论文部分内容阅读
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。
In order to improve the precision of fault diagnosis of engine rotating components, aiming at the problems of traditional shallow network and support vector machine (SVM) methods, such as lack of generalization ability and local optimal solution, A tremendous breakthrough has been made in the diagnosis of failure of performance of engine components due to the deep confidence network (DBN) that simulates the human brain multi-layer structure. In order to improve the performance of deep belief networks, an improved algorithm (ad_DBN) that adaptively adjust weights during unsupervised and supervised training phases is proposed. Taking the turbofan engine as an object, the two kinds of DBN algorithms, BP, RBF and SVM methods are compared and analyzed comprehensively from the aspects of diagnostic precision, calculation time and anti-noise ability. The results show that the diagnostic accuracy of DBN algorithm is significantly better than BP neural network, Radial Basis Function (RBF) neural network and Support Vector Machine (SVM). Thanks to the adaptive adjustment of weights, the average accuracy of ad_DBN diagnosis Up to 97.84%, its anti-noise ability is also significantly better than other algorithms, can improve the effectiveness and reliability of fault diagnosis.