中文摘要：

为提高发动机转动部件性能衰退故障诊断精度,针对传统的浅层网络和支持向量机（SVM）方法在诊断时存在泛化能力欠缺、易产生局部最优解等问题,引入近年来在模式识别领域取得巨大突破,模拟人脑多层结构的深度置信网络（DBN）进行发动机部件性能衰退故障的诊断。为改进深度置信网络性能,提出一种在无监督和有监督训练阶段都可自适应调整权值的改进算法（ad_DBN）。以涡扇发动机为对象,将两种DBN算法与BP,RBF和SVM方法从诊断精度、计算时间、抗噪能力三方面进行综合比较分析。结果表明DBN算法诊断精度明显优于反向传播（BP）神经网络,径向基（RBF）神经网络和支持向量机（SVM）方法,得益于权值的自适应调整,ad_DBN诊断的平均精度高达97.84%,其抗噪声能力也明显优于其他算法,能够提高故障诊断的有效性和可靠性。

英文摘要：

In order to improve the accuracy of engine rotating component performance degradation diagnosis and overcome the problems of low generalization ability,easily trapped in local optimal solution caused by traditional shallow network and Support Vector Machine（SVM） diagnosis method, deep belief network（DBN） is introduced to diagnose engine rotating component performance degradation defect. DBN imitates the multiple layer structure of human brain and has made great achievement in pattern recognition area in recent years. In addition,to improve the performance of deep belief network,an improved algorithm（ad_DBN） is put forward,which can upgrade weights adaptively both on unsupervised and supervised learning stages. The article compared the two DBN methods with Back Propagation（BP） network,Radical Basis Function（RBF）network and Support Vector Machine（SVM） methods in diagnosis accuracy,computation time and anti-noise ability using a certain type of turbine fan engine data. Results show that two DBN methods have obvious advantage over the other three methods in diagnosis accuracy. Own to the strategy of adaptive upgrading weigh,ad_DBN diagnosis mean accuracy is as high as 97.84%,and shows better anti-noise ability than other algorithms. It helps to diagnose the engine component performance degradation fault more effectively and reliably.