HHMM has many advantages for state recognition and more accurately calculates recognition results in the form of probability, in comparison with traditional hidden Markov model(HMM).Model parameters increased exponentially with the increasing equipment state.In view of this,dynamic Bayesian network was introduced,which can effectively reduce the computational complexity and decrease the inference time. Accordingly, HHMM was expressed as dynamic Bayesian network, which identified

health status by utilizing vibration signals of pretreatment. In order to avoid the limitations of the current state classifications,the optimization of the condition numbers was proposed,on the basis of combination of K-means algorithm and cross-validation.It also investigated the basic framework for HHMM state recognition and calculation process based on full life test for gearbox, which provides a new way for state recognition of complex equipment.

hierarchical hidden Markov model(HHMM), state recognition, dynamic Bayesian network, optimization of , condition number

与传统的隐Markov模型(HMM)相比较而言，应用分层隐Markov模型(HHMM)对设备进行状态识别有诸多优点，而且能以概率的形式更为精确地计算识别结果。针对模型参数随着设备状态的增加呈指数倍增这一问题，引入动态贝叶斯网络这一新的方法，由于该方法可以有效地降低模型的计算复杂度并缩短推理时间，所以将HHMM表达为动态贝叶斯网络，利用预处理的振动信号对设备的健康状态进行识别;针对现有状态分类方法的局限性，提出了基于K均值算法和交叉验证方法相结合的状态数优化方法;以齿轮箱全寿命实验为依据，对该模型实现状态识别的基本框架和计算过程进行了研究，研究结果为复杂设备的状态识别提供了新的思路。

分层隐Markov模型, 状态识别, 动态贝叶斯网络, 状态数优化