Accurate description of the condition of engineering structures is important for ensuring structural safety. Traditional analysis methods based on simplified physical mechanisms cannot accurately characterize the structural condition and neglect the value of the large amount of data generated during the construction process. This paper proposes a data-driven analysis framework that combines physical principles, dimensionality reduction techniques and ensemble learning models to trace back the deep-seated connections between data, achieving multi-factor analysis of structural defects. Using concrete structural cracks in a certain project as an example, the framework considers full life-cycle data, including material, environment, and construction processes, to construct an assessment model. The results show that by establishing a mapping relationship between construction data and structural condition, and integrating cumulative indicators from different construction stages, a reference for describing the structural safety condition can be provided to some extent, along with optimization suggestions, offering an analytical perspective for solving complex structural problems in engineering.