Multi-Objective Robust Optimization Reconstruction Algorithm for Electrical Capacitance Tomography

Electrical capacitance tomography holds significant potential for multiphase flow parameter measurements, but its application has been limited by the challenge of reconstructing high-quality images, especially under complex and uncertain conditions. We propose an innovative multi-objective robust optimization model to alleviate this limitation. This model integrates advanced optimization methods, multimodal learning, and measurement physics, structured as a nested upper-level optimization problem and lower-level optimization problem to tackle the challenges of complex image reconstruction. By integrating supervised learning methodologies with optimization principles, our framework synchronously achieves parameter tuning and performance enhancement. Utilizing the regularization theory, the multimodal learning prior image, sparsity prior, and measurement physics are incorporated into a novel lower-level optimization problem. To enhance the inference accuracy of the prior image, a new multimodal neural network leveraging multimodal data is developed. An innovative nested algorithm that mitigates computational difficulties arising from the interactions between the upper- and lower-level optimization problems is proposed to solve the proposed multi-objective robust optimization model. Qualitative and quantitative evaluation results demonstrate that the proposed method surpasses mainstream imaging algorithms, enhancing the automation level of the reconstruction process and image quality while exhibiting exceptional robustness. This study pioneers a novel imaging framework for enhancing overall reconstruction performance.