数据融合在APMP盘磨故障诊断中的应用研究

本资料包含pdf文件1个，下载需要1积分

In fault diagnosis of alkaline peroxide mechanica pulp （APMP） disc refiner, it is difficult to distinguish actual conditions of various faults in the same symptom domain. Aiming at this situation, by adopting the diagnosis information of other symptom domains, the global information integration is carried out and researched. Different data fusion algorithms are proposed respectively for detection layer, feature layer and decision making layer. In detection layer, wavelet packet analysis is adopted to extract fault feature; in feature layer~ through neural network fusion, the probability assignment is provided for the D-S evidence theory of decision making layer. The experimental data indicate that because of the complementary advantages of these three layers, the probability assignment of the evidence theory may not fully depend on subjective experience of experts; and the diagnostic accuracy can be improved by adopting the redundant and complementary information of various failures.针对碱性过氧化氢化学机械浆（APMP）盘磨故障诊断在同一征兆域中很难区分多种故障的实际情况,利用其他征兆域的诊断信息进行了全局信息融合的研究。在数据处理的检测层、特征层、决策层上分别提出了不同的数据融合算法,即检测层采用小波包分析的融合方法提取故障特征,特征层通过神经网络的融合为决策层的D-S证据理论提供可信度分配。试验数据表明,通过三个层面的优势互补,可以使证据理论的可信度分配不再完全依赖主观专家经验;利用各种故障的冗余和互补信息,可提高诊断的准确率。