Underground concrete structures are exposed to a multi-ion groundwater and seepage–leakage coupling environment for a long time, and it is difficult to observe visually, which makes it difficult to accurately characterize important boundary conditions and defect states, resulting in significant time-varying and spatially differing characteristics of the concrete deterioration process. Therefore, its durability assessment and life prediction are significantly different from those of above-ground structures. Aiming at the complex prediction problem of limited service information of underground concrete, this paper summarizes and combs the evolution process of underground concrete life prediction methods, and puts forward the evolution process of five generation prediction frameworks: from a deterministic mechanism model (Gen-1) to a multi-physical field coupling model (Gen-2), a probabilistic reliability framework (Gen-3), a data-driven and physical information fusion method (Gen-4) and then to a digital twin framework for online update and system integration (Gen-5). Differently from the traditional review by model category, this paper reveals the internal logic of life prediction from single life point values to time-varying risk assessment from the perspective of the transformation of prediction targets and problem structures. Based on the comparison of typical underground service environments, it is further shown that the key constraints of prediction ability are usually derived from insufficient observability and limited parameter identifiability, as well as model structure errors introduced by deterioration mechanism switching and local defects, rather than physical model complexity. On this basis, this paper proposes the selection idea of life prediction methods for different underground scenes, emphasizing measurable characterization, hierarchical verification and hierarchical calculation as the core, and effectively connecting the mechanism model, uncertainty analysis, data update and operation and maintenance decisions. In this paper, the life prediction of underground concrete is redefined as a dynamic evaluation process embedded in the whole life management of infrastructure, which provides a theoretical framework and research direction for the construction of a reliable and deployable life prediction system of underground concrete.

Aiming at the problem of synergistic regulation of setting time and strength of brucite-based magnesium phosphate cement (BMPC), this study used ammonium chloride (AC) as a variable, and revealed the regulation mechanism of AC on the hydration behavior of BMPC through the tests of setting time, fluidity, and compressive strength, as well as the monitoring of pH-ion concentration, and the microanalysis of XRD-TG-MIP. The results showed that the optimal performance combination of BMPC (setting time of 16 min, fluidity of 120 mm, and compressive strength of 20.5/30.7/54.5 MPa at 3 h/1 d/28 d, respectively) was obtained when AC was doped at a dosage of 4%. The mechanism of retardation stems from the fact that the addition of AC inhibits the dissolution rate of ADP and retards the hydration reaction of Mg²⁺ and $\mathrm{P}{\mathrm{O}}_4^{3-}$. An appropriate amount of AC can optimize the pore structure of the BMPC matrix and improve the compressive strength of the matrix. The BMPC system based on complete replacement of magnesite by brucite not only significantly reduces carbon emission and cost, but also provides a new path for the development of low-carbon MPC.