- By Preston Johnson
- September 23, 2024
- ISA
- Feature
Summary
A digital twin represents a physical asset and its function; it contains intelligence to evaluate static and real-time data.
Chemical refining plants are asset intensive and employ a wide range of rotating and static equipment. Chemical refining plants make biofuels, alcohols, automotive coatings and chemicals used downstream in other chemical processes. Centrifugal pumps are used to move fluids, including water, used for steam, cooling water for heat exchangers, mineral oils as feed stock for refining and resulting product. Pumps are an important asset for refineries.
Enter: Digital twin
A digital twin of an important asset helps measure and evaluate asset health and performance parameters. With improved measurements (and monitoring), it becomes possible to manage and improve the health and performance of the asset. Financial benefits accrue in the form of lower maintenance costs, increased availability of the asset for production, lower power consumption and lower use of natural resources.
A digital twin represents a physical asset and its function. The digital twin contains intelligence to evaluate static and real-time data. Inherently, the digital twin acts like a model, evaluating data to provide actionable information that, when acted upon, maintains the asset’s health and performance.
In practice, companies often start the digital journey with equipment and process reliability in mind. A typical initial task is to create an equipment hierarchy down to replaceable or repairable components using ISO 14224, ISA 95, or similar hierarchical maps. The hierarchical maps organize equipment around units, systems, subsystems and equipment (Figure 1).
Equipment is then categorized into groups or classes and ranked in criticality or importance to plant performance. In this case, centrifugal pumps are chosen due to their wide use in distillation column units, and significant importance in the boiler unit that creates steam for plant heat exchangers.
Digital twin elements
Digital twin data models contain or connect to engineering data, physics models, real-time data, historical trend data and event data including maintenance activities. Engineering data and physics models describe expected performance and expected behavior of the asset and nearby process. Real time and historical trend data describe the process and asset’s current state and trend history leading to the current state. These inputs to the digital twin are summarized in Figure 2.
The digital twin connects to its data sources using database queries to connect to engineering data and maintenance data and using secure standards such as OPC-UA or MQTT to connect to streaming process controls data and data historian tag connections.
Engineering data including physical properties and physics are the foundation for physics-based models. The physics models also yield likely failure modes for the equipment. In a pump, failure modes can be hydraulic, mechanical, lubrication, or electrical in nature. The physics models should be detailed enough to yield expected measurements under intended operating conditions. For example, vibration, motor current, temperature, flow, pressure and lubrication sensors can be used to monitor for abnormal values and generate an alert. Monitoring sensors against alert levels is the basis of condition monitoring and fault diagnostics in equipment. With the fault type and severity of the alert, it is possible to roughly estimate the time to failure. The physics model becomes a predictor of future performance based on the presence and severity of a defect detected by a sensor. We can use a prevention-failureperceive failure (P-F) curve to visualize the degradation of the pump.
The P-F curve, also known as the potential failure to functional failure curve, is a way of visualizing how a mechanical system degrades over time. The physics models help to identify leading indicator sensors that provide early detection of defects with ample time to plan and schedule mitigating maintenance activities (Figure 3). This helps lower maintenance costs and improve equipment uptime and availability. Lagging indicators help confirm diagnosis made from leading indicators.
The P-F curve is typically nonlinear, so predicting how soon maintenance action should be taken is problematic. Very detailed physics models likely can make accurate predictions, yet the complexity of the model and computational resources needed can make the model hard to maintain and deploy.
When we combine historical data and maintenance event data, it is possible to include historical patterns of data-driven models such as machine learning (ML) used for anomaly detection and pattern recognition. Data driven models with advanced statistics are good at predicting future data, or the future health of the centrifugal pump. However, data-driven models often require a large amount of data to find patterns. Data-driven models are fast to execute once trained yet can be hard to generalize.
Combining methods
The benefits of both data-driven and physics-based methods can be leveraged by using hybrid approaches. Furthermore, the drawbacks can be reduced or mitigated, making hybrid approaches ideal for maximizing the value of analytics technologies. Some of the ways physics and ML can be combined to produce a hybrid approach include the following, also shown in Figure 4.
- Physics-informed ML methods, e.g., PINNs, evolutionary algorithms.
- Estimation of engineering parameters from data.
- Reduced-order modeling (ROM).
- Virtual sensors.
- Uncertainty quantification and propagation (UQ).
Together, the data-driven and physics-based models operating within the digital twin improve the accuracy of predicting the remaining useful life (ROL) of the equipment in its current state without mitigating maintenance activities. The expectation of functional failure can be visualized as exemplified in Figure 5.
Methods come together
Digital twin definitions vary depending on the application and expected function. In this article, a digital twin represents a physical asset and its function. The digital twin contains intelligence to evaluate static and real-time data. Inherently, the digital twin acts like a model, evaluating data to provide actionable information that when acted upon maintains the asset’s health and performance.
In this article, a combination of engineering data, physics based models, maintenance records, process data, condition monitoring data and data-driven models come together to provide not only actionable information but also the timeframe for which mitigating action should be performed.
About The Author
Preston Johnson, senior delivery manager at Novity, provides solutions for Industrial IoT with a focus on condition monitoring and predictive maintenance. Preston builds on 28 years in industrial instrumentation and machine condition monitoring systems at National Instruments, four years at Allied Reliability building predictive maintenance programs with digital technologies, and four years at CB Technologies bringing IT to predictive maintenance. He is an active member of the International Society of Automation's Smart Manufacturing and IIoT Division (also known as SMIIoT), the newest and fastest-growing division of ISA aimed at helping members grow professionally and technically.
Preston draws from his broad functional background in people and project management, technical and domain expertise in equipment monitoring applications, and business development to create and implement comprehensive business strategies for new product development, market introduction and growth. He works with business development and operations teams to deploy condition monitoring and predictive maintenance systems and services, reliability solutions, training and hardware/software systems that improve machinery uptime, reliability, sustainability and ultimately, production capacity. Preston has held roles in Applications Engineering, Field Sales, Sales Management, Product Management, Business Development, and Field Implementation and Training.
At Novity, Preston led the delivery of TruPrognostics solutions including hardware and software for condition monitoring and predictive maintenance. Preston resides and works in Austin, Texas. In his spare time, he enjoys his family, the outdoors and music.
Did you enjoy this great article?
Check out our free e-newsletters to read more great articles..
Subscribe