论文部分内容阅读
针对传统神经网络在故障诊断中因测点信息多而导致的网络庞大、收敛困难等问题,引入集成神经网络,提高了融合诊断效率;同时引入基于D-S证据理论,这种决策融合方法解决了集成神经网络各个子网诊断结果不一致的问题。在应用于柴油机故障诊断时,首先对测取的正常和故障样本进行小波包AR谱分析,同时提取各个特征频带的能量分别作为集成神经网络对应子网的输入进行诊断,当其无法确定诊断结果时,再运用证据理论进行决策融合输出最终诊断结果。试验证明:基于集成神经网络和D-S证据理论的两级综合诊断模型提高了诊断的准确性和可靠性。
Aiming at the problems of large network and difficult convergence caused by more information of measuring points in traditional fault diagnosis, the introduction of integrated neural network improves the efficiency of fusion diagnosis. At the same time, based on DS evidence theory, this decision fusion method solves the problem of integration Neural network sub-network diagnosis of inconsistent problems. When applied to diesel engine fault diagnosis, firstly, the wavelet packet AR spectrum analysis is performed on the normal and faulty samples and the energy of each characteristic band is extracted as the input of the corresponding subnet of integrated neural network respectively. When it can not confirm the diagnosis result When the evidence theory and then use the decision-making fusion output final diagnosis. Experiments show that the two-level synthetic diagnosis model based on integrated neural network and D-S evidence theory improves the accuracy and reliability of diagnosis.