中文摘要：

针对设备运行状态识别与故障预测问题，提出一种基于时变转移概率的隐半Markov模型。该模型将设备历史运行信息融人Markov状态转移概率矩阵的估计过程中，使Markov状态转移概率矩阵具有时变特性。基于改进前向后向算法研究了相应的隐半Markov模型参数估计方法，使其能够不断综合利用历史运行信息进行自我更新，以更加符合设备真实运行的过程。同时以该模型为基础，利用故障率方法建立了对设备剩余使用寿命进行预测的基本步骤。通过某滚动轴承运行状态识别实例演示了该模型的建模过程，证明了基于该模型的设备状态识别与预测方法比传统隐半Markov模型方法更为有效。

英文摘要：

Aiming at the problem of equipment operation state identification and fault prognosis, a Duration-Depend- ent Hidden SemFMarkov ModeI（DD-HSMM）was proposed. In this model, the historical operation information was merged into estimation process of Markov state transition probability matrix, thus the matrix had time variant char- acteristics. Furthermore, the parameter estimation method of Hidden Semi-Markov Model（HSMM）was studied based on improved forward-backward algorithm to make self-renewal by using historical operation information. The basic steps for predicting the Remaining Useful Life（RUL）of equipment was built by using fault rate method. Through a case of a rolling hearing＇s operation state to demonstrate the modeling process of proposed model, and the result showed that the proposed method was more effective than traditional HSMM model.