Bayesian-LSTM Neural Network-based Remaining Useful Life Prediction and Uncertainty Estimation of Rollers in A Hot Strip Mill
朱挺 1陈兆祥 1周笛 2陈震 1胡兵 3潘尔顺1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 上海交通大学机械与动力工程学院 上海 200240
2. 东华大学机械工程学院 上海 201620
3. 上海宝信软件股份有限公司 上海 201203
折叠
摘要
轧辊性能直接影响钢铁轧制流程的生产效率和生产质量,结合轧辊的复杂运行环境和波动工况条件,精准预测轧辊运行状态的时序变化特征与剩余寿命对生产流程精细化、智能化、高效化尤为重要.考虑轧辊服役过程中的动态不确定性,提出一种结合贝叶斯神经网络的长短期记忆网络(Bayesian long short term memory,Bayesian-LSTM)方法,提取健康指标以评估轧辊健康状态,并智能预测轧辊剩余寿命,量化其分布特征的区间不确定性,进一步讨论Bayesian-LSTM网络结构参数对寿命区间的动态影响.通过某热轧厂的实际运行数据验证了方法的有效性,结果表明:所提出方法相对传统数据驱动方法,预测精度达到96.90％,实现了热轧轧辊寿命智能预测和不确定性评估.
Abstract
Roller's performance directly affects the production efficiency and quality of the steel rolling process.Combined with the complex operating environment and fluctuating working conditions,it is especially important to accurately predict the time-series change characteristics of the roller's operating status and the remaining useful life(RUL)for the refined,intelligent and efficient production process.Considering the dynamic uncertainty in operation,Bayesian neural network-long short term memory network(Bayesian-LSTM)method is proposed to extract the roller health index to evaluate the roller's health status and intelligently predict the RUL of the roller and quantify interval uncertainty of the distribution feature.The influence of the constructed Bayesian-LSTM network structure parameters on the lifetime interval is further discussed.And the effectiveness of the method is verified by using the actual industrial data from a hot strip mill.Results show that compared with the traditional data-driven method,the proposed method can not only achieve the prediction accuracy of96.90％,but also evaluate the uncertainty of roller RUL.
关键词
热轧/轧辊/剩余寿命预测/不确定性评估/Bayesian-LSTM
Key words
hot strip mill/rollers/remaining useful life prediction/uncertainty estimation/Bayesian-LSTM
维普
万方数据