【作者】 田玉玲；

【导师】 熊诗波；

【作者基本信息】 太原理工大学 ， 机械电子工程， 2009， 博士

【摘要】 人类长期生存在充满传染性病原体的环境中,可是大部分情况下,能够抵御这些感染,那是因为我们的免疫系统保护机体抗御病原体的侵害。生物免疫系统是具有异常检测、多样性、交互性、连续学习、记忆等特性的复杂多层防御系统。受生物免疫系统启发,论文以机电设备故障诊断为目标,研究基于人工免疫系统的故障诊断系统设计及应用。论文分析了生物免疫系统的多层防御结构,特别是固有免疫系统和适应性免疫系统及其相互关系,固有免疫系统的重要性表明人工免疫系统也应该将固有免疫系统和适应性免疫系统进行整合,建立具有多种检测机制的诊断模型,以提高系统的故障识别率。借鉴生物免疫系统的分层防御机理以及层次间的相互作用,本文提出了用于机电设备故障诊断的多层免疫诊断模型(MIFD),将固有免疫系统扩增到人工免疫系统中,增强了人工免疫系统的性能。多层免疫诊断模型采用三层结构,第一层是固有诊断层,主要实现对故障在发生概率上相互独立的已知故障类型的快速及时的故障检测与诊断。第二层是故障传播诊断层,解决故障单元的影响通过整个系统进行传播的问题。第三层是适应性诊断层,以自适应的方式解决未知及早期故障的识别与诊断。层次之间通过提呈抗原以及激活信号进行信息传递与交互。该模型借鉴了生物免疫系统的多种重要机制,不仅考虑了系统对抗原的多层防御结构及其作用时相,还包括了自体/非自体模式识别。模型采用了将固有应答与适应性应答的相互激活机制,克隆选择算法与免疫网络模型的相互激励机制相结合的原理,同时还使用了初次免疫应答理论处理系统中出现的未知故障的诊断。将否定选择算法、克隆选择算法及免疫网络模型等分散的人工免疫算法及模型组合为一个较完整的人工免疫系统结构,共同完成故障诊断任务。在故障传播诊断层,利用B细胞免疫网络理论建立故障传播模型,将设备系统中单元之间的故障传播的因果关系映射为免疫系统中细胞之间的交互识别关系。由B细胞网络描述故障节点传播关系,T细胞描述测点回路,实现基于免疫网络的故障传播模型的设计,将免疫网络的动力学特性应用于故障诊断中,并且采用粗糙和精确分步诊断方法,实现准确的故障定位。在适应性诊断层,提出了B-PCLONE连续学习算法,采用B细胞和抗体双重学习机制概括在抗原数据中发现的模式。通过对诊断知识的不断补充和完善,克服了故障知识的不完备问题,使系统的诊断能力达到最优。在抗体学习中,采用粒子群优化算法来指导抗体的变异方向,使抗体能够向着有益的方向发展,提高了最佳亲和力的收敛速度。同时系统通过记忆初次响应的诊断结果,可以更快更准确地实现类似故障的二次响应。在本文设计的人工免疫系统中,提出了区分B细胞与抗体功能的思想。基于生物体液免疫机理,将故障检测器定义为B细胞及其所包含的若干抗体结构。将故障类型映射为B细胞,将各种故障征兆映射为抗体种群,用B细胞内包含若干抗体更准确地逼近故障征兆与故障的对应关系,在空间区域的分布上属于同一故障的各种故障征兆发生在这种故障的范围内。采用这种检测器机制不仅解决了故障征兆的混叠使得各故障难以明确区分问题,而且可以提高故障检测的效率与准确率,提高连续学习的精度。最后以异步电动机为例,设计了试验方案。模拟了各种故障,对所提出的理论方法及所完成的相关成果进行了验证。更多还原

【Abstract】 We are constantly being exposed to infectious agents and yet, in most cases, we are able to resist these infections. It is our immune system that enables us to resist infections. The natural immune system is a complex multi-layer defense system, with its cell diversity, anomaly detection, interactive, Life-long learning and memory of information processing characteristics. Inspired by theoretical immunology, in this study, we explore the design and application of fault diagnosis system based on artificial immune systems (AISs) for electromechanical device.This thesis examines multi-layered defense structure of Biology immune system, especially on innate immunity system, adaptive immunity system and interaction between them. The importance of the innate immune system suggests that AISs should also incorporate models of innate and adaptive immune system to structure diagnosis model with multifold detection mechanisms and improve the performance of the approach.Inspired by the multi-layer defense mechanism and incorporates the feedback mechanism in the nature immune system, the paper proposes a multi-layer immune model for fault diagnosis (MIFD) which incorporate both innate and adaptive immune system mechanisms on the purpose of enhancing the performance of the artificial immune system. In the multi-layer model, the innate immune layer directs recognition of known fault that could not cause influence to other nodes, the propagation immune layer adopts the structure of the B-lymphocyte network to construct the fault propagation network for the fault localization, finally, the Adaptive immune layer learns the unknown and incipient fault pattern. Layers interact with each other through activation signals and presenting antigen.The proposed MIFD draws its inspiration from variety of cells and different type of mechanisms in the natural immune system. It not only adopts multi-layered defense structure and works in tandem with each other, but also involves the self/non-self discrimination. The model considers the activation between clonal selection algorithm and immune network, utilizing the theory of primary response and B-lymphocyte secrete antibodies to deal with the unknown fault pattern. In order to implement fault diagnosis, we combine negative selection algorithm, clonal selection algorithm and immune network to build the framework for engineering artificial immune systems.The fault propagation layer adopts the structure of the B-lymphocyte network to construct the fault propagation model for the fault localization. In this structure, the cause-consequent relationship of fault propagation of systems corresponds to the interaction between B-lymphocytes in the immune system. In this model, with the network of B-lymphocytes representing the influence of fault node throughout system propagates and T-lymphocytes representing sensors loop. It realizes time-continuous fault diagnosis, incorporate the dynamics of the immune networks to the fault propagation model and exploit preliminary and precision processes diagnosis for the fault localization.In adaptive immune layer, we suggest a B-PCLONE learning algorithms which recognise the pattern of antigen by using double-learning mechanisms of B-lymphocyte and antibody. Through the continuous supplement and improvement of diagnostic knowledge, the system overcome limited size of learning sample and successfully make the system achieve optimal diagnostic results. In learning algorithms of antibody, we introduced Particle swarm optimization (PSO) into clonal selection algorithm and every candidate detector is regarded as a Particle. The algorithm used Particle optimization evolution equations to guide mutation direction of antibodies for global optimum corporately. Meanwhile, the consequent memory of the primary response of pathogen enables the immune system to mount a more rapid and efficient secondary response to similar faults.This thesis introduces the idea about the distinction between the function of B-lymphocyte and antibody. In the biological humoral immunity, the B- lymphocyte can secrete large numbers of antibodies to recognize and eliminate the antigens. Inspired by the relationship of B-lymphocyte and antibody, the detectors are defined as B-lymphocyte and antibodies which the B-lymphocyte produces. The fault is mapping to B-lymphocyte and the omens is mapping to antibodies. A same fault shows various omens represent as the B-lymphocyte produces antibodies. In a shape-space the various omens of a fault range at the fault. By using such a mechanism, it not only solves the problems that how to distinguish the faults which aroused by the overlap of the omens, but also improves the efficiency and accuracy of the fault detection and enhances the accuracy of continuous study.At last, experiments are undertaken to assess the effectiveness of the proposed model in an induction motor. The results of the detection show that the implemented MIFD can detect the antecedents to the faults. The effects of the continuous learning feature are demonstrated.更多还原