The synthesis of nanomaterials is extremely sensitive to various factors under experimental conditions. Therefore, for controlling synthesis, it is important to ascertain comprehensively the relations between the conditions and nanomaterial properties. This study is intended to acquire the relations in data sets from combinatorial syntheses by means of an artificial neural network-based method toward property optimization. Recently, 3404 data sets were obtained systematically using microreactorbased combinatorial CdSe nanoparticle (NP) syntheses for examining conditionproperty relations. However, it is time-consuming to acquire the relations for the following reasons: (i) massiveness and complexity of the multivariate data sets, (ii) small numbers of points permitted for each experimental parameter to avoid 'combination explosion', and (iii) errors and missing data attributable to experimental reasons. In this work, an NN-based data analysis method was developed and applied for analyzing the data sets to acquire the relations. In the method, an exhaustive 1600 training processes and the following ensemble approach are performed for obtaining preferred NNs. Results show that NNs extract essential patterns on the condition-property relations on a realistic time scale. The trained NNs are capable of predicting the NP properties even for new experimental conditions with high accuracy. Moreover, data interpolation and sensitivity analysis based on the NNs provide us the relations as accessible descriptions such as multidimensional condition-property landscapes and key parameters for controlling the synthesis. Such information can guide us when optimizing the NP properties. Our approach is suitable to extract condition-property relations rapidly from the combinatorial synthesis data and is expected to be effective for various types of target materials, even with unknown properties, because of the flexibility of the NN analysis.
!!!All Science Journal Classification (ASJC) codes