Fault diagnosis of the mine ventilation system based on OCKIELM

To address the difficulties in obtaining fault samples from mine ventilation systems and the lack of real-time online diagnostic theories, a method combining One-Class Kernel Extreme Learning Machine (OCKELM) with incremental learning is proposed for online fault diagnosis of sequentially arriving sample data. Based on sample data from the normal operation of the ventilation system, the kernelized form of OCKELM is provided, and the expressions for the kernel function and kernel weight vector are derived. The kernel weight vector is updated, and the output values of the samples are estimated as new samples are absorbed according to the incremental learning method. Finally, the model’s test threshold is determined based on two threshold criteria. The proposed method is applied to the University of California Irvine (UCI) datasets and Dongshan Coal Mine. Experimental results indicate that the algorithm can quickly and accurately identify faulty branches, achieving a fault diagnosis accuracy of 94.8% with millisecond-level time consumption. Compared to support vector data description, principal component analysis, and one-class support vector machine methods, this method exhibits superior performance across F1, area under the curve, and G-mean metrics.