基于大模型的柴油机装配冷试数据分析

赵旭辉¹，徐卓¹，王相成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.山东浪潮傲林大数据科技有限公司，山东 济南  250101

摘要：为提高柴油机装配质量和冷试性能，基于柴油机装配冷试基础数据集，选取加州大学欧文分校（University of California Irvine，UCI）机器学习资料库标准数据集中的Seeds、Wine、Wdbc三种数据集，对比支持向量机（support vector machines,SVM）模型、组合智能算法改进后SVM模型、Transformer模型应用于冷试异常数据的分析效果。结果表明：SVM、改进后SVM，Transformer模型对正常数据和异常数据分类的准确率分别为85.20%、92.54%、97.94%；相比SVM、改进SVM模型，Transformer模型的分类准确率有较大的提高，可用于分析参数异常；排气压力与转矩关系密切，排气压力较大造成转矩增大；排气门开启时间过长导致进气真空度异常，验证了Transformer模型对发动机装配异常识别方法的有效性。

关键词：柴油机装配；冷试；异常检测；SVM；Transformer模型架构

Data analysis of cold test for diesel engine assembly based on large models

ZHAO Xuhui¹, XU Zhuo¹, WANG Xiangcheng², YAN Wei^1*, LI Guoxiang¹

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

2.Shandong Inspur Aolin Big Data Technology Co., Ltd., Jinan 250101, China

Abstract: To improve the assembly quality and cold test performance of diesel engines, based on the basic dataset of diesel engine assembly and cold testing, three standard datasets, namely Seeds, Wine, and Wdbc, from the University of California Irvine(UCI) machine learning repository are selected. The analysis effects of the support vector machines (SVM) model, the SVM model improved by combined intelligent algorithm, and the Transformer model on abnormal cold test data are compared. The results show that the classification accuracies of the SVM, improved SVM, and Transformer models for normal and abnormal data are 85.20%, 92.54%, and 97.94%. Compared with the SVM and improved SVM models, the Transformer model has a significantly higher classification accuracy and can be used to analyze parameter anomalies. Exhaust pressure is closely related to torque, higher exhaust pressure leads to increased torque; an excessively long exhaust valve opening time results in abnormal intake vacuum. The effectiveness of the Transformer model in identifying engine assembly anomalies is verified.

Keywords: diesel engine assembly; cold test; anomaly detection; SVM; Transformer model architecture