Abstract: Plentiful significant information about the operation status of bearings，which is potential for the fault diagnose after processed properly，is contained in their acoustic signals.In this paper，a new fault diagnosis scheme using acoustic signals is proposed for the bearings by introducing Continuous Gaussian mixture Hidden Markov Model（CGHMM） method，in which the data processing error due to vector quantization is avoided，and therefore the diagnosis precision is improved.Besides，a clustering algorithm and a scaled coefficient algorithm are introduced for parameters initiation and the forward and backward algorithms to simplify the complexity in the computation and improve the training and recognizing speed and diagnosis precision.At last，experiment results of a diagnosis precision achieve to 98.75% and demonstrate the feasibility and potential for applications of the presented scheme.

Key words: bearing, fault diagnosis, Continuous Gaussian mixture Hidden Markov Model（CGHMM）, acoustic signal

摘要： 轴承音频信号包含其运行状态的重要信息，通过分析这些信息就能对轴承故障进行有效诊断。率先引入基于连续高斯混合密度隐马尔可夫模型的轴承故障音频诊断方法，避免矢量量化带来的数据处理误差，提高了系统诊断精度；引入基于聚类算法的模型参数初始化方法和标定系数的前向-后向算法，简化系统复杂度，加快了训练和诊断速度，进一步提高了诊断精度。实验结果表明，诊断精度达到98.75%，具有很好的应用前景。

关键词: 轴承, 故障诊断, 连续高斯混合密度隐马尔可夫模型, 音频信号