A Visual Analysis Method for Predicting Material Properties Based on Uncertainty

Abstract

The traditional way of studying fluorinated materials by adjusting parameters throughout multiple trials can no longer meet the needs of the processing and analysis of multi-source, heterogeneous, and numerous complex data. Due to the high confidentiality of fluorinated materials’ data, it is not convenient for the plant to trust the data to third party professionals for processing and analysis. Therefore, this paper introduces a visual analysis method for material performance prediction supporting model selection, MP2-method, which helps with researchers’ independent selection and comparison of different levels of prediction models for different datasets and uses visual analysis to achieve performance prediction of fluorinated materials by adjusting control parameters. In addition, according to the Latin hypercube Markov chain (LHS-MC) model of uncertainty for visual analysis proposed in this paper, the uncertainty of the control-parameter data is reduced, and their prediction accuracy is improved. Finally, the usefulness and reliability of MP2-method are demonstrated through case studies and interviews with domain experts.

Keywords

• machine learning
• autonomous model selection
• visual analytics
• performance prediction
• uncertainty analysis