High Throughput Rheology and Calorimetry of Reactive Resins Using Machine Learning Inference

During frontal ring-opening metathesis polymerization (FROMP), a polymerization front propagates spatially, driven by its exothermic heat of reaction. Only a small amount of energy is required for polymerization, thus offering a route to energy-efficient polymer manufacturing. Furthermore, recent studies have shown such highly efficient thermosetting polymers can be deconstructed or even repossessed. The low environmental footprint of these materials both in terms of energy consumption and reusability make FROMP an attractive manufacturing method for casting and direct ink writing of polymeric materials.<br/>The chemistry space of resins which support a frontal curing reaction is extensive given the vast number of monomers, initiators, and inhibitors that can be used to undergo FROMP. Although a few different monomers and formulations have been characterized, these spaces remain rather unexplored. Our work here aims to enable high throughput characterization of new resin formulations through two modules that can be easily integrated into a robotic lab setup. The first module aims to infer the viscosity evolution of the resins due to their background reactivity. This module uses the concept of protorheology, i.e. inference of rheological properties from nonuniform flow fields, to identify the time windows where these resins can be used for casting or direct ink writing. A series of vials filled with resin are flipped every 10 minutes and videos of the flow are recorded with an optical camera. A convolutional neural network trained with videos of fluids of known viscosity infers the viscosity of the samples from the recorded videos. The evolution of the resins’ viscosity is then tracked as a function of time to find the workable pot life. Although not as accurate as rheometers, this module can test several samples at once and can be simply integrated with a robotic arm. The second module aims to infer front speeds, thermal properties, and reaction kinetics. The module thermally initiates an array of 8 fronts in parallel. We measure the spatiotemporal temperature field during the reaction using an IR camera and the energy input with heat flux sensors. We use heat flux data, image analysis, and different regression models to calculate front velocities and infer thermal conductivity, specific heat, and heat of reaction of the resins. This is 30 to 100 times faster than differential scanning calorimetry, at a lower accuracy. We hope that the future integration of these modules into a robotic lab accelerates the material discovery to enable higher manufacturing efficiency, better reusability, and longer storage times of these promising polymeric systems.