Decision support systems based on heterogeneous data driven models for a safe and optimal operation of industrial process systems

包括化工、石油和天然气工业、制药、采矿、金属和纸浆造纸在内的过程工业在加拿大经济中发挥着重要作用。今天，操作员只在预期的情况下参与，因为在正常操作条件下，过程是自主的。这种脱节被称为“自动化的讽刺”。工业过程系统需要更高水平的自主性来解决这一悖论。自主机器人和自动驾驶汽车等自主操作的最新进展开辟了全新的方法，可以处理很少或非结构化的输入信息，并在高度不确定的环境中产生大量反应。这反过来又极大地提高了公众对自主系统的接受度。过程系统的完全自主操作是一个复杂的问题，因为与自动驾驶不同，它需要与多个人类角色(控制室操作、现场操作、过程优化和生产调度)相对应的多个自主关键功能，而不是单一的汽车司机。一些必要的支撑技术(工业物联网、大数据、高性能计算和新型传感技术)最近才出现。横河最近发布的一项调查发现，三分之二的流程工业公司预计到2030年将完全自主运营。在此背景下，该研究计划的长期目标是创建虚拟控制室助理，为操作员提供工厂异常的早期检测、准确诊断和纠正措施建议。为此，与过程模型可解释性和在线可移植性相关的基本问题将在这项发现赠款中得到解决，具体解决方案将针对真实世界的过程系统进行量身定做，并通过后续的联盟赠款在工业试验中进行演示。通过未来五年的四个博士和三个硕士项目，这项研究计划将通过与行业合作伙伴的密切合作来实现以下短期目标，从而产生高度影响：(1)从第一原则模型中提取在线可移植和可解释的代理模型(影响：提高执行速度)。(2)预测流程异常事件(影响：增加稳健性)(3)将名义和异常事件流程模型集成到一个单一框架中(影响：促进工业吸收)。前述调查显示，54%的受访者预计未来3年将增加对新冠肺炎的自主投资。这项研究计划的结果将有助于开发在线运营商决策支持系统。使用这些系统将导致(1)通过提高效率和工厂可用性来降低生产成本，(2)通过先进的分析来提高最终产品质量，(3)通过远程操作来实现更安全的工作环境。

Process industries including chemicals, oil and gas industries, pharmaceutics, mining, metals and pulp and paper play an important role in the Canadian economy. Today, operators are involved only in expectational circumstances because processes are autonomous under normal operating conditions. This disconnection is termed "the irony of automation". A higher level of autonomy of industrial process systems is needed to resolve this paradox. Recent advances in autonomous operations such as autonomous robots and self-driving cars have opened up completely new ways to deal with little or unstructured input information and to generate a large range of reactions in highly uncertain environments. This has in turn greatly increased public acceptance of autonomous systems. Fully autonomous operation of process systems is a complex problem because unlike autonomous driving, it requires multiple autonomous key features corresponding to multiple human roles (control room operations, field operation, process optimization and production scheduling) instead of a single one, the car driver. Some of the necessary underpinning technologies (Industrial Internet of Things, Big data, High-performance computing and novel sensing technologies) have only recently emerged. A recent survey released by Yokogawa finds two-thirds of process industry companies are anticipating fully autonomous operations by 2030. Within this context, the long-term objective of the research program is to create virtual control room assistants providing operators with early detection of plant anomalies, accurate diagnosis and suggestions of corrective actions. For this purpose, fundamental problems related to process models interpretability and online portability will be solved within this Discovery Grant with specific solutions tailored to real-world process systems and demonstrated in industrial trials via subsequent Alliance grants. Through four PhD and three MSc projects over the next five years, this research program will achieve a high impact by working closely with industrial partners to achieve the following short-term objectives: (1) Extract online portable and interpretable surrogate models from first principles models (impact: increase execution speed). (2) Predict process abnormal episodes (impact: increase robustness) (3) Integrate nominal and abnormal episodes process models into a single framework (impact: facilitate industrial uptake). The aforementioned survey indicates that 54% of respondents expect to increase their autonomy investment over the next 3 years as a direct result of COVID-19. The outcome of this research program will be instrumental in the development of online operator decision support systems. Using these systems will lead to (1) Reduced production costs through improved efficiency and higher plant availability (2) Improved end-product quality through advanced analytics (3) Safer working environments through remote operations.