China Nonferrous Metallurgy

To mitigate the frequent failures of Isasmelt furnace lance during copper smelting process and enhance the identification accuracy of few-shot fault data, this study introduces a novel approach, integrating exponential linear unit (ELU), global average pooling (GAP) convolutional neural network (EGCNN) and rime-ice optimization algorithm (RIME) optimized support vector machine (SVM). Initially, ELU is employed as the activation function for the convolutional neural network (CNN), enhancing robustness against noise and input variations, thereby expediting model convergence. Subsequently, GAP replaces the full connect (FC) layer to strengthen the correlation between process parameters and fault categories, which reduces the number of model parameters and mitigates the risk of overfitting. Ultimately, SVM is implemented as the final classifier in lieu of the traditional Softmax function. RIME is employed to optimize the penalty factor and kernel parameter of the SVM, thereby further enhancing the accuracy of the model. The results indicate that the proposed method achieves an accuracy of 97.08%, precision of 97.08%, recall of 97.10%, F1-score of 97.07% and Kappa coefficient of 0.9611 in identifying Isasmelt furnace lance faults. The proposed method exhibits superior fault identification performance.