<p>Symmetric image registration estimates bidirectional spatial transformations between images while enforcing inverse consistency. Its capability of eliminating bias introduced inevitably by generic single-directional image registration allows more precise analysis in manufacturing quality control applications, e.g., PCB component placement verification and defect detection. However, most existing symmetric registration techniques for industrial inspection images are limited by low speed from commonly used iterative optimization, hardship in exploring complex component-to-pad spatial relations, or high labor cost for labeling alignment ground truth. We propose FlowReg-Diff to shatter these limits, which is a novel flow matching-based approach to symmetric image registration. We formulate symmetric registration of PCB component images as a conditional diffusion process and train it with a semi-supervised strategy. The registration symmetry is realized by introducing a loss encouraging that the cycle composed of the geometric transformation from one component view to another and its reverse should bring the image back. The flow matching framework enables efficient single-step inference while preserving the high-quality generation capability of diffusion models. The semi-supervised learning enables both precious labeled alignment data and large amounts of unlabeled PCB inspection images to be fully exploited. Experimental results from multiple PCB component insertion datasets demonstrate the superiority of FlowReg-Diff to several existing state-of-the-art methods in terms of registration accuracy, inverse consistency, and inference speed for industrial inspection applications.</p>