This study presents a scenario-based fault detection method for intelligent customer service systems in the telecommunications industry. By integrating a multimodal dialogue intent understanding system, the method combines textual and visual modalities to address the limitations of existing intelligent customer service systems in handling complex user needs and troubleshooting. The system includes a scene text recognition module, a fault target detection module, and an inference module, utilizing Optical Character Recognition (OCR) technology and the YOLO v7 model for text and image processing, respectively. An external knowledge base is also used to enhance fault detection accuracy and response speed. Experimental results show that the fault target detection accuracy is 82.43%, scene text recognition accuracy is 79.32%, and overall accuracy is 70.95%, significantly improving the efficiency and user experience of intelligent customer service.