In stepped eddy current thermography nondestructive testing, the structure of the electromagnetic exciter has an important influence on the distribution and strength of the electromagnetic field and heat transfer field. In this article, an integrated optimization system is designed and applied to the square electromagnetic exciter structure to improve the comprehensive detection performance of volumetric defects in oil and gas storage tanks or large-diameter pipe sections. First, the influence of the variation of each structural parameter of the exciter on the detection performance is analyzed, and the structural parameters with significant cumulative influence are selected as optimization variables by principal component analysis. The objective assignment method is used to give weights to nine detection performance indicators according to the amount of information, and three comprehensive indicators about temperature rise, uniformity and detection efficiency are obtained. Then the discrete optimization space constructed from the key structural parameters about the composite indexes is mapped to the continuous optimization space of the machine learning model. A multi-objective particle swarm algorithm is utilized to obtain the Pareto front. Finally, the optimal structures are ranked according to the approximate ideal solution ranking method. Compared with the initial structure, the optimized structure improves the temperature rise, uniformity, and efficiency indexes by 18.88%, 2.46%, and 73.61%, respectively. compared with the initial structure. In addition, a set of experimental systems is established to verify that the optimized structure has significantly better detection performance than the comparative exciter and can be used to detect defects in steel plates with a thickness of 8 mm and a diameter-to-thickness ratio greater than 3. By integrating and optimizing the electromagnetic exciter structure of the system design, it significantly improves the stepped eddy current thermography detection performance, and experimentally verifies its efficient detection capability of internal volume defects in tank materials, providing an effective solution for the detection of volumetric defects in oil and gas storage tanks and large-caliber pipeline segments.