摘要
研究了信息融合技术在电控发动机故障诊断中的应用。研究结果表明,基于神经网络的特征层信息融合诊断效果明显优于单一传感器,而且可实现信息压缩,进行实时处理与诊断;基于Dempster—Shafer证据推理的决策层信息融合,可对异质传感器信息进行非同步处理,对发动机故障分类准确性高、可靠性强,但融合精度不及特征层融合方法,预处理代价高。在实际应用中,应根据传感器类型、信号预处理方式、系统的复杂程度等合理选择信息融合方法。
This paper described the application of data fusion technology to fault diagnosis for automotive electronically controlled engine. An example of the technique was the use of fault detection and diagnosis for SANTANA AJR engine.The conclusion is reached as follows:(1)feature level fusion of multisensor data based on neural network provides significant advantages over single source data,implements information compression so as to perform real time data processing and fault diagnosis;(2)decision level fusion based on Dempster -Shafer evidence theory processes asynchronously data acquired through heterogeneous sensors, achieves good accuracy, high reliability and more specific inferences than could be achieved by use of a single sensor alone. But decision level fusion has lower accuracy and higher data preprocessing cost than feature level fusion. In practical application of engine fault diagnosis, the methods of data fusion must be reasonably chosen by means of sensor types, data preprocessing manner and complexity degree of the diagnosis system designed.
出处
《内燃机学报》
EI
CAS
CSCD
北大核心
2003年第5期374-378,共5页
Transactions of Csice
基金
航天"863"资助项目(2002AA722073)
山东省自然科学基金资助项目(Y2002F17)。