Abstract: In the era of industry 4.0, with the wide application of IoT, fault detection of industrial has great significance to improve the reliability of the equipment. In the real industry scene, it’s difficult to use an unchangeable model to predict the status of equipment, because the relationships among devices are complex and changeable during the running time. In recent years, with the prevalent development of deep learning, it has become a mainstream for fault detection. This paper presents an online fault detection model based on Long Short-Term Memory（LSTM） neural network. A curve-registration method of correlation maximization algorithm is used to feature extraction for multi sensors. Then the paper applies the LSTM neural network to develop a fault detection model, and realizes the online detection and update of the model with the help of sliding window technology. The effectiveness of the proposed model is demonstrated by examining real cases in a power plant. The experimental results show the effectiveness of the proposed method.

Key words: fault detection, feature extraction, Long Short-Term Memory（LSTM） neural network, online update

摘要： 在工业4.0时代，随着IoT的广泛应用，工业设备的故障检测对于提高设备的可靠性具有重要的意义。在实际的工业场景中，由于设备之间的关系复杂多变，难以用统一的模型来表示设备的运行状态。近年来，随着深度学习技术的不断发展与进步，深度学习技术成为故障检测的主流解决方案。提出了一种基于长短记忆神经网络的在线故障检测模型，采用曲线排齐方法对传感器数据进行特征提取，基于长短时记忆神经网络（LSTM）开发故障检测模型，最后借助滑动窗口技术实现了设备故障的在线检测以及模型的在线更新。基于真实的发电厂传感数据进行了实验验证，实验结果表明了该方法的有效性。

关键词: 故障检测, 特征提取, 长短时记忆神经, 在线更新