Development of novel methods to predict the strength properties of thermally treated sandstone using statistical and soft-computing approach

The knowledge of thermally modified strength properties is crucial in processes such as underground coal gasification, disposal of radioactive wastes in deep geological reservoirs and restoration of fire-damaged structures. However, due to the unavailability of dedicated laboratory equipment, scientists and engineers need to go across various laboratory facilities for testing their desired sample. Further, it delays the process and increases the cost of the project. Since there is a lack of empirical equations that can provide the strength of a thermally modified rock from the strength at room temperature, new predictive models have been developed using multivariate regression analysis (MVRA), artificial neural network and adaptive neuro-fuzzy inference system (ANFIS), to predict the uniaxial compressive strength (UCS) and tensile strength (TS) of both uncooled (NC) and heat-treated (WC) samples of a fine-grained Indian sandstone that has been widely used as a major building material in most of the Indian monuments. The UCS and TS of NC and WC samples have been predicted from the physical properties, viz. temperature (T), density (D), porosity (P), thermal expansion coefficients (EL and EV) and ultrasonic wave velocities (VP and VS). The performance and prediction efficiency of the models have been compared on the basis of performance indices, namely the correlation coefficient (R2), the mean absolute percentage error, the root mean square error and the variance account for. Based on the comparative study, it was concluded that ANFIS provides a relatively closer estimation as compared to MVRA.