【作者】 阳小燕；

【导师】 刘义伦； 周国荣；

【作者基本信息】 中南大学 ， 机械电子工程， 2008， 博士

【摘要】 排烟风机是国民经济建设中不可缺少的关键设备,对其进行故障诊断方法的研究具有重要意义。当前排烟风机故障诊断手段主要是频谱诊断,随着智能诊断技术的发展,相继出现了针对单个领域的诊断专家系统,如转子故障诊断系统、电机故障诊断系统等,但由于排烟风机结构差异、环境恶劣等因素的影响,特别是在冶金、矿山等恶劣环境,诊断效果较差,为了提高排烟风机监测与故障诊断的准确性,本文作者综合运用信息融合理论、提升小波信号预处理方法、盲源分离故障诊断方法、BP-ART2神经网络故障诊断、多专家协同诊断理论等先进理论和算法,对多传感器信息在多层结构上进行多诊断方法的信息融合,并在理论研究的基础上,开发了排烟风机运行状态监测与故障诊断微机集中式和DSP分布式两种系统。信号预处理方法中,在提升小波信号分析的基础上,设计了改进型小波去噪阈值函数和平滑递变的自适应提升小波函数,提出了基于信号局部特征的自适应提升小波信号去噪方法,该方法在大型排烟风机故障信号去噪处理中取得了良好的应用效果。在故障源数不确定情况下的动态源数估计中设计了引入拓展四阶累积量矩阵的盲源分离动态源数估计算法,并研究了根据源数与传感器数的关系(正定、超定、欠定),选择相应分离算法的自适应盲源分离故障诊断方法,该方法在数据融合层面能有效地识别和诊断排烟风机动态故障。在特征融合层,研究了综合BP网络与ART2自适应共振网络二者优点的改进型BP-ART2神经网络故障诊断方法,在ART2结构的输入层增加非线性映射隐层,通过非线性映射降低输入特征的维数,从而提高ART2神经网络的诊断效率。在故障聚类中,提出了ART2警戒阈值的局部自适应调整算法,对每个聚类设置各自的警戒阈值,并根据聚类结果与期望值的误差来调整隐层映射权值;在聚类评判指标上,采用双重评判指标,将信号与相应聚类中心的幅值差,与警戒阈值一起作为判断聚类的评判指标,当两者同时满足时聚类成功。从而提高了排烟风机故障诊断的效率和准确性。针对排烟风机转子故障诊断、电气故障诊断以及机电耦合故障诊断等各种诊断方法,在排烟风机机械诊断与电气诊断的基础上,研究了综合时域诊断与频域诊断相融合、机械诊断与电气诊断相融合的黑板型多专家协同诊断系统,实现了机械与电气双重角度的故障诊断。并将多专家诊断黑板结构按照诊断逻辑划分为8个信息层,分别包含相应的诊断条件、诊断方法和诊断结论,并建立了相应的黑板监督机制,设计了多专家融合诊断算法。在决策融合层,模仿诊断专家综合考虑多个传感器诊断信息,设计了多传感器加权激励融合方法,实现多个传感器诊断结果相互比较与应证,根据两两传感器诊断故障之间的相关加权激励系数矩阵,分析各故障的相互激励与增强程度,计算加权融合结果,最后将所有两两传感器加权融合结果进行综合融合并归一化,得出多传感器故障融合诊断结果。并对多种诊断方:法得到的局部诊断结果,采用D-S证据理论决策融合得到全局诊断结论。在本文所研究的信号处理与故障诊断方法的基础上,结合排烟风机的力学分析与针对现场干扰信号的信号处理以及故障诊断的要求,研究开发了排烟风机运行状态监测与故障诊断微机集中式监测和DSP分布式监测两种系统,并已成功应用于生产实践的风机监测现场,实现了大型排烟风机状态实时监测与故障诊断。更多还原

