基于关联规则的柴油机装配冷试异常数据分析

赵旭辉¹，徐卓¹，王辉²，闫伟^1*，李国祥¹

1.山东大学核科学与能源动力学院，山东 济南 250061；2.潍柴控股集团有限公司，山东 潍坊 261001

摘要：为分析某柴油机装配冷试试验中异常数据间的关系，提出一种基于关联规则的装配冷试异常数据分析方法，采用KIH-means聚类方式，针对该柴油机装配异常构建关联规则样本库；通过支持度-置信度试验，确定最佳的最小支持度和最小置信度阈值组合；采用Apriori算法进行关联规则挖掘，揭示异常数据的关联关系，得到针对该柴油机装配异常的挖掘结果。结果表明：该柴油机最小置信度为0.75，最小支持度为0.225%为最佳阈值组合，此时关联规则条数变化最平稳；KIH-means聚类方式与Apriori算法结合，通过对关联规则样本库数据挖掘，确定异常参数的关联规则，提出有效优化措施，从而提高发动机装配质量及装配性能一致性。

关键词：柴油机；冷试测试；异常检测；关联规则

Association rules based on analysis of anomalies in a diesel engine assembly cold test

ZHAO Xuhui¹, XU Zhuo¹, WANG Hui², YAN Wei^1*, LI Guoxiang¹

1. School of Nuclear Science and Energy Power Engineering, Shandong University, Jinan 250061, China；

2. Weichai Holding Group Co., Ltd., Weifang 261001, China

Abstract: To analyze the correlations among abnormal data in the cold test of a certain diesel engine assembly, an abnormal data analysis method for assembly cold test based on association rules is proposed. This method adopts the KIH-means clustering method to construct an association rule sample library for the diesel engine assembly data. Through the support-confidence test, the optimal combination of minimum support and minimum confidence thresholds is determined. The Apriori algorithm is used for association rule mining to reveal the correlation among abnormal data, and the mining results for the diesel engine assembly anomalies are obtained. The results show that the optimal threshold combination for the diesel engine is a minimum confidence of 0.75 and a minimum support of 0.225%, at which the change in the number of association rules is the most stables. The combination of KIH-means clustering method and Apriori algorithm can determine the association rules of abnormal parameters through data mining of the association rule sample library. On this basis, targeted optimization measures can be effectively proposed, thereby improving the engine assembly quality and the consistency of assembly performance.

Keywords: diesel engine; cold test; anomaly detection; association rule