Real‐time fault detection in multicomponent nuclear‐waste slurries through data fusion of spectroscopic sensors

Abstract Three instruments–Raman spectroscopy, attenuated total reflectance–Fourier transform infrared spectroscopy, and focused beam reflectance measurement–were used to detect sensor faults, mixing faults, and unanticipated chemistry in a system of multicomponent slurries. Data from the three instruments were combined via a data fusion scheme utilizing principal component analysis, Hotelling T, and squared prediction error. Uncertainty was quantified yielding a three‐sigma region of normal operation to identify faults. The three instruments allowed for the detection of a variety of process faults that may appear in either phase of a slurry. This work represents a major step forward in (1) monitoring radioactive slurry processes by reducing the need for offline monitoring, (2) accurately detecting faults in the presence of probe fouling, and (3) enabling the use of commercially available spectroscopic sensors to detect a variety of faulty process states in real‐time and remotely.