Extended Shapelet Learning-Based Discriminant Dictionary for Froth Flotation Fault Recognition

Accurate fault type recognition is crucial for the stable and optimal operation of froth flotation production. However, the complexity of the industrial flotation process presents a challenge due to the numerous fault states and the similar froth image features associated with these faults, which hampers accurate identification. To address this issue, we introduce a novel method that combines extended shapelet operators and dictionary learning to enhance flotation fault recognition performance. First, we propose an extended shapelet learning (ESL) method using process parameters and froth videos to develop spatio-temporal feature expression. Then, we construct a discriminative dictionary (DD) model to encode these features, boosting both intraclass compactness and interclass separation. Finally, an incremental learning-based real-time update strategy is introduced to improve the generalization performance of the model to unknown faults. The proposed method was validated on the froth flotation dataset and demonstrated excellent accuracy and robustness, with at least a 2.49% improvement in accuracy compared to conventional methods.