动态服役工况下盾构机刀盘系统性能退化机制与集成深度学习监测评估方法研究

The shield machine is a high-end major equipment urgently needed for the construction of urban underground space infrastructure in China. The cutterhead system is the core component of the shield machine, and its performance degradation directly affects the quality, efficiency, safety and cost of the tunneling process. Affected by the complex and variable construction environment, there is no effective real-time monitoring and assessment method for performance degradation of the cutterhead system at present. The degradation mechanism is still unclear. This project proposes a decoupling extraction method for degraded features of the cutterhead system under dynamic operating regimes, which solves the problem of the system's true degradation information being masked by the working conditions. We propose the deep degraded feature extraction and enhancement methods combining multi-source information, to achieve self-learning of degraded features and improve their predictive ability, meeting the different requirements of degraded pattern recognition and state assessment. We establish the hybrid degradation pattern recognition model and state assessment model based on ensemble deep learning, to achieve comprehensive and accurate perception of the degraded evolution process, and to identify the coupling effects of geological environment and tunneling control parameters on degradation, revealing the degradation mechanism of the cutterhead system under dynamic operating regimes. The research results will be verified in the field of shield tunneling. This project is expected to make innovative breakthroughs in several aspects, including the extraction and enhancement of the degraded features of the shield cutterhead system and the real-time perception of the degradation state under dynamic operating regimes. The research results will provide theoretical basis and technical support for the intelligent maintenance of shield cutterhead system.

盾构机是我国城市地下空间基础设施建设亟需的高端重大装备。刀盘系统是盾构机掘进的核心部件，其性能退化直接影响掘进质量、效率、安全和成本。受复杂多变、恶劣施工环境影响，当前还未有有效的刀盘系统性能退化实时监测评估手段，其退化机制尚不明确。本项目提出动态服役工况下刀盘系统退化特征的运行工况解耦提取方法，解决系统真实退化信息受工况变化掩蔽问题；面向退化模式识别和状态评估不同需求，提出多源信息融合退化特征深度提取与增强方法，实现退化特征自主学习并提高其预测辨识能力；建立基于集成深度学习的混合退化模式识别和状态评估模型，实现退化演变过程全面准确感知，探明地质环境与掘进控制参数对退化的耦合作用规律，揭示动态服役工况下刀盘系统性能退化机制。研究成果将在盾构掘进现场应用验证。本项目研究可望在动态服役工况下盾构机刀盘系统退化特征提取与增强、退化状态实时感知方面有所创新突破，为其智能维护提供理论基础和技术支撑。

盾构机是我国城市地下空间基础设施建设亟需的高端重大装备。刀盘系统是盾构机掘进的核心部件，其性能退化直接影响掘进质量、效率、安全和成本。受复杂多变、恶劣施工环境影响，当前还未有有效的刀盘系统性能退化实时监测评估手段，其退化机制尚不明确。项目提出动态服役工况下盾构机刀盘系统性能退化机制研究。针对土压平衡盾构机，系统研究了其刀盘系统机械结构，开展了滚刀失效FMECA分析、刀具磨损机理分析、刀具磨损影响因素分析，并结合专家经验进行了施工参数匹配分析。针对动态服役工况下刀盘系统退化特征的运行工况解耦提取问题，联合采用机理知识和数据驱动的特征构建方法，提出融合抗工况干扰能力、单调性、趋势性的综合特征筛选指标，从而提高特征表现刀具磨损退化的能力和鲁棒性。在此基础上提出基于无监督学习的刀盘退化状态评估方法，构建以LSTM为编码器、解码器的seq2seq模型，通过与其他输入参数和特征构造方法及算法模型结果的对比，显示所提出的技术方案获得最佳的表现。针对多源信息融合退化特征深度提取与退化状态评估问题，提出基于CNN-GRU网络的刀盘健康评估方法，综合利用CNN网络和GRU网络分别提取空间信息和时序信息的能力进行深层退化特征提取，对比单纯的CNN网络和GRU网络，显示所提出的综合网络模型具有最高的准确率，验证了其特征提取能力。针对刀盘系统性能退化问题，构建了以刀盘系统历史健康状态指标曲线作为因变量，以与刀盘健康状态相关的其他运行状态和工况参数作为协变量的退化趋势预测模型。针对单步预测，提出基于LSTM网络的预测模型，对比普通BP神经网络和RNN网络，显示获得最佳预测效果。针对多步预测，提出融入注意力机制的Seq2Seq模型，以Bi-LSTM作为编码器，以LSTM作为解码器，进行未来十个时间步的退化趋势预测，对比显示所提出的模型具有最高的预测精度。本项目研究可为大数据驱动的动态服役工况下盾构机刀盘系统退化特征提取与增强、退化状态实时感知和预测提供理论基础和技术支持，助力实现刀盘系统智能维护。

内容获取失败，请点击重试