The role of digital twins in energy and utilities management

Submitted by Victor de Souza on Wed, 02/07/2018 - 17:45
digital_twins

Industry 4.0 (Industrie 4.0) started as an initiative of the German government, specifically of the Ministry of Research and Education and of the Ministry of the Economy and Technology. This strategy is a new industrial revolution that uses the concepts of cyber-physical systems, Internet of Things and cloud computing. In this way industrial production can be self-organized, and therefore professionals, machines, plants, logistics, and products communicate and cooperate directly, optimizing not only industrial production but also the production chain as a whole. On the basis of these concepts, Smart Factories arise: highly integrated, intelligent, and efficient industrial systems.

See also infografic: From the first Industrial Revolution to Industry 4.0

Benefits and challenges in the creation of Digital Twins

One important component of intelligent industrial systems are digital models of equipment and subsystems: so-called digital twins), initially conceived for industry, are models that can be used to create simulations in equipment or other systems, generally with the objective of reducing operational costs and improving process efficiency. With the advance of technology and of the sophistication of the models, other characteristics can also now be modeled, such as product demand. For Industry 4.0, this type of model is vital for the digitization of plants, equipment and processes. Continuous utilization of models of this type can be extremely beneficial for production and for the energy efficiency of industrial machinery; benefits which, if applied well, are reflected in entire plants, providing a massive economization of energy that also contributes to the reduction in environmental impact. This type of modeling can bring various types of benefits, but its implementation is not trivial. There are several challenges in the creation of a digital twin.

Sophisticated models based on data are remarkably powerful, precise and accurate. For a digital twin to be viable an abundance of high-quality data is needed Today it is already possible to obtain data in sufficient scale to supply these models, in view of the fact that with the digitization of sensors the volume of data collected is always increasing. However, in many cases it is seen that the quality of the meters used may not be sufficient, or that a significant part of the recorded data is missing. Even if it is possible to use these data to train the models, this type of problem directly interferes with their results and analyses. The instrumentation is a very important aspect, as it is directly related to the quality of the data.

Even given the availability of data with quality and sufficient volume, the construction of a digital twin model is still not a simple procedure. The first step is the definition of the problem and the identification of which variables and characteristics of the equipment or process are relevant for that modeling. A common example is the modeling of a machine for the purpose of improving its energy efficiency. For this case it is necessary to define which machine is to be modeled and to identify the relevant static factors and variables of the system before proceeding to the next steps.

The modeling itself is a process that varies greatly based on the process and data. To create a digital twin a set of models can be used, or even a single model with greater complexity.

The biggest advantage of using sets of models is that you can divide a complex problem into several smaller problems, each handled by a subset of models, each having the most appropriate characteristics for solving each specific problem. In addition, it is possible to use less-complex models to find the solution of a complex problem, which is a good idea if a model needs to be explainable without sacrificing performance. Today there is much effort to make models more interpretable to better understand decisions recommended by intelligent systems. A common problem with this type of approach occurs when there is a need to group data by similar characteristics. In this way some groups may have few points. This type of situation is difficult to model, although there are techniques for handling these cases.

The use of a single model can be the solution to some of these problems. Using all the available data, a more complex model with multiple exits may have higher accuracy, in addition to modeling some problems better. Due to its high complexity, this type of model is more likely to present problems of overfitting and to have less interpretability at the same time as it manages to handle more difficult problems with better accuracy than would be obtained by simple models.

Another problem that frequently occurs in industrial applications is the lack of maintenance of the models. Like any model based on data, digital twins are subject to obsolescence. As changes occur in the process, it is difficult for the model to follow them without help. A few different approaches can be used to solve this problem. The most common strategy is to define a time interval, for example three months, and retrain the model with more recent data. Another possibility is to also model the degradation of the equipment over time.

Conclusion

Digital twins are tools that can bring exceptional benefits to industry. Like all tools, some prerequisites need to be met in order to obtain effective results. By following a structured methodology it is possible to create models that are effectively digital replicas of physical systems. To this end it is extremely important to have solid instrumentation and a modeling that is consistent with the system.

The Viridis platform is extensively based on the concepts of the Internet of Things, Big Data, and Industry 4.0, as it is capable of capturing, storing and processing large volumes of data on the consumption of energy and utilities of any operational unit, whether a specific machine or a production process as a whole. These data, combined with information about the context in which they were recorded (production plans, process variables, operational parameters, among others), makes possible, using the techniques of Machine Learning, the construction of predictive models that are combined to create a digital representation of that asset – a digital twin. The large integration capacity of the platform facilitates the capture of relevant data, while its modeling and visualization capabilities make the use, interpretation, and evