Self-Healing UI Test Automation via Multi-Modal Fusion

Traditional UI automated testing relies heavily on application implementation details (such as XPath and element ID), resulting in high script maintenance costs and poor robustness. This paper proposes a multimodal self-healing framework, which combines the semantic reasoning ability of Large Language Model (LLM), computer vision positioning technology to achieve dynamic repair of test scripts. Experiments show that the success rate of this method in GUI change scenarios is 89%, and the maintenance cost is reduced by 67%. The industrial practicability of this method is verified through the business case of China Mobile.