【Abstract】 Smoke exhauster fan is important equipment in economic construction. It is of great significance to research intelligent fault diagnosis method. At present, the main fault diagnosis method of smoke exhauster fan is frequency spectrum diagnosis. As the development of intelligence diagnosis technology, diagnosis expert system have appeared consecutively, specifically for single fields, for instance, rotator fault diagnose system, motor fault diagnose system etc.. Since the effect factor of machinery structure difference and adverse circumstances, especially in metallurgy, mine etc., therefore, diagnosis effect is relatively poor. In order to improve accuracy of monitoring and diagnose of smoke extaction fan, the author apply the advanced theory and algorithm synthetically such as information fusion theory, signal pretreatment method of lifting wavelet, blind source separation, BP-ART2 neural networks fault diagnose, multi-experts collaborative diagnose theory, and so on, to fuse multi-sensor information by multi-diagnosis method on multi-level structure, on basis of research of theory, and develop two kind monitorings and fault diagnosis systems of smoke extaction fan: PC concentrated and DSP distributed.In signal pretreatment of information fusion theory, a kind improved wavelet threshold function and splitting changing function of lifting wavelet is designed, and an adaptive de-noising method of lifting wavelet on basis of local feature of signals is brought forward , which gain fine signal treatment effect.On the data level of information fusion theory, under uncertain situation of number of fault source in dynamic estimation of source number, a source number estimation algorithm based on exhibition fourth-order cumulants is designed, and an adaptive fault diagnosis method of blind source separation which chooses the relactive algorithm such as determine, over-determined and Under-determined, according to the relation of the source number and the sensor number is researched. This method is able to distinguish and diagnose of the dynamic fault of smoke extaction fan effectively.On the feature level of information fusion theory, an improved BP-ART2 neural network fault diagnosis method which utilizes synthetically both benefit of BP neural network and ART2 adaptive resonance theory is researched. On the iput layer of ART2 structure an nonlinear hidden layer is adden to reduce the dimension of input feature and inprove the diagnosis efficiency of ART2 neural network . In the fault cluster of ART2 an local adaptivly adjustive algrithm based on warning threshod of ART2 is designed, and respective threshod is interposed to every cluster , and the hidden layer mapping weight is adjusted according to the error between clustering result and the expected value. On the index of clustering deciding double indexes is adopt to judge the threshod, the first one is the amplitude error between the signal and corresponding clustering centre, the second one is the warning threshod . When the both indexes are satisfied clustering is successful.Secondly, On the feature level of information fusion theory, aimed at various diagnose method of smoke exhauster fan such as rotor fault diagnosis, electricity fault diagnosis, electromechanical coupling fault diagnosis, and so on, on the basis of machinery diagnosis and electricity diagnosis, a multi-experts coordination diagnosis system based on blackboard is researched which fuses synthetically time-domain diagnosis and frequency-domain diagnosis, machinery diagnosis and electricity diagnosis, and realizes double fault diagnosis of machinery and electricity diagnosis. According to the diagnosis logic structure of the blackboard multi-experts coordination diagnosis system is divided to 8 layers, each layer contains corresponding diagnosis condition, diagnosis method and diagnosis conclusion. A corresponding blackboard supervision mechanism is built and multi-experts fusion diagnosis algorithm is designed.On the decision level of information fusion theory, a multi-sensor weighted incentive fusion method is designed, which imitates diagnosis experts to consider many sensors diagnosis information synthetically, and realizes several sensor diagnosis results comparing and certificating mutually, and according to the weighted stimulates fusion modulus matrix between each two sensors, analyzes stimulating and strengthening degree, and calculates weighted fusion result, fuses and normalizes synthetically all two sensors weighted fusion results, and comes to a conclusion of multi-sensor fault fusion diagnosis, and aimed at the local diagnosis results of multi-method of diagnosis adopts D-S evidence theory to make global fusion and come to decision-making fusion conclusion.Adopting the signal treatment and fault diagnosis methods which are researched in this dissertation, according to the actual state of smoke exhauster fan, two kind monitoring and fault diagnosis systems of smoke exhauster fan are developed: PC concentrated and DSP distributed, which already have been successfully applied to the smoke exhauster fan of several enterprise, and realized condition real-time monitoring and fault diagnosis.更多还原