基于时间序列的柴油机空气系统健康状态评估与预测方法

王永康，文冠华，王彦岩^*，沈照杰，马琮淦，纪兆圻，林波，邵麓铭

哈尔滨工业大学（威海），山东 威海  264200

摘要：为解决柴油机空气系统因结构复杂、故障频发、全生命周期数据不足及故障精确表征困难导致的柴油机空气系统健康状态评估难题，为其预测性维护提供支撑，提出一种基于时间序列的柴油机健康状态评估方法，根据单位时间内空气系统健康状态构建初始健康指标，采用自编码器模型对初始指标进行特征加权优化以提高表征精度，按照时间序列结合滑动窗口法对历史健康状态进行平滑处理，构建长短期记忆（long-short term memory，LSTM）网络，预测柴油机空气系统的健康状态。结果表明：设计的基于时间序列的柴油机空气系统健康状态评估方法有效且预测准确度较高，健康状态预测结果与实测结果的均方误差为2.11×10^-5，均方根误差为0.004 6，平均绝对误差为0.003 1，该方法可以为柴油机空气系统预测性维护提供可靠支撑。

关键词：柴油机；健康状态评估；健康状态预测；LSTM网络

Health status assessment and prediction method for a diesel engine air system based on time series

WANG Yongkang, WEN Guanhua, WANG Yanyan^*, SHEN Zhaojie, MA Conggan,

JI Zhaoqi, LIN Bo, SHAO Luming

School of Automotive Engineering, Harbin Institute of Technology (Weihai), Weihai 264200, China

Abstract: To address the challenge of health status assessment for diesel engine air systems, when considering their complex structure, high fault frequency, insufficient full-life-cycle data, and difficulties in accurate fault characterization, and to provide support for predictive maintenance, a time-series-based health status assessment method for diesel engines is proposed. Initial health indicators are constructed according to the health status of the air system per unit time. An autoencoder model is adopted to perform feature weighting optimization on the initial indicators for improved characterization accuracy. Combined with the sliding window method, time-series analysis is applied to smooth historical health status data. Finally, a long short-term memory (LSTM) network is established to predict the health status of the diesel engine air system. The results demonstrate that the designed time-series-based assessment method is effective with high prediction accuracy. The mean squared error (MSE) between the predicted and measured results of health status is 2.11×10^-5, the root mean squared error (RMSE) is 0.004 6, and the mean absolute error (MAE) is 0.003 1. This method can provide reliable support for the predictive maintenance of diesel engine air systems.

Keywords: diesel engine; health status assessment; health status prediction; LSTM network