Concrete production accounts for approximately 8% of global anthropogenic CO₂ emissions, yet the same material possesses a latent capacity for post‑hardening carbonation that can partially reabsorb emitted CO₂ over decadal timescales; critically, however, no existing system enables real‑time, co‑located tracking of both carbon sequestration progress and structural cracking. Here we demonstrate an embedded microsensor network within concrete that simultaneously monitors CO₂ absorption, detects crack initiation and propagation, and provides data for durability forecasting. A suite of MEMS nondispersive infrared (NDIR) CO₂ sensors, miniature piezoelectric acoustic emission transducers, and complementary metal‑oxide‑semiconductor (CMOS) relative humidity and temperature sensors were embedded during casting into cement‑fly ash‑biochar ternary mixes (water‑to‑binder ratio = 0.45). Specimens were subjected to 90 days of accelerated carbonation (5% CO₂, 65% relative humidity, 23°C) combined with cyclic compressive loading (0.3–0.7 fc′) to induce controlled microcracking. The sensor data stream—comprising CO₂ concentration, acoustic emission counts and amplitudes, and hygrothermal conditions—was integrated into a physics‑informed neural network (PINN) that couples Fickian diffusion with crack‑enhanced transport kinetics. Post‑mortem characterization via thermogravimetric analysis, phenolphthalein spraying, and X‑ray computed tomography validated sensor performance. The embedded sensors survived the 90‑day exposure with an 85% functional yield, with failures attributed to lead‑wire fatigue rather than sensor body degradation. Cumulative CO₂ uptake measured by embedded sensors showed strong agreement with conventional titration‑based methods (root‑mean‑square error = 1.2 kg CO₂/m³ concrete). Acoustic emission signals detected cracks as narrow as 50 µm with 94% classification accuracy (precision = 0.92, recall = 0.91) against post‑test dye penetrant verification. The PINN, trained on the first 60 days of sensor data, forecast carbonation depth at day 90 with R² = 0.91 and mean absolute error of 3.1 mm—outperforming a classical Fickian model (R² = 0.78). This smart sensor‑embedded concrete establishes, for the first time, a real‑time, non‑destructive framework for concurrent carbon accounting and structural prognosis, enabling a paradigm shift toward low‑carbon, resilient infrastructure that self‑reports both its environmental and mechanical status.
Keywords: Accelerated carbonation, Acoustic emission (AE), Carbon capture, Embedded MEMS sensor, Physics-informed neural network (PINN), Self-sensing concrete, and storage (CCUS), utilization
