软测量辅助下丝状菌污泥膨胀的成因分析和预测维护

Diagnosis and maintenance of filamentous sludge bulking efficiently and timely are of great importance to guarantee the safety of a wastewater treatment system. The main bottlenecks limiting their wide applications are the difficult-to-capture faulty signals, hard-to-measure induced factors and unavailability of efficient maintenance strategies. Firstly, types, modes, failure mechanism and dynamic evolution law of faults with respect to filamentous sludge bulking are investigated under the effect of internal and external uncertainties; Secondly, Relevant Vector Machine (RVM) based and Gaussian Processes (GP) based adaptive and multi-output soft-sensors are proposed to mine hard-to-measured induced factors and fault information; Thirdly, deep kernel and broad kernel Canonical Correlation Analysis (CCA) as well as nonlinear multivariate Granger Causality are studied, which are able to explore the fault determination, propagation and location; Finally, non-stationary kernel, multi-step and multi-output soft-sensors are proposed for Remaining Useful Life (RUL) prediction, also, a reliability-centered repair and preventive maintenance decision-making expert system is constructed to provide scientific basis of optimal maintenance. This study contributes to the theories of fault diagnosis and maintenance for filamentous sludge bulking, and is of great significance in theory and practical value in engineering.

及时有效地诊断和维护丝状菌污泥膨胀过程是保障污水处理系统安全运行的关键。然而，污泥膨胀故障的难以捕捉、致因因素的复杂难测、维护决策难以为继成为了制约污泥膨胀管控的一个主要瓶颈。本项目首先研究内部和外部不确定性作用下污泥膨胀的故障发生的类型、模式、失效机理和动态演变规律；其次，研究基于自适应多输出相关向量机和自适应多输出高斯过程模型对难以测量致因因素和故障信息的挖掘；再次，研究深度核和宽度核典型相关分析模型、非线性多元格兰杰因果模型，探明故障判定、故障传播、故障定位的机理；最后，研究基于非稳定核的多步多输出软测量模型的寿命预测方法以及建立以可靠性为中心的维修与预防性维护多属性专家决策模型，为操作人员制定最佳维修与预防性维护方案提供科学合理的依据。本项目的研究成果将为构建丝状菌污泥膨胀故障诊断和维护一体化的监控系统奠定理论和技术基础，具有重要的理论意义和工程应用价值。

及时有效地诊断和维护丝状菌污泥膨胀过程是保障污水处理系统安全运行的关键。然而，污泥膨胀故障的难以捕捉、致因因素的复杂难测、维护决策难以为继成为了制约污泥膨胀管控的一个主要瓶颈。本项目实现了内部和外部不确定性作用下污泥膨胀的故障发生的类型、模式、失效机理和动态演变规律;其次，实现了基于自适应多输出相关向量机和自适应多输出高斯过程模型对难以测量致因因素和故障信息的挖掘; 再次，实现了深度核典型相关分析模型、非线性多元格兰杰因果模型，探明故障判定、故障传播、故障定位的机理;最后，实现了多步多输出软测量模型的寿命预测方法以及建立以可靠性为中心的维修与预防性维护多属性专家决策模型，为操作人员制定最佳维修与预防性维护方案提供科学合理的依据。相关算法最终在BSM1、BSM2、北京某污水厂的数据上对提出的软测量建模方法、故障诊断、故障预测和维护框架进行了仿真和验证。本项目的研究成果为构建丝状菌污泥膨胀故障诊断和维护一体化的监控系统奠定理论和技术基础，具有重要的理论意义和工程应用价值。相关研究发表SCI论文23篇，EI论文8篇，核心论文1篇，申请发明专利9项，访问丹麦科技大学2人次，提交年度进展报告3份。围绕本课题的研究内容，培养博士研究生3名，硕士研究生4名。