence of glutaraldehyde, it can be tedious to test a matrix of recipes and determine the effects of each component on the electrical properties. To assist in the design of highly conductive aerogels machine learning was used as it uses a data-driven approach to analyze the effect of inputs, sample composition in this case, to predict a set of inputs that will return a desired output, which is a highly conductive aerogel.Aerogels of various compositions were fabricated and their resistances and sensitivities to applied pressure were measured to screen for highly conductive recipes and for strain insensitive samples. Of these samples, a strain insensitive sample recipe and a strain sensitive sample recipe were selected for Joule heating tests. Low voltages of 2 Volts and below, were applied to the aerogel samples and the temperature increase was measured. The stability of these samples under multiple heating and cooling cycles were tested both with and without applied compression. Through these tests we determined a strain insensitive aerogel recipe for stable temperature control regardless of pressure applied. This aerogel recipe was found to have a thermal conductivity comparable to common insulating materials at a much lower density. A machine learning model was then trained from the aerogel compositions and measured resistance values, and a prediction model with a low mean relative error of 19% was developed to assist in conductive aerogel recipe formulation.