开题报告---汽车ABS系统智能故障诊断.doc

毕业设计（论文）材料之二（2）安徽工程大学机电学院本科毕业设计(论文)开题报告题目： 汽车ABS系统智能故障诊断 课 题 类 型： 设计 实验研究 论文 学 生 姓 名： 陈首雨学 号： 3092105224专 业 班 级： 自动化2092教 学 单 位： 电气工程学院指 导 教 师： 田丽开 题 时 间： 2013.3.102013年 3月10日开题报告内容与要求一、 毕业设计论文内容及研究意义(1)防抱死制动系统（ABS，Anti-Brake System）是一种汽车主动安全装置，它在制动过程中根据“车辆-路面”状况，采用电子控制方式自动调节车轮的制动力矩来达到防止车轮抱死的目的，增加行车的安全性1。针对ABS系统，研究其执行器和传感器的故障诊断有着重要理论意义及现实意义。（2）神经网络诊断原理在ABS系统执行器和传感器故障诊断中的应用【2】，本文试图从ABS系统执行器和传感器故障诊断的角度研究神经网络诊断的理论问题，即BP神经网络故障诊断原理和方法。（3）利用MATLAB进行仿真来对汽车制动防抱死系统（ABS）进行故障诊断，从而验证神经网络在汽车ABS故障诊断系统中的应用。（只写了研究内容没有写研究意义），再加一点研究意义二、毕业设计（论文）研究现状和发展趋势（文献综述）随着汽车行驶速度提高及道路行车密度的增大，对汽车的行驶安全性的要求越来越高，汽车防抱死制动系统ABS是一种在汽车上日益普及的主动安全装置。它通过轮速传感器检测车轮轮速，经过信号处理后的轮速传输至计算机，计算机根据轮速以一定的算法和控制方法来控制电磁阀增减制动压力，防止车轮抱死。ABS能避免汽车制动过程中的侧滑、跑偏、甩尾和丧失转向操纵能力3，提高汽车的操纵性和稳定性，缩短制动距离；还能避免轮胎的局部磨损，提高轮胎的使用寿命，具有一定的经济价值。普通制动系统在湿滑路面上制动，或在紧急制动的时候，车轮容易因制动力超过轮胎与地面的摩擦力而完全抱死。而ABS是常规刹车装置基础上的改进型技术，可分机械式和电子式两种。它既有普通制动系统的制动功能，又能防止车轮锁死，使汽车在制动状态下仍能转向，保证汽车的制动方向稳定性，防止产生侧滑和跑偏，是目前汽车上最先进、制动效果最佳的制动装置。由于人们对汽车驾驶安全性要求的不断提高以及ABS系统在汽车中的普及，通过对ABS系统故障诊断技术的研究，及时有效的判断其状态，使其长期、安全可靠的运行，对于提高汽车制动系统的可靠性具有十分重要的意义。而目前ABS系统的自诊断系统只能对于断路、短路一些电气故障进行电气检测，当ECU检测到故障时，立即停止ABS功能，并将故障信息以故障码的形式存入到存储器中。如果对故障进行维修后，不及时清除存储器中的故障码，很有可能造成新的故障码与旧的混杂，造成误诊断。因此对于ABS系统智能故障诊断技术的进一步研究是非常必要的。三、毕业设计（论文）研究方案及工作计划（含工作重点与难点及拟采用的途径）设计的重点与难点：1应用汽车整车运动的力学模型，分析制动过程中的运动情况2利用MATLAB-SIMULINK对整车系统进行建模，并建立ABS执行器和传感器发生故障时的故障模式，采集故障数据，应用BP神经网络进行泛化，从而进行故障诊断。拟采用的途径：1.调研，查阅相关资料，搜集样本数据；2.确定神经网络的输入和输出向量；3.抽取部分样本数据作为训练样本，利用BP神经网络进行训练；4.将剩余的样本数据作为检验样本，用上述训练好的神经网络分别进行仿真检验，通过对诊断结果与实际故障类型的比较、分析，找出故障诊断准确率相对最高的神经网络。具体流程如下： 查阅资料和搜集数据 确定神经网络的输入和输出向量收集训练样本，利用BP神经网络进行训练收集检验样本，用上述训练好的神经网络分别进行仿真检验验验比较仿真结果，进行分析，找出故障诊断准确率相对最高的神经网络设计（论文）进度计划起止日期（日/月）内 容 进 程2/28-3/93/9-3/153/16-3/223/23-3/293/30-4/54/6-4/124/13-4/194/20-4/264/27-5/35/4-5/105/11-5/175/18-5/245/25-5/316/1-6/76/8-6/146/15-6/21与导师联系,获得课题,写开题报告搜索相关资料（包括图书馆和网上检索）整理、消化资料，理清思路写读书报告对BP神经网络故障诊断和汽车ABS系统基本了解研究当执行器和传感器发生故障时会产生什么样的现象建立汽车ABS系统故障诊断模型，并进行仿真对神经网络原理和方法学习和消化建立基于BP神经网络的汽车ABS智能故障诊断的设计方案对方案进行初步拟稿最终设计确定利用MATLAB进行仿真对调试结果进行分析评价得出预期的结论撰写毕业论文检查并修改毕业设计最终定稿准备答辩四、主要参考文献（不少于10篇，期刊类文献不少于7篇，应有一定数量的外文文献，至少附一篇引用的外文文献（3个页面以上）及其译文）1 周志立，徐斌. 汽车ABS原理与结构M. 机械工业出版社, 2004.1101122 陈丙珍.人工神经网络在过程工业中的应用J.中国有色金属学报（工学版），2004.5，14(1)：106111.3 陈朝阳，张代胜，任佩红.汽车故障诊断专家系统的现状与发展趋势J.机械工程学报，2003.11，39(11)：16.4 王耀南，孙炜 智能控制理论及应用J.机械工业出版社，2011.7，4869.5 王海英，袁丽英 吴勃 控制系统的MATLAB仿真与设计，高等教育出版,2011.8 1221996 王仲生.智能故障诊断与容错控制M.西北工业大学出版社,2005.4， 240250.7 肖永清,杨忠敏.汽车制动系统的使用与维修M.中国电力出版社,2004，3123548 董长虹. 神经网络与应用.北京：国防工业出版社，2005，103-1059 陈丙珍.人工神经网络在过程工业中的应用J.中国有色金属学报（工学版），2004.5，14(1)：10611110 Henrik NiemannFault.Tolerant Control based on Active Fault Diagnosis,1996,12：95135.11 BHARITKAR S,MENDELJM.The hysteretic Hopfield neural networkJ.IEEE Trans on Neural Networks,2000,11(4)：897888附:参考外文文献及其译文Soft computing methods in motor fault diagnosisAbstract During the last decade, soft computing (computational intelligence) has attracted great interest from different areas of research. In this paper, we give an overview on the recent developments in the emerging field of soft computing-based electric motor fault diagnosis. Several typical fault diagnosis schemes using neural networks, fuzzy logic, neural-fuzzy, and genetic algorithms, with descriptive diagrams as well as simplified algorithms are presented. Their advantages and disadvantages are compared and discussed. We conclude that soft computing methods have great potential in dealing with difficult fault detection and diagnosis problems. 