Rapid estimation of coal ash using integrated X-ray transmission and fluorescence features

To enhance the accuracy and stability of coal ash content detection and overcome the limitations of X-ray transmission (XRT) under conditions of severe coal quality fluctuations, this study proposes a multi-source modeling approach that integrates XRT and X-ray fluorescence (XRF) features. Three detection techniques were compared for estimating the ash content of artificially blended coal samples: the XRT based on high-energy attenuation rate, low-to-high energy ratio, and attenuation ratio, the XRF based on the contents of Si, Fe, K and the effective atomic number, and the multisource model fused XRT and XRF characteristics. Moreover, three machine learning models, multiple linear regression, ridge regression, and support vector regression, were established, and the estimation performance of ash content under different modeling methods was systematically evaluated. The results show that the XRT-XRF feature fusion strategy significantly outperforms the single-source approaches based on either XRT or XRF. Among all models, the ridge regression model based on fused features achieved the best performance, with R2 of 0.8807, RMSE of only 0.6816 and a proportion of samples with absolute estimation errors within ±1% reaching 85.71%. The findings provide theoretical support for the rapid estimation of coal ash, but independent validation using real coal is still necessary.