Identifying Signals of Failure Freeze–Thaw Instability Fissured Sandstone Based on CPO-RF and Acoustic Emission

The brittle nature of rock materials results in insignificant failure precursors, often leading to sudden and catastrophic events, particularly in cold regions. To address this issue, this study conducted freeze–thaw cycle tests and uniaxial compression tests on fissured sandstone specimens. The compression process was monitored and analyzed using acoustic emission (AE) technology. Subsequently, the collected AE signals were combined with machine learning techniques to construct a multifeature and multicondition data set. A novel crested porcupine optimizer-random forest (CPO-RF) machine learning model was proposed to identify failure signals in sandstone. The results demonstrate that failure cracks propagate from both ends of the prefabricated fissure until the final failure of the specimen. Under freeze–thaw cycles, the generation of additional tensile cracks leads to a transition in the failure mode from shear-dominated to tension-dominated cracking. Sudden changes in the AE ringing count, b-value, the rise time to amplitude, and event rate are reliable indicators of impending rock failure. The AE failure signals for frozen sandstone occur approximately 11, 17, and 24 s earlier than those for unfrozen sandstone. Additionally, the failure signals advance by 24, 19, 11, 14, and 16 s as the prefabricated fissure angle increases, respectively, compared to intact sandstone. The proposed CPO-RF method effectively captures failure signals with an accuracy of 0.91, an area under the curve value of 0.92, and a kappa value of 0.78 compared with RF, support vector machine, and back propagation models. The CPO algorithm is compared with the introduced bitterling fish optimization, puma optimizer, genetic algorithm optimization algorithms, and it is found that R2 = 0.85 and mean square error = 0.059 in the training process, demonstrating better accuracy and stability. This indicates that the proposed model has good stability and generalization ability, providing valuable insights for the early warning system of rock instability in cold regions.