基于隐马尔可夫模型与EM算法的复杂机械系统故障诊断的研究

发布时间：2018-04-10 11:15

本文选题：隐马尔可夫模型（HMM） + EM算法　；参考：《华中科技大学》2012年硕士论文

【摘要】：二十世纪以来，随着工业生产的不断发展和进步，机械系统的可靠性和安全性问题日益突出。及时了解和掌握机械设备在运行过程中的状态，整体或局部是正常还是异常，如果机器发生故障，如何识别故障类型，这些对于生产的监控系统来说是至关重要的。隐马尔可夫模型（HiddenMarkovModel，HMM）是一种基于统计学理论的模式识别方法，已经广泛应用于语音识别领域。基于振动信号和语音信号的的相似性，将隐马尔可夫模型（HMM）应用于机械故障诊断中。本文系统地介绍了隐马尔可夫模型的基本理论并利用旋转机械故障诊断的例子来说明隐马尔可夫模型在故障诊断中的应用。HMM的基本理论主要包括HMM的基本元素、基本假设，以及HMM的三个基本问题及其解决方法。并详细地推导了解决HMM估计问题的前向-后向算法，解决HMM译码问题的Viterbi算法。介绍了EM(ExpectationMaximization)算法，在此基础上，采用组合的方法—多观测序列概率是各单观测序列概率的组合，更方便对多观测序列给出不同的相关性假设，并引入了相关定理，具体地推导出基于多观测序列的HMM参数重估公式。引入了时间可逆的隐马尔可夫链和可逆的判定定理。高阶HMM考虑了状态转移概率及观测信号的输出概率这两个概率和系统历史状态的关联性，对观测信号具有有更强的识别能力，本文介绍了高阶HMM的定义，以及高阶HMM如何等价地转化为一阶HMM，使得一阶HMM的理论方法能够应用于任一高阶HMM。最后本文讲述了隐马尔可夫模型在机械故障诊断中的应用技术，，以及在MATLAB环境下的计算机实现。
[Abstract]:Since the 20th century, with the continuous development and progress of industrial production, the problems of reliability and safety of mechanical system have become increasingly prominent.It is very important to know and master the state of mechanical equipment in the process of operation, whether the whole or part of the machine is normal or abnormal, and how to identify the type of failure if the machine breaks down, which is very important for the monitoring system of production.Hidden Markov Model (HMMM) is a pattern recognition method based on statistical theory and has been widely used in the field of speech recognition.Based on the similarity between vibration signal and speech signal, hidden Markov model (HMMM) is applied to mechanical fault diagnosis.And three basic problems of HMM and their solutions.A forward backward algorithm for HMM estimation and a Viterbi algorithm for HMM decoding are derived in detail.In this paper, EMN expectation maximization algorithm is introduced. On the basis of this, the method of combination is adopted-the probability of multiple observation sequences is the combination of the probabilities of each single observation sequence, it is more convenient to give different correlation hypotheses for multiple observation sequences, and the correlation theorem is introduced.The reestimation formula of HMM parameters based on multiple observation sequences is derived in detail.In this paper, we introduce a time-reversible hidden Markov chain and a decision theorem of invertibility.High order HMM considers the correlation between the state transition probability and the output probability of the observed signal and the historical state of the system. It has a stronger ability to recognize the observed signal. The definition of high order HMM is introduced in this paper.And how the high-order HMM can be equivalent to first-order HMMs, so that the theoretical method of first-order HMM can be applied to any high-order HMMs.Finally, this paper describes the application technology of hidden Markov model in mechanical fault diagnosis, and the computer implementation in MATLAB environment.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：O211.62;TH165.3

【参考文献】