1. Introduction The ac and dc motors are intensively applied in various industrial applications . Changing working environment and dynamical loading always strain and wear motors and cause incipient faults such as shorted turns, broken bearings, and damaged rotor bars .These faults can result in serious performance degradation and eventual system failures, if they are not properly detected and handled. Improved safety and reliability can be achieved with appropriate early fault diagnosis strategies leading to the concept of preventive maintenance. Furthermore, great maintenance costs are saved by applying advanced detection methods to find those developing failures. Motor drive monitoring, fault detection and diagnosis are, therefore, very important and challenging topics in the electrical engineering field.Soft computing is considered as an emerging approach to intelligent computing, which parallels the remarkable ability of the human mind to reason and learn in circumstances with uncertainty and imprecision. In contrast with hard computing methods that only deal with precision, certainty, and rigor, it is effective in acquiring imprecise or sub-optimal, but economical and competitive solutions to real-world problems. As we know, qualitative information from practicing operators may play an important role in accurate and robust diagnosis of motor faults at early stages. Therefore, introduction of soft computing to this area can provide us with the unique features of adaptation, flexibility，and embedded linguistic knowledge over conventional schemes . An up-to-date presentation of motor fault detection and diagnosis methods was recently published on a special section. This overview is organized as follows. First, we give a concise introduction to the conventional motor fault diagnosis in Section 2. Soft computing-based approaches, including operating principles, system structures, and computational algorithms, are then discussed in the following sections. We present a few interesting motor fault diagnosis schemes using soft computing methods, such as neural networks, fuzzy logic, neural-fuzzy, and genetic algorithms (GAs) in Sections 36, respectively. Their advantages and disadvantages are also briefly reviewed and compared. Some conclusions are finally drawn at end of the paper.2. Conventional motor fault diagnosis methods There exist numerous conventional approaches for motor fault detection and diagnosis. The most straightforward method is the direct inspection. It requires careful check-over of the condition of individual motor components to find defective faults. A similar procedure is named particle analysis of lubricate oil of the motor, if the motor has a gear box with oil lubrication. The oil is first sampled and then taken for laboratory check, which detects the possible faults. This will, however, result in a time consuming and costly examination. The above two approaches are more suitable, on the other hand, for routine maintenance. Classical parameter estimation methods can also be reasonably applied for motor fault detection and diagnosis problems. The underlying idea is that based on some measurement signals from the actual motor, we use parameter identification techniques to estimate relevant information of the motor working condition. Fig. 1 illustrates this kind of fault detection process. The parameter estimation strategy is well-suited for real-time cases. Nevertheless, it requires a deep understanding of the operating principle of the motor as well as