Crack coalescence prediction and load-bearing mechanism of defective specimen based on computer vision recognition model

The coalescence of cracks in rock, triggered by external stress or geological activities, plays a pivotal role in determining the mechanical properties and stability of rock structures. Consequently, the prediction method of crack coalescence become critical tools in ensuring the safety and stability of geotechnical engineering facilities. In this paper, based on the strain field data obtained by digital image correlation (DIC), a set of programs was developed to automatically identify the values and percentage changes of different strain intervals in the strain field of the specimen. Then, a computer vision recognition (CVR) model is established to study the process and prediction of crack coalescence in sandstone specimens with open flaws. This method overcomes the limitations, subjectivity and unpredictability of the traditional method of identifying cracks through artificial vision. The results show that the increase of the flaw width causes the displacement trend to be squeezed and deflected along the width direction, thereby changing the angle of crack initiation and exhibiting different forms of crack coalescence. The increase in the inclination angle of the flaw leads to an increase in the effective compressive area (ECA) of sandstone, and the magnitude of ECA is positively correlated with the uniaxial compressive strength (UCS) of the sandstone. Additionally, the CVR model identifies two types of numerical fluctuation signals before crack coalescence, namely the plastic characteristic signal (PCS) in the early stage of the experiment and the early-warning signal (EWS) near the period of crack coalescence, where EWS can be used as a predictive signal for crack coalescence. The research results provide a new algorithm technical support for the process analysis and prediction of crack coalescence, and provide a basis for early warning of rock mass engineering disasters.