Mechanical properties evaluation of waste gangue-based cemented backfill materials based on an improved response surface model.

Traditional mining methods damage the cultivated land and produce gangue waste that often contaminates the environment. Yet, these problems can be mitigated by transforming the waste into gangue-based cemented backfill material (GCBM), whose mechanical properties are crucial for surface protection. Therefore, in this study, an intelligent model based on laboratory tests was developed to evaluate the GCBM's mechanical properties. The strength tests and polynomial response surface model (PRSM) were used to analyze the non-linear correlation between the influencing factors and the uniaxial compressive strength (UCS). Meanwhile, the importance of multidimensional factors was analyzed by the mean impact value, revealing that concentration and gangue proportion are the most sensitive factors. In addition, an intelligent response surface model (IRSM) based on support vector regression model was constructed by enhancing an optimization algorithm with chaotic mapping and adaptive methods. The performance of the traditional PRSM and the novel IRSM was compared, and the IRSM was validated. The IRSM can predict UCS more efficiently and effectively than the traditional PRSM under high-dimensional factors, with R2 of 0.96 and MBE of 0.05. This indicated that the IRSM has the potential to promote coal mine waste reduction and environmental protection.

Evaluation of rheological and mechanical performance of gangue-based cemented backfill material: a novel hybrid machine learning approach.

Waste discharge and surface damage are the unavoidable consequences of coal mining. However, filling waste into goaf can help reuse waste materials and protect the surface environment. In this paper, it is proposed to fill coal mine goaf with gangue-based cemented backfill material (GCBM), while the rheological and mechanical performances of GCBM influence the filling effect. A method that combines laboratory experiments and machine learning is proposed to predict the GCBM performance. The correlation and significance of eleven factors that affect GCBM are analyzed using random forest method, and the nonlinear effects of the main factors on the slump and uniaxial compressive strength (UCS) are analyzed. The optimization algorithm is improved, and the improved algorithm is combined with a support vector machine to build a hybrid model. The hybrid model is systematically verified and analyzed using predictions and convergence performances. The results demonstrate that the R2 of the predicted and measured values is 0.93 and the root mean square error is 0.1912, indicating that the improved hybrid model can effectively predict the slump and UCS and can promote sustainable waste use.