【摘要】 声发射信号具有频率范围宽、蕴含信息丰富的优点，被广泛应用于柴油机故障诊断中，但诊断判据在多工况和变机型下适应性较差，以柴油机喷油器喷孔堵塞故障为例，分析缸盖声发射信号燃烧段特征，发现不同机型燃烧段的缸盖声发射信号包含多个对喷孔堵塞故障较为敏感的衰减振荡，研究了适应性较好的特征参数提取方法．柴油机运行工况多变，难以采集到所有工况充足的故障样本，基于实例的TrAdaBoost迁移学习算法，利用已有工况的数据作为源域辅助训练数据，结合少量目标域数据构成联合训练集，通过对各故障类别的权重迭代，提高了同机型不同工况的故障诊断算法的鲁棒性和故障识别率，并经不同型号的柴油机试验验证表明，该方法将诊断准确率从55%提高到90%以上，有较强的跨机型适用性．更多还原

【Abstract】 Acoustic emission(AE) signal is widely used in diesel engine fault diagnosis due to its advantages of wide frequency range and rich information,but it has poor adaptability in multi-working conditions and variable engine types. In this paper,AE signal of a cylinder head during the combustion section was analyzed to diagnose the nozzle clogging fault of a diesel engine injector. It has been found that the AE signal during the combustion section from different diesel engines contains multiple attenuation oscillations that are sensitive to the nozzle clogging fault. Therefore,a feature parameter extraction method was studied in order to obtain a good fault diagnosis adaptability. However,it is difficult to collect sufficient fault samples under all working conditions because the operating condition of diesel engine is highly varying. Such,the Tr Ada Boost transfer learning algorithm was proposed,in which the data of existing working conditions are used as the auxiliary training data in the source domain,combined with a small amount of target domain data to form a joint training set. Through the weight iteration of each fault category,the robustness and fault recognition rate of the fault diagnosis algorithm under different working conditions of the same diesel engine type was significantly improved. The test results from different type diesel engines show that the proposed method obviously improves the diagnostic accuracy from 55% to more than 90%,and also has strong applicability among different type engines.更多还原