Open-Source Concrete Stress-Strain Acquisition from Comoression Machines Using Comouter Vision and Microcontroller-Based Position Sensors

Obtaining concrete stress–strain curves requires synchronized load and deformation measurements, but many laboratories use legacy machines with only visual load displays and no digital output. This study presents an open-source, non-invasive, event-based architecture for automated acquisition during compression tests. A smartphone recorded the machine display, while a Python/OpenCV pipeline using YOLOv8 detected digits and reconstructed load at nominal 1 s intervals. Two linear potentiometric position sensors connected to an ESP32 measured axial deformation and transmitted timestamped data to a cloud database at nominal 1 Hz. A web application then generated stress–strain curves and PDF reports. Calibration against a mechanical dial indicator using 44 paired observations yielded a mean absolute error of 0.01562 mm, root mean square error of 0.01930 mm, and R² = 0.999991. The digit detector achieved 0.920 precision, 0.932 recall, 0.964 mAP50, and 0.749 mAP50–95. Two independent groups of five companion cylinders from the same batch reached mean peak compressive stresses of 26.85 ± 4.03 MPa and 27.86 ± 1.64 MPa, with corresponding peak axial strains of 0.002050 ± 0.000876 and 0.002976 ± 0.000462. The workflow produced mechanically plausible responses without modifying the machine electronics, using automated offline video processing for load reconstruction, and fully automated digital reporting.