Abstract: In the coal mining processes, the operating environments of the belt conveyors were harsh and the working conditions were complex. These resulted in a limited amount of sensor monitoring data and a large amount of noise interference, which seriously limited the accuracy of the residual life prediction. Aiming at this problem, a MDT learning residual life prediction method was proposed. To predict the residual life of key component roller bearings accurately, multiple working condition data of belt conveyors accumulated in coal flow transportation could be fully used. Firstly, integrating a multi-scale convolutional neural network and bidirectional gated recurrent unit(MCNN-BiGRU), a degradation feature extraction model was constructed. The particle swarm optimization(PSO) was used to determine the model hyperparameters. Then, using MDT learning and multiple working condition data, the residual life prediction was carried out. Combining loss of maximum mean discrepancy(MMD) with correlation alignment(CORAL) the data distribution difference of each source domain was narrowed. This might solve the problem of low training accuracy of the model due to the small amount of data. Finally, the actual production data sets of a coal mine were used for verification. The results show that the prediction effectiveness of the MDT-MCNN-BiGRU model is better, and the model performance is further improved after the Savitzky-Golay filter denoising. Using the IMS dataset and comparing with the existing methods, the proposed method has high prediction accuracy and is of some significance in guiding the health management of coal mine transportation equipment.

Key words: belt conveyor, residual life prediction, multiple working condition, feature extraction, multi-source domain transfer（MDT） learning

摘要： 煤矿开采过程中，带式输送机运行环境恶劣、工况复杂，致使获得的传感监测数据量有限且存在大量噪声干扰，严重限制了其剩余寿命预测的准确度。针对该问题，提出了一种多源域迁移学习剩余寿命预测方法，充分利用煤矿运输过程中积累的带式输送机多工况数据，以达到准确预测其关键零部件托辊轴承剩余寿命的目的。首先构建集成多尺度卷积神经网络和双向门控循环单元（MCNN-BiGRU）的设备退化特征提取模型，对单工况数据进行特征提取挖掘，并使用PSO算法确定模型超参数。在此基础上，加入多源域迁移学习（MDT）方法，利用多个工况数据进行剩余寿命预测，通过最大均值差异(MMD）与相互关系对齐（CORAL）联合损失拉近各源域数据分布差异，解决因数据量少导致的模型训练精度不高的问题。最后以煤矿实际生产数据集为例进行实验，结果表明：MDT-MCNN-BiGRU模型的预测效果较好，Savitzky-Golay滤波去噪后模型性能得以进一步提升；使用IMS数据集与现有方法进行比较，发现所提方法预测准确度较高，对煤矿运输设备健康管理具有一定的指导意义。

关键词: 带式输送机, 剩余寿命预测, 多工况, 特征提取, 多源域迁移学习