Abstract Fouling is a phenomenon where material accumulates on the exterior of convective heat exchangers (HX) and other surfaces. In boilers fired by waste-derived or biomass fuels, these surfaces are cleaned frequently to maintain adequate heat transfer between the flue gas and fluid. However, excess cleaning of HX surfaces wastes money and resources, and the common practice of soot removal at fixed time intervals is not an optimal strategy. An adaptive timing method would be beneficial; however, real-time knowledge of HX condition is hard to obtain. In this paper, we present (1) a state estimation approach for fouling monitoring in a Circulating Fluidized Bed (CFB) boiler, fusing knowledge from a physical model with process measurement data, and (2) a novel condition monitoring scheme based on modal-vibrational sensing, with potential for a directly estimating the degree of fouling on heating surfaces. The results are demonstrated on a full-scale commercial CFB. Combining physical models, machine learning, and modal analysis in mutually supporting ways provides a solid basis for future sootblowing optimization efforts and improved fouling management.