Abstract: Coal flotation is a critical stage in modern coal preparation, and its intelligentization plays a key role in the overall transformation of coal-washing plants. The ash content of the flotation concentrate directly reflects product quality and impurity removal efficiency. Precise control of concentrate ash not only guides reagent dosing and process optimization, but also supports maximal coal recovery and economic returns. To address the challenges of delayed ash-content measurements and the underutilization of heterogeneous data, we propose an intelligent detection method for flotation coal ash based on multi-source information fusion. The method integrates X-ray fluorescence (XRF) spectral data, real-time flotation process variables, and tailings imagery features. We use continuous projection and multiple linear regression for spectral dimensionality reduction and apply the Hilbert-Schmidt independence criterion (HSIC) for time-series alignment to handle asynchronous data streams. A regularized Random Vector Functional-Link network with dynamic node adjustment is then developed to model nonlinear relationships between multi-source data and ash content. Experimental results show that the root mean square error is 0.113, the coefficient of determination is 0.787, and the pass rate is 100% when the absolute error of ash content is 0.3. The developed intelligent ash-content detection system has been successfully deployed on-site, achieving real-time, high-precision measurements that meet the production process requirements. This system provides comprehensive technical support and decision-making data for the flotation process, enhancing detection accuracy and process control.