AI‑Driven Automated Question Generation for Aviation Maintenance Training

The maintenance department of JD Aviation faces a rapid increase in personnel and a growing number of training manuals (about 12 handbooks) and regulatory documents, resulting in a massive and constantly changing pool of exam questions that are difficult to manage manually.

To improve training effectiveness, the organization plans to shift from a single‑attempt, 70‑point pass system to an unlimited‑attempt, 100‑point pass system, requiring a large, high‑quality question bank that stays synchronized with document updates.

The proposed solution combines a vector store with a large language model (LLM). Documents are vectorized and stored in the AutoBots platform, enabling fast retrieval of relevant manual content. The LLM then generates single‑choice, multiple‑choice, true/false, and short‑answer questions based on retrieved knowledge and user‑specified parameters (question type, difficulty, quantity, knowledge scope).

Key steps include:

Tool selection: using the AutoBots platform for configurable LLM, knowledge base, and plugin integration.

Overall workflow design: users upload files, AutoBots parses and stores them, users issue commands, the system retrieves relevant knowledge, the LLM generates questions, and results are previewed and can be added to the question bank.

LLM prompt design: a detailed role‑play prompt defines the question‑generation skills, output formats, and constraints (e.g., four options for single‑choice, JSON array output).

The article also provides an example of generated JSON questions wrapped in a code block:

[
    {
        "type": "singleChoice",
        "question": "根据CCAR-396-R3规定，民航局对在民用航空安全信息管理工作中做出突出贡献的单位和个人给予什么？",
        "options": [
            {"optionName": "A", "optionContent": "罚款"},
            {"optionName": "B", "optionContent": "表彰和奖励"},
            {"optionName": "C", "optionContent": "警告"},
            {"optionName": "D", "optionContent": "培训"}
        ],
        "correctAnswer": "B"
    },
    ...
]

Additional features include batch question generation, support for whole‑document processing, duplicate‑rate filtering, and periodic validity checks of existing questions against updated manuals.

Overall, the AI‑driven approach reduces manual effort, improves question relevance, and can be extended to other domains requiring automated test or questionnaire generation.