基于改进证据理论的多传感器信息融合故障诊断

针对传感器信号不确定会产生冲突证据的问题,提出了一种基于改进证据理论的多传感器信息融合故障诊断方法。提出了基于遗传神经网络的原始证据生成方法，利用遗传算法优化神经网络参数，提高网络训练速度;定义了向量空间和方向相似度，利用分类准则函数区分冲突证据和相似证据，通过可信度修正冲突证据，降低了因不确定性产生的冲突对合成结果的影响。通过齿轮泵故障实验验证了改进方法的有效性,改进方法的诊断正确率明显高于单一传感器的诊断正确率，并通过设置适当的阈值提高了方法的灵活性和适用性。

Multi-sensor Information Fusion Fault Diagnosis Based on Improved Evidence Theory

Aiming at conflict evidence resulting from uncertainty of sensor signals, a new multi-sensor information fusion fault diagnosis approach was proposed based on improved evidence theory. Firstly, a method to create original evidence was put forward using genetic neural network, where genetic algorithm was used to optimize neural network parameters so as to enhance the training speed. Secondly, vector space and direction similarity were defined and classification rule function was built to distinguish conflict evidence and similar evidence. Credibility modified conflict evidence to decrease the conflict effect from uncertainty. Finally, gear pump fault tests prove the validity of improved method, whose diagnosis precision is higher than that of single sensor diagnosis evidently. The threshold setup increases the flexibility and applicability.