基于特征加权连续隐马尔可夫模型的故障诊断方法研究

发布时间：2019-03-15 11:51

【摘要】：在现代工业生产中,随着科学技术的进步与发展,作为主要生产工具的机械设备一方面不断向复杂、高速、高效、大型自动化方面发展,另一方面却又面临更加苛刻的工作和运行环境。滚动轴承是石化、电力、冶金、机械、航空航天以及军事工业部门中使用最广泛的机械零件,也是最易损伤的部件之一,其工作状态是否正常对于整个机械设备乃至整条生产线的运行状态有着重大的影响。因此,如何有效地诊断和评估设备的运行状态,从而能够及时采取措施以防止突发事故的发生是当前迫切需要解决的问题。一般来说,机械设备在运行过程中,都会经历从正常直至完全失效的过程,在这过程中,机械设备总会经历一系列不同程度的性能退化状态。因此,在机械设备性能退化的过程中,如果能够准确地监测到其性能退化程度,那么就可以有针对性地制定合理的机械设备维护计划,从而既可以防止设备因故障而临时失效,又可以合理安排生产。当设备发生异常时,性能退化评估可以及时发现并进行故障诊断,防止故障进一步加深,从而提高设备的利用率,缩短设备的停机维修时间。为此,本文以滚动轴承为研究对象,深入开展了基于特征加权连续隐马尔可夫模型(CHMM:Continuous Hidden Markov Model)的故障诊断与性能退化评估的理论体系和技术方法的研究,具体内容如下: 1)从理论分析与工程应用的角度出发,简要阐述了论文的选题背景和研究意义。针对机械设备故障诊断,论述了信号分析与处理技术、智能故障诊断方法、性能退化评估方法等方面的国内外研究现状与发展趋势。在此基础上,给出了本文的主要研究内容。 2)详细介绍了HMM的基本理论,并着重论述了连续HMM的理论,针对其算法中存在的数据下溢、参数初始化等问题给出相应的解决方法。最后,本文简单介绍基于连续HMM故障诊断的基本思想与流程。 3)研究了基于二阶循环平稳分析的谱相关密度组合切片特征提取方法,并将其与连续HMM相结合,提出了一种适用于滚动轴承故障诊断的方法。通过对其进行实验分析,验证了该方法的有效性与可行性。通过与多种特征提取方法比较,结果表明,该方法具有分类准确率高,分类离散度大等优点,可适用于滚动轴承的故障诊断。 4)研究了基于特征加权的连续HMM故障诊断方法。在描述设备状态时通常需要提取多种不同特征,从而形成高维特征。本文研究了基于距离评估技术的特征降维方法,并对其进行参数补偿,对高维特征进行分析,可以得到对分类特征明显的敏感特征,有效的解决了因为人为随机选取特征进行分类造成的分类结果可靠度不高及高维特征灾难等问题。最后通过两个轴承实验进行了验证,与传统方法相比,该方法在保证分类准确率的前提下,有效降低了模式分类器的计算复杂度,提高模式类别的可分性,增强了分类结果的可靠性。 5)研究了基于不完备数据与完备数据两种情况下的滚动轴承性能退化评估方法,特别是在完备数据情况下,提出了一种基于特征加权的连续HMM性能退化评估方法。利用滚动轴承的加速疲劳试验得到的全寿命周期数据,对该评估方法进行了验证,试验分析结果表明基于特征加权的连续HMM性能退化评估方法具有识别性能好,计算量小等优点。在完备数据情况下,考察了性能退化评估模型的推广性,并用试验数据进行了交叉验证,结果表明同一种退化模式下的模型具有较好的推广性。
[Abstract]:In the modern industrial production, with the progress and development of science and technology, the mechanical equipment, which is the main production tool, is continuously developing in the aspects of complex, high-speed, high-efficiency and large-scale automation, and on the other hand, it faces more demanding work and operation environment. The rolling bearing is one of the most widely used mechanical parts in the petrochemical, electric, metallurgical, mechanical, aerospace and military industrial sectors. It is also one of the most damaged parts. The working state of the rolling bearing is of great influence on the operation state of the whole mechanical equipment and the whole production line. Therefore, how to effectively diagnose and evaluate the operation state of the equipment, so as to be able to take measures in time to prevent the occurrence of a sudden accident is a problem that is urgently needed to be solved. In general, in the course of operation, the mechanical equipment will experience a process from normal to complete failure, during which the mechanical equipment will experience a series of varying degrees of performance degradation Therefore, in the process of the degradation of the performance of the mechanical equipment, if the performance degradation degree of the equipment can be accurately monitored, a reasonable maintenance plan of the mechanical equipment can be established in a targeted manner, so that the equipment can be prevented from being temporarily disabled due to failure, and the equipment can be arranged reasonably When the equipment is abnormal, the performance degradation evaluation can timely find and fault the fault, and prevent the fault from further deepening, so as to improve the utilization rate of the equipment and shorten the equipment shutdown and maintenance. In this paper, the theoretical system and technical method of the evaluation of fault diagnosis and performance degradation based on the feature-weighted continuous hidden Markov Model (CHMM: Continuous Hidden Markov Model) are carried out in this paper. (1) From the point of view of the theory analysis and the engineering application, the paper gives a brief account of the background and the research of the thesis. In view of the fault diagnosis of mechanical equipment, the present situation and development of the research and abroad of signal analysis and processing technology, intelligent fault diagnosis method and performance degradation assessment method are discussed. In this paper, the main research of this paper is given. In this paper, the basic theory of HMM is introduced in detail, and the theory of continuous HMM is discussed in detail. The problems of data underflow, parameter initialization and so on in the algorithm are given. Finally, this paper briefly introduces the basic principle of fault diagnosis based on continuous HMM In this paper, the feature extraction method of spectral correlation density combined slice based on the second-order cyclic stability analysis is studied and combined with the continuous HMM, a new method is presented for rolling bearing. The method of fault diagnosis is verified by the experimental analysis. The results show that the method has the advantages of high classification accuracy, large classification, and the like, and can be applied to rolling. The fault diagnosis of the bearing (4) is based on the feature-weighted continuous fault diagnosis. in describe that state of a device, it is often necessary to extract a number of different features In this paper, the feature reduction method based on distance assessment technology is developed, and the parameter compensation is carried out to analyze the high-dimensional features and the classification can be obtained. is characterized by obvious sensitive characteristics and effectively solves the problem that the reliability of the classification result caused by the classification of the human-made random selection characteristic is not high, At last, it is verified by two bearing experiments. Compared with the traditional method, the method effectively reduces the computational complexity of the mode classifier, improves the separability and the enhancement of the mode class, The reliability of the classification results is given.5) The method of evaluating the performance of rolling bearing based on incomplete data and complete data is studied, especially in the case of complete data, a continuous evaluation method based on feature weight is proposed. The evaluation method of the performance degradation of the HMM is carried out by using the full-life cycle data obtained by the accelerated fatigue test of the rolling bearing, and the test result shows that the method for evaluating the performance degradation of the continuous HMM based on the characteristic weighting has the advantages of identification, In the condition of complete data, the generalization of the performance degradation evaluation model is investigated, and the cross-verification is carried out with the test data. The results show that in the same degradation mode
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【参考文献】