【摘要】 针对固体推进剂生产过程中机理复杂、工艺参数众多，导致质量缺陷频发的问题，提出了一种基于知识图谱的固体推进剂质量缺陷根因变量识别的方法。首先通过历史数据挖掘对质量缺陷进行预测，并通过SHAP模型对预测结果进行具体解释。再将数据挖掘结果、专家经验与工艺机理进行结构化表示与融合，建立固体推进剂质量缺陷领域的知识图谱。最后将图谱中包含参数信息和结构关系的子图映射到贝叶斯网络中进行参数学习，从而推断出不同特征参数导致推进剂存在质量缺陷的后验概率。最终通过实际生产数据样本进行验证，结果表明：该方法能有效识别生产过程中造成固体推剂质量缺陷的根因变量，准确识别率可达85%。更多还原

【Abstract】 Considering the frequent quality defects in producing solid propellants due to the complex nature of the production process and the many process parameters, a knowledge graph-based method for identifying the root cause variables of these defects was proposed. Firstly, having the quality defects predicted by historical data mining, and the prediction results explained by SHAP model; and then, having the data mining results, expert experience and process mechanism structurally represented and fused to establish a knowledge graph in the field of solid propellant quality defects; and finally, having the subgraphs containing parameter information and structural relations in the graph spectrum mapped to the Bayesian network for parameter learning so as to infer the posterior probability of different characteristic parameters leading to the existence of mass defects in the propellant. The verified results from data samples show that this method can effectively identify the root cause variables which causing the quality defects of solid propellant in the production process, and the accurate recognition rate can reach 85%.更多还原