an accurate mathematical model. In addition, with the aging of the motor, the original model becomes less accurate. During the past few years, soft computing has been employed to overcome the aforementioned difficulties that conventional diagnosis strategies are facing. In general, soft computing methods consist of three essential paradigms: neural networks, fuzzy logic, and GAs (evolutionary computation) .In our paper, we discuss the recent progresses of soft computing methods-based motor fault diagnosis. The applications of neural networks, fuzzy logic, and GAs together with their fusion, e.g. neuron-fuzzy, in this motor fault detection and diagnosis area will be presented in the following sections, respectively.Fig. 1. Motor fault diagnosis using parameter estimation scheme3. Neural networks-based motor fault diagnosisDue to their powerful nonlinear function approximation and adaptive learning capabilities, neural networks have drawn great attention in the motor fault diagnosis field. Chow and his colleagues have carried out comprehensive investigation on various neural networks-based fault detection schemes .They proposed a typical Back-propagation (BP) neural network structure for incipient motor faults diagnosis, as illustrated in Fig. 2 . The incipient faults here refer to the turn-to-turn insulation and bearing wear in a split-phase squirrel-cage induction motor. In Fig. 2, I is the steady-state current of the stator, the rotor speed, and Nc and Bc are the conditions of the motor winding insulation and bearing. From the characteristic equations of an induction motor, we know that the relationships between inputs (I, ) and outputs (Nc, Bc) are highly nonlinear. Thus, a BP neural network is applied to approximate this relationship. The training structure is shown in Fig. 3. The values of I and can be obtained easily from the on-line measurement data. In fact, the inputs of the BP neural network in Fig. 2 could be further expanded to include higher orders of I and , e.g. I2 and 2, which would increase the convergence speed. On the other hand, Nc and Bc should be evaluated by a human expert as Fig. 3 shows. More precisely, based on the observation of the working condition and qualitative fault diagnosis knowledge of a training motor, the values of Nc and Bc, which quantitatively describe the motor, are classified into three condition levels, good, fair，bad, to yield Nc and Bc, respectively. After the neural network has been trained to learn diagnosis experience from the expert, it is employed on-line as illustrated in Fig.4. Judging from the motor operating condition, stator current and rotor speed, the neural network can indicate incipient faults according to the above three fault levels. Filippetti et al. proposed a similar BP neural network-based motor fault diagnosis scheme to detect the number of broken rotor bars. The training data for the neural network is acquired from healthy as well as simulated faulty machines. Their promising scheme has the diagnosis accuracy of 100% in simulations.Fig. 2. BP neural network for incipient fault detectionFig. 3. Training phase for neural network-based motor fault detectionFig. 4. Neural network-based motor fault detection From the discussions above, it is concluded that the motivation of employing neural networks for motor fault diagnosis is due to their self-adaptation and nonlinear approximation abilities, which can set up the relationship between the indication of faults and available measurement signals. However, the critical shortcoming of neuralnetworks-based motor fault diagnosis is that qualitative and linguistic information from the operator of motors cannot be directly utilized or embedded in the neural networks because of their numerically oriented black-box structures. Additionally, it is even difficult to interpret the input and output mapping of a trained neural network into meaningful fault diagnosis rules.4. Fuzzy logic-based motor fault diagnosis To take advantage of linguistic fault diagnosis knowledge explicitly, numerous motor fault diagnosis methods using fuzzy logic have been studied. Nejjari and Benbouzid applied fuzzy logic to the diagnosis of induction motor stator and phaseconditions. Their diagnosis structure, whose kernel is just a representative fuzzy reasoning system including a fuzzification interface, inference engine, fuzzy rulebase, and a defuzzification unit, is illustrated in Fig. 5. The conditions of the stator and phases are represented with three rectangular membership functions, i.e. good, damaged, and seriously damaged. Totally, there are 12 heuristic IFTHEN fuzzy inference rules applied to detect the two aforementioned faults, for instance1. IF Ib is small THEN the stator is damaged.2. IF Ic is medium THEN the stator is in good condition. This diagnosis approach achieves 91.7% accuracy in detecting severe conditions and 100% accuracy at both good and bad conditions of the bearing. Fuzzy logic-based motor fault diagnosis methods have the advantages of embedded linguistic knowledge and approximate reasoning capability. However, the design of such a system heavily depends on the intuitive experience acquired from practicing operators. The fuzzy membership functions and fuzzy rules cannot be guaranteed to be optimal in any sense. Furthermore, fuzzy logic systems lack the ability of self-learning, which is compulsory in some highly demanding real-time fault diagnosis cases. The above two drawbacks can be partly overcome by the fusionof neural networks and fuzzy logicneural-fuzzy technique.5. Motor fault diagnosis using neural-fuzzy techniqueAs we know, both neural networks and fuzzy logic have their own advantages and disadvantages. The major drawbacks of BP neural network are its black-box data processing structure and slow convergence speed. On the other hand, fuzzy logic has a similar inference mechanism to the human brain, while it lacks an effective learning capability. Auto-tuning the fuzzy rules and membership functions may be difficult in a classical fuzzy logic system. In a word, neural networks are regarded as model free numerical approaches, and fuzzy logic only deals with rules and inference on a linguistic level. Therefore, it is natural to merge neural networks and fuzzy logic intoa hybrid systemneural-fuzzy, so that both of them can overcome their individual drawbacks as well as benefit from each others merits. In fact, neural-fuzzy technique has found many promising applications in the field of motor fault diagnosis. Although fuzzy neural networks own the advantages from both neural networks and fuzzy logic, most of the existing models, such as ANFIS, cannot deal with fuzzy input/output information directly. A bearing fault diagnosis problem is employed as a test bed for this approach. Simulations demonstrated that their method cannot only successfully detect bearing damages faults but also provide a corresponding linguistic description.6.Genetic algorithms-based motor fault diagnosis A GA is a derivative-free and stochastic optimization method 31. Its orientation comes from ideas borrowed from the natural selection as well as evolutionary process. As a general purpose solution to demanding problems, it has the unique features of parallel search and global optimization. In addition, GA needs less prior information about the problems to be solved than the conventional optimizationschemes, such as the steepest descent method, which often require the derivative of objective functions. Hence, it is attractive to employ a GA to optimize the parameters and structures of neural networks and fuzzy logic systems instead of using the BP learning algorithm alone. In principle, the training of all the motor fault diagnosis methods discussed above can be implemented using GAs. For instance, Vas introduced GA into the parameter estimation of an induction motor.Betta et al. discussed the use of GA to optimize a neural network-based induction motor fault diagnosis scheme, which is conceptually illustrated in Fig. 5. The diagnosis performance is encouraging: the percentage of correct single-fault detection is higher than 98%. Moreover, it can also cope with double-fault, with correct diagnosis of both faults in about 66% of the considered cases and of at least one fault in about 100% of the cases.Fig. 5. Application of GA in neural network-based motor fault diagnosisSince GA is only an auxiliary optimization method, it cannot be applied independently in practice. The combination of GA with other motor fault diagnosisschemes has demonstrated enhanced performance in global and near-global minimum search. However, optimization with GA often evolves heavy computation, and is therefore quite time-consuming. Targeted at real-time fault diagnosis, fast GAs with parallel implementation to improve the convergence speed have to be developed.7. ConclusionsIn this paper, we gave an overview on the recent progresses of soft computing methods-based motor fault diagnosis systems. Several motor fault diagnosis.techniques using neural networks, fuzzy logic, neural-fuzzy, and GAs were concisely summarized. Their advantages and drawbacks were discussed as well. Based on our observations, we conclude that emerging soft computing methods can provide uswith improved solutions over classical strategies to challenging motor fault diagnosis problems. However, they are not supposed to compete with conventional methods. Instead, more accurate and robust diagnosis approaches should be developed based on the fusion of these two categories of methodologies, soft computing and hard computing. This overview paper is the starting point for our future research activitiesin the field of soft computing-based fault diagnosis of electric motors. Acknowledgements The authors would like to thank the anonymous reviewer for his insightful comments and constructive suggestions that have improved the paper. This researchwork was funded by the Academy of Finland.电机故障诊断的软计算方法摘 要在过去的十年里，软计算(计算智能)引起了来自不同领域研究的极大兴趣。在本文中，我们对基于软计算的电机故障诊断这个新兴领域的最近发展事态进行了概述。几个典型故障诊断方案运用生动的图表以及简化算法来利用神经网络、模糊逻辑、神经模糊和遗传算法。他们的优缺点被进行了比较和讨论。我们认为软计算方法有极大的潜力在于处理困难的故障检测和诊断问题。1. 介绍 交流和直流电机广泛应用在各种工业应用。改变的工作环境和动态加载总是拉紧和磨损电动机而且导致例如短路、轴承破碎和转子条损坏的潜在故障。如果得不到正确的检测与处理，这些错误可能导致严重的性能退化和最终的系统故障。提高安全性和可靠性才能实现良好的早期故障诊断策略，这个策略会让预防性维护保养的概念得以产生。此外, 采用先进的检测方法去寻找那些发展中的失败，使得大量的维护成本得以保留。因此,马达驱动监测、故障检测与诊断在电气工程领域是非常重要的和富有挑战性的课题。软计算方法作为一个新兴的智能计算,匹敌人类头脑推理和学习不确定性和不精确情况