AI Agent-driven system development for promoting experimental self-learning and fault diagnosis skills

In higher education, laboratory-based courses play a critical role in bridging theoretical knowledge and practical competence. However, students often encounter substantial challenges in independently mastering complex experimental procedures and resolving unexpected errors during laboratory sessions. Traditional instructional models, which rely heavily on instructor guidance and static teaching materials, are insufficient to address the individualized and real-time learning needs of students. With the rapid development of artificial intelligence and conversational agents, there is an increasing demand for intelligent systems capable of providing dynamic guidance, supporting autonomous learning, and improving diagnostic efficiency when students encounter experimental difficulties. This study addresses this need by developing and evaluating an artificial intelligence (AI) Agent–based intelligent tutoring system specifically designed for experimental teaching scenarios. The system supports a closed-loop learning process integrating self-learning, hands-on operation, and fault diagnosis. The primary objective is to improve laboratory task completion efficiency, enhance the quality of students’ self-directed learning paths, and increase the success rate of fault diagnosis while fostering greater engagement and satisfaction.

The system was designed using a modular and service-oriented architecture consisting of four main components: a front-end interaction layer, an AI Agent module, a knowledge base system, and a log analysis module. The front-end interaction layer provides students with an intuitive and responsive interface that integrates multimodal content delivery, semantic highlighting, and a conversational window for natural language interaction, ensuring accessibility across devices. The AI Agent module functions as the intelligent core and incorporates natural language understanding, intent recognition, context modeling, and response generation. By integrating a large language model with customized prompt strategies, the Agent delivers adaptive feedback and targeted recommendations. A hybrid knowledge base was constructed by combining rule-based structures for rapid keyword matching with vector-based semantic retrieval to address complex or ambiguous queries. The knowledge base organizes experimental procedures, common error cases, and semantic links between conceptual knowledge and operational steps, enabling fine-grained alignment between theory and practice. To support personalized recommendations and adaptive interventions, a log analysis module continuously records and analyzes student interactions, including behavioral trajectories, error frequencies, and system responses. Empirical validation was conducted in an experimental class of the Computer Organization course at a university. An experimental group used the AI-supported system, whereas a control group followed conventional instructional practices. Data collection included task completion time, error resolution rate, quality of recommended self-learning paths, and post-course satisfaction surveys.

The experimental evaluation demonstrated that the system produced notable improvements across multiple dimensions. Compared with the control group, students in the experimental group completed laboratory tasks with approximately 20% greater efficiency, reflecting the benefits of streamlined guidance and real-time support. The quality and adaptability of self-directed learning paths improved markedly, as the AI Agent generated context-aware recommendations that reduced redundant exploration and directed students toward more effective solutions. Fault diagnosis performance also improved substantially, with the success rate of problem identification and resolution exceeding 85%, significantly higher than that of the control group. In addition, survey results indicated high levels of student satisfaction with the system. Students particularly valued its ability to provide timely assistance, explain complex concepts in accessible terms, and promote greater autonomy during laboratory work. Qualitative feedback further suggested that the system encouraged independent learning by reducing reliance on instructors for immediate troubleshooting and supporting active problem-solving.

The findings demonstrate that the AI Agent–based intelligent tutoring system effectively enhances laboratory teaching by addressing both cognitive and operational challenges encountered by students. By integrating semantic modeling of experimental tasks, multi-source fault knowledge bases, and dialog-driven intent recognition, the system provides a comprehensive solution supporting the full cycle of self-learning, practical experimentation, and diagnostic reasoning. Its modular architecture enables adaptation to different subject domains, while the hybrid knowledge base and real-time log analysis provide a foundation for continuous improvement and scalability. The observed improvements in task efficiency, learning path optimization, fault resolution, and student satisfaction highlight the system’s potential to transform experimental pedagogy in higher education. Beyond its immediate educational benefits, this study proposes a replicable framework for applying AI Agents in educational environments, offering guidance for future research and practice in human–AI collaborative learning. Overall, the results underscore the transformative potential of artificial intelligence in promoting autonomous learning, reducing instructional bottlenecks, and advancing the modernization of laboratory teaching.