Abstract: During the service process of aerospace vehicles, they are subjected to extreme environments such as alternating high and low temperatures, strong ultraviolet radiation, atomic oxygen erosion, and micrometeoroid impacts, which impose higher requirements on the temperature resistance, radiation resistance, and self-healing properties of polyurethane (PU) protective materials. Due to the strong designability of PU properties and the highly nonlinear correlation between soft and hard segment composition and macroscopic properties, traditional trial-and-error research methods are time-consuming and costly, and it is difficult to analyze the performance evolution mechanism at the molecular level. In recent years, machine learning (ML) is changing the existing paradigm of material research and development with its powerful data feature extraction and modeling capabilities. This paper systematically reviews the research progress of ML in the structure-property relationship of PU soft and hard segments, as well as its application in aerospace coatings, thermal insulation, structural and functional integration, etc. It outlines the chemical characteristics of PU soft and hard segments and the basic logic of ML prediction models, and systematically summarizes representative works on the prediction and mechanistic interpretation of PU thermal behavior, stress-strain behavior, surface properties, and other properties by ML. Additionally, it discusses the current status and gaps of machine learning in predicting the characteristic properties of polyurethane for aerospace applications. Based on this, it further introduces the practical progress of ML-assisted material design for PU-based coatings, foams, and fibers, focusing on the microstructure identification of foam materials, flame retardant performance optimization, thermal conductivity prediction, and active learning inverse design. It also discusses the potential applications in the field of lightweight thermal insulation for aerospace, and finally points out the current research deficiencies and development directions.

Key words: polyurethane, machine learning, structure-property relationship, intelligent design, performance prediction