(Invited) Scanning Microscopic Light Scattering with Machine Learning for Quality Assurance of 3D-Printed Hydrogels

Abstract

Scanning microscopic light scattering (SMILS) is a powerful tool for the characterization of soft and wet matter, including polymer gels, polymer solutions, and the dispersion of nanometer particles. Specially the SMILS has a special advantage to overcome the inhomogenities of the soft and wet matter. For example, the network structure of polymer gels is so stable that the inhomogenities due to the fluctuation under the formation of chemical reaction, i.e. gelation, tends to be fixed in the network structure. This kind of the static inhomogeneites causes the position dependence of the observation of the polymer gels in the experiment of light scattering. The SMILS is able to scan the sample of the soft and wet matter and observes many data in different positions in the same sample to obtain the space-averaged data of the time-averaged correlation functions of scattered photons. Then the SMILS determine the ensemble-averaged correlation function.However, through the actual experiences of the SMILS experiments, we knew that the set of many data in different positions has several bad data due to dust, stray light, and mechanical vibration. By thinning the set of data to remove the bad data, the net result of the ensemble-averaged correlation function tends to improve very well. In the previous studies, we sometimes obtained the 30-100 data in different positions and thinned out the data to remove the 1-70% based on the judgement of the observer. There is a problem that the thinned data may depend on the observer.To overcome this problem, we tried to apply machine learning for the SMILS experiments. In the machine-learning procedure, artificial intelligence was made to learn the thinning operation of the observer as a teacher so that the same thinning operation as the observer could be performed. The below-shown figure shows one of the results of this procedure. It is shown that the raw ensemble-averaged correlation function has the ambiguous four peaks of relaxation modes. However the thinned correlation function by the machine-learning procedure will improve and show only one nice-seeming peak. We will exam the reliability of this novel automation procedure especially for the 3D-printed hydrogel samples and will establish the novel AI-aimed SMILS system to be applied for quality assurance of 3D-printed hydrogels in near future. Figure 1