An approach for fault diagnosis using a novel hybrid fuzzy clustering algorithm

In this paper a hybrid algorithm using fuzzy clustering techniques is presented. The algorithm is applied in a fault diagnosis scheme with online detection of novel faults and automatic learning. The proposal, initially identifies the outliers based on data density. Later, the outliers are removed and the clustering process is performed. To extract the important features and improve the clustering, the maximum-entropy-regularized weighted fuzzy c-means is used. Then, the use of kernel functions is performed for clustering the data, where there is a non-linear relationship between the variables. This allows achieving greater separability among the classes, and reducing the classification errors. Later, an step is used to optimize the parameters m (regulation factor of the fuzziness of the resulting partition) and sigma (bandwidth and it indicates the degree of smoothness of the Gaussian kernel function). The approach proposed was validated using the Tennessee Eastman (TE) process benchmark. The results obtained indicate the feasibility of the proposal.