Edge-Cloud Collaborative Calibration of Industrial Digital Twin Models

Abstract

This article proposes an edge-cloud collaborative model calibration framework for industrial digital twins that decomposes the traditional Kennedy–O'Hagan (KOH) Bayesian calibration into cloud-side surrogate training and edge-side bias compensation, enabling online calibration under resource constraints. Our key innovation lies in three aspects: 1) a formal task decomposition of the KOH framework that enables asynchronous cloud-edge calibration while preserving theoretical consistency; 2) a positive information gain active learning (PIG-AL) algorithm that selectively filters noise-contaminated samples for robust edge-side model training; and 3) a bidirectional synchronization protocol that maintains calibration accuracy across different update frequencies. Results from simulations and an industrial fused magnesia furnace case study demonstrate the effectiveness and practical feasibility of the proposed approach.