A novelty detection approach for detecting faulty batches in a photo-Fenton process

Abstract A novelty detection approach for detecting novel faults has been developed and applied to a photo-Fenton process in a fully monitored pilot plant. The proposed approach consists of two stages: Fault Detection or binary classification stage, and Fault Diagnosis (FD) or multi-class classification stage. Batches under nominal operating conditions, where coffee samples are degraded, are defined according to an experimental design and three possible faults are found in the process. Experimental batches under such normal and abnormal conditions are used to construct the classification models and therefore, the novelty detection approach. Two faulty batches, not learned by the models, were used to validate the approach. The successful results obtained encourage performing potential models for other pollutants that allow further detecting novel contaminants in wastewater treatments.