Novel Adaptive Multiscale Vertical Federated KELM Algorithm for Cross-unit Fault Diagnosis in Interconnected Industrial System

A key challenge that needs to be addressed in data-driven fault diagnosis technology based on artificial intelligence algorithms is how to effectively perform cross-unit fault diagnosis in interconnected industrial system while preserving the data privacy of each subunit. Fortunately, it is found from our research that the kernel extreme learning machine (KELM) algorithm with the vertical federated (VF) learning mechanism shows great effect for solving this challenge. However, it assigns single and equal contribution scale to the multiple feature variables of each subunit participated in VF learning, which leads its fault diagnosis rate (FDR) is still insufficient. To address this shortcoming, a novel adaptive multiscale vertical federated kernel extreme learning machine (AMVF-KELM) algorithm is proposed in this paper. For AMVF-KELM, the multiple feature variables of each subunit were assigned different contribution scales according to Laplacian Score, that is, multiscale. In addition, the adaptive adjustment factor is designed in AMVF-KELM for the multiscale, which enables the algorithm to adapt more flexibly to the complexity of industrial process data, that is, adaptive. The effectiveness and superiority of our method is validated on the benchmark Tennessee-Eastman industrial system datasets. The results demonstrated that our method can achieve better FDR while preserving the data privacy of each subunit effectively.