中文摘要：

机械装各正在朝着高速、高精、高效方向发展，为了确保这些装备的健康运行，健康监测系统采集了海量数据来反映机械的健康状况，促使机械健康监测领域进入了“大数据”时代。机械大数据具有大容量、多样性与高速率的特点，研究和利用先进的理论与方法，从机械装备大数据中挖掘信息，高效、准确地识别装备的健康状况，成为机械装备健康监测领域面临的新问题。深度学习理论作为模式识别和机器学习领域最新的研究成果，以强大的建模和表征能力在图像和语音处理等领域的大数据处理方面取得了丰硕的成果。结合机械大数据的特点与深度学习的优势，提出了一种新的机械装备健康监测方法。该方法通过深度学习利用机械频域信号训练深度神经网络，其优势在于能够摆脱对大量信号处理技术与诊断经验的依赖，完成故障特征的自适应提取与健康状况的智能诊断，因此克服了传统智能诊断方法的两大缺陷：需要掌握大量的信号处理技术结合丰富的工程实践经验来提取故障特征；使用浅层模型难以表征大数据情况下信号与健康状况之间复杂的映射关系。试验结果表明，该方法实现了多种工况、大量样本下多级齿轮传动系统不同故障位置不同故障类型的故障特征自适应提取与健康状况准确识别。

英文摘要：

Mechanical equipment in modem industries becomes more automatic, precise and efficient. To fully inspect its health conditions, condition monitoring systems are used to collect real-time data from the equipment, and massive data are acquired after the long-time operation, which promotes machinery health monitoring to enter the age of big data. Mechanical big data has the properties of large-volume, diversity and high-velocity. Effectively mining characteristics from such data and accurately identifying the machinery health conditions with advanced theories become new issues in machinery health monitoring. To harness the properties of mechanical big data and the advantages of deep learning theory, a health monitoring and fault diagnosis method for machinery is proposed. In the proposed method, deep neural networks with deep architectures are established to adaptively mine available fault characteristics and automatically identify machinery health conditions. Correspondingly, the proposed method overcomes two deficiencies of the traditional intelligent diagnosis methods： （1） the features are manually extracted relying on much prior knowledge about signal processing techniques and diagnostic expertise; （2） the used models have shallow architectures, limiting their capability in fault diagnosis issues. The proposed method is validated using datasets of multi-stage gear transmission systems, which contain massive data involving different health conditions under various operating conditions. The results show that the proposed method is able to not only adaptively mine available fault characteristics from the data, but also obtain higher identification accuracy than the existing methods.