Hybrid digital twins: Incorporating AI to advanced monitoring and optimization of energy recovery systems

本项目主要研究受污染的HEXs网络的动态行为。传统的HEXs设计是隔离和稳定运行的，但结垢是一个动态过程；它随着时间的推移而积累，逐渐降低单个单元和网络的性能，并在流程系统级别上产生一组复杂的交互。HEXs的运行方式对当地条件（温度、流动状态或成分）非常敏感，在实践中，这些条件经常偏离设计目标（例如启动、关闭、清洁、配方变化、旁路、远离网络的装置提供冷/热流）。这使得在传统设计方法中被忽视的历史效应对理解实际能量回收系统在服役期间的演变至关重要。在这个项目中，我们将研究HEX网络中的关键性能指标（例如污垢阻力，导热性，速率，厚度）如何响应实际系统运行期间的扰动（例如传热流体的性质，操作程序，变化或复合污垢机制）。我们将使用先进的统计（即人工智能/机器学习）来严格表征现有数字孪生的不确定性，并开发和训练新的混合模型来监测效率，预测其演变（即新的沉积/去除模型，新的模型识别和参数估计工具），并可能以最优控制策略实时响应。工业正在缓慢地从过时的（稳定的）模式过渡到先进的（动态的）数字双胞胎，以最大限度地提高大规模能源回收系统的回报。该项目将通过对HEX网络的智能监控，引领创建准确、响应更快的数字孪生模型，并减少数万吨二氧化碳当量。

The project focuses on studying the dynamic behavior of HEXs networks undergoing fouling. HEXs are traditionally designed in isolation and steady operation but fouling is a dynamic process; it builds up over time, gradually reducing the performance of individual units and networks, and causing a complex set of interactions at the process system level. The way HEXs foul is very sensitive to local conditions (temperature, flow regime or composition), which in practice deviate often, and substantially, from a design target (e.g. start-ups, shutdown, cleaning, formulation changes, bypasses, units far from the network delivering cold/hot streams). This makes history effects, neglected in a traditional design approach, of a critical importance to understand how real energy recovery systems evolve during service. In this project we will investigate how key performance indicators in a HEX network (e.g. fouling resistance, thermal conductivity, rate, thickness) respond to perturbations during the operation of real systems (e.g. properties of heat transfer fluids, operational procedures, varying or compounded fouling mechanisms). We will use advanced statistics (i.e. AI/machine learning) to rigorously characterize the uncertainty of existing digital twins and develop and train new hybrid models to monitor efficiency, predict its evolution (i.e. new deposition/removal models, new model identification and parameter estimation tools) and potentially, respond in real-time with an optimal control strategy. Industry is slowly transitioning from obsolete (steady) models into advanced (dynamic) digital twins to maximize the return of large-scale energy recovery systems. This project will lead the way in creating accurate, more responsive digital twins and reducing tens of thousands of tons of CO2e via intelligent monitoring of HEX networks.