基于GA-BP神经网络的内燃机系统故障诊断研究

发布时间：2018-01-11 10:37

本文关键词：基于GA-BP神经网络的内燃机系统故障诊断研究　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：内燃机在诸多领域都起到十分重要的作用。例如,大型农业装备、工程机械设备、军工重装武器设备、军舰船舶等,内燃机是名副其实的工业动力。捣固车是大型铁路养护设备,是典型的大机。其动辄百吨的自重工作时完全依靠内燃机为其提供动力,包括车载卫星小车的全部动力和系统供电等。因此,针对捣固车而言,内燃机系统是整个设备的核心之一。当内燃机出现故障时,如果不能被及时发现这可能导致内燃机持续受损,给社会生产带来隐患,而机械设备自身也会带伤运行,影响机械寿命。采用定期入库检修虽然可以缓解这一现象,但是在内燃机运转良好的情况下进行检修既浪费了人力物力,又降低了工作效率。倘若内燃机已经发生故障,而此时按计划未到检修时间,机械设备将带病工作。因而,对内燃机系统进行故障诊断分析,及时发现问题及时处理,这显得重要而有意义。本文通过分析当前故障诊断的常用手段和方法发现,有些诊断模型始终围绕机械振动进行研究、而有些则简单的利用单一算法进行分析。这些都或多或少的存在一些不足和局限。受仿生技术的启发,本文的研究对象和研究手段都进行了改进。鉴于医生对病人的检查方法,本文选取内燃机的排放物作为研究对象,通过内燃机排出的碳氧化物、氮氧化物、碳氢化合物、以及排放物的温湿度、排出物的速率等相关数据进行分析研究。因为内燃机的故障形成都通过其机械状况、气缸内部燃烧情况反映了出来。例如,内燃机燃油系统的喷油嘴堵塞,会造成进入燃烧室的油料偏少,这会导致混合气过稀,燃烧不充分,在排出物中直接反应就是碳氧化物含量增多。本文结合生物神经网络介绍了人工神经网络,经过分析,BP神经网络在某些区域容易陷入极优状态,故提出引入动量项和自适应学习率进行改进。改良优化后的BP神经网络表现出良好的训练性能,但其缺陷不能被彻底消除。而遗传算法GA在全局搜寻方面性能很强,所以将两者进行融合。融合手段即优化BP神经网络的权值和阈值,还可以用遗传算法优化其网络架构。BP算法在网络训练过程中,通过误差反向修正,获得网络待输入的权值和阈值。如此,两种者优点互相补偿,就能充分发挥彼此的算法优点。在建模和实验过程中,基于改良遗传算法优化的BP神经网络,本论文构建了内燃机故障诊断模型,利用DEUTZ的F12L413F风冷柴油机排出物的数据进行研究,并利用MATLAB对故障诊断模型进行仿真分析。对比BP神经网络和优化后的GA-BP神经网络模型的仿真结果可以发现,BP神经网络大致经过342步训练完成收敛,而优化后的GA-BP神经网络经过约60步训练完成收敛。实验数据表明,新改良的GA-BP神经网络结合新的故障诊断对象(内燃机排出物),使内燃机系统的故障诊断精确度有显著提升。
[Abstract]:Internal combustion engines play a very important role in many fields, such as large-scale agricultural equipment, construction machinery equipment, military heavy weapons equipment, warships, and so on. Internal combustion engine is a real industrial power. Tamping car is a large railway maintenance equipment, is a typical large machine. Its 100 tons of deadweight work is completely dependent on the internal combustion engine to provide power for it. Therefore, for tamper, the internal combustion engine system is one of the core of the whole equipment. If it can not be detected in time, it may cause continuous damage to the internal combustion engine, bring hidden trouble to the social production, and the mechanical equipment itself will also run with injury. Although this phenomenon can be alleviated by the use of periodic storage maintenance, it is a waste of manpower and material resources to carry out maintenance under the condition of good operation of internal combustion engine. If the internal combustion engine has broken down and the time of maintenance is not up to schedule, the mechanical equipment will work with malfunction. Therefore, the fault diagnosis and analysis of the internal combustion engine system will be carried out. It is important and meaningful to find problems in time and deal with them in time. By analyzing the common methods and methods of fault diagnosis at present, this paper finds that some diagnosis models are always focused on mechanical vibration. Some are simple to use a single algorithm for analysis. These are more or less there are some shortcomings and limitations. Inspired by the bionic technology. In view of the doctor's examination method of patients, this paper selects the emission of internal combustion engine as the research object, the carbon oxide, nitrogen oxide discharged through the internal combustion engine. Data on hydrocarbons, temperature and humidity of emissions, rate of exhaust, etc., are analyzed because the failure of the internal combustion engine is reflected by its mechanical condition, combustion inside the cylinder, for example. The blockage of fuel injection nozzle in internal combustion engine fuel system will result in less oil entering the combustion chamber, which will lead to the mixture being too thin and the combustion is not sufficient. The direct reaction in the exhaust is the increase of carbon oxides. This paper introduces the artificial neural network combined with the biological neural network. The BP neural network is easy to fall into the optimal state in some areas. Therefore, it is proposed to introduce momentum term and adaptive learning rate to improve the improved BP neural network with good training performance. But its defects can not be completely eliminated. Genetic algorithm (GA) has a strong performance in global search, so the fusion method is to optimize the weight and threshold of BP neural network. Genetic algorithm can also be used to optimize its network architecture .BP algorithm in the process of network training, through the error reverse correction, to obtain the network input weights and thresholds, so that the advantages of the two kinds of compensation each other. In the process of modeling and experiment, based on BP neural network optimized by improved genetic algorithm, the fault diagnosis model of internal combustion engine is constructed. Using DEUTZ's F12L413F air-cooled diesel engine exhaust data were studied. The simulation results of BP neural network and optimized GA-BP neural network model can be found by using MATLAB to simulate the fault diagnosis model. The BP neural network is approximately trained to complete convergence in 342 steps, while the optimized GA-BP neural network has been trained to complete convergence in about 60 steps. The improved GA-BP neural network combined with the new fault diagnosis object (internal combustion engine exhaust) can improve the accuracy of fault diagnosis of internal combustion engine system.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U216.631;TP183

【参考文献】