The military equipment question-answer system (MEQAS) is designed to provide information related to military equipment and has made significant progress with the rapid advancement of artificial intelligence. The quality of MEQAS is particularly important, as errors in responses can lead to substantial losses. To evaluate the quality of MEQAS, high-quality test cases need to be generated. However, due to the specialized, closed, and difficult-to-access nature of data in the military equipment domain, the cost of relying solely on domain experts to manually construct datasets is prohibitively high and not effectively feasible. To improve the efficiency of generating test cases for MEQAS, this paper proposes MQATG, an automatic military equipment question-answer test case generation framework, which is based on large language models. MQATG encompasses the generation, evaluation, and selection of question-answer test cases. The question-answer test cases generation utilizes a large language model, guided by a generation prompt, to analyze the document and produce initial question-answer test cases. The question-answer test cases evaluation employs an evaluation prompt to engage the large language model in scoring each generated test case. The question-answer test cases selection sorts the question-answer test cases according to the scoring results, and allowing for the identification of the high-ranked cases as the final test cases. We evaluated the question-answer test cases for military equipment generated by MQATG using five metrics and calculated the comprehensive score. The experiment results demonstrate that MQATG can produce high-quality military equipment question-answer test cases, and is more effective than other method MQATG_b.