College of Mechanical and Vehicle Engineering, Taiyuan University of Technology
Shanxi Key Laboratory of Fully Mechanized Coal Mining Equipment

刮板输送机链传动系统由于承受复杂载荷作用导致故障频发，然而传统的故障诊断需要大量的先验知识和主观干预，对技术人员要求高。为实现刮板输送机链传动系统故障预警的自主性、准确性与高效性，利用深度学习强大的数据挖掘能力，提出了基于LSTM−Adam的刮板输送机链传动系统故障预警方法。首先，基于组态技术搭建刮板输送机工况监测系统，采集减速器输出轴转矩及转速、中部槽中板压力、刮板竖直方向振动加速度及刮板链运行方向应变等刮板输送机实时运行数据，并对数据进行清洗和min−max归一化处理，为故障预警提供数据支撑；然后，基于LSTM搭建预测模型，并采用Adam优化算法对其进行训练和优化，得到最优LSTM−Adam预测模型；最后，将刮板输送机实时运行数据导入LSTM−Adam预测模型，得到刮板输送机运行参数预测值，使用滑动加权平均法计算预测值与真实值之间的残差，并将正常运行工况下同类数据的最大残差作为预警阈值，当残差超过预警阈值时进行预警。试验结果表明：LSTM−Adam预测模型能够准确预测出刮板链应变数据的变化趋势，并对卡链与断链故障准确做出预警。

The scraper conveyor chain transmission system is prone to frequent faults due to its complex load bearing capacity. However, traditional fault diagnosis requires a large amount of prior knowledge and subjective intervention, which requires high technical personnel. In order to achieve the autonomy, accuracy, and efficiency of fault warning for the scraper conveyor chain transmission system, a fault warning method for the scraper conveyor chain transmission system based on LSTM-Adam is proposed using the powerful data mining capability of deep learning. Firstly, a monitoring system for the working conditions of the scraper conveyor is built based on configuration technology. The system collects real-time operating data of the scraper conveyor, such as the torque and speed of the output shaft of the reducer, the pressure of the middle groove plate, the vibration acceleration in the vertical direction of the scraper, and the strain in the running direction of the scraper chain. The data is cleaned and normalized in min-max to provide data support for fault warning. Secondly, a prediction model is built based on LSTM and trained and optimized using the Adam optimization algorithm to obtain the optimal LSTM Adam prediction model. Finally, the real-time operating data of the scraper conveyor is imported into the LSTM-Adam prediction model to obtain the predicted values of the scraper conveyor operating parameters. The sliding weighted average method is used to calculate the residual between the predicted value and the true value. The maximum residual of the same type of data under normal operating conditions is used as the warning threshold. When the residual exceeds the warning threshold, an early warning is given. The experimental results show that the LSTM-Adam prediction model can accurately predict the trend of strain data of the scraper chain and provide accurate warnings for stuck chain and broken chain faults.

国家自然科学基金项目（52204149）；山西省科技重大专项计划“揭榜计划”项目（202101020101021）；山西省基础研究计划项目（202103021223080）；山西省研究生实践创新项目（2023SJ052）；太原理工大学校基金项目（2022QN006）。

李博，郭星燃，李娟莉，等. 基于LSTM−Adam的刮板输送机链传动系统故障预警方法[J]. 工矿自动化，2023，49（9）：140-146.

LI Bo, GUO Xingran, LI Juanli, et al. A fault warning method for scraper conveyor chain transmission system based on LSTM-Adam[J]. Journal of Mine Automation，2023，49（9）：140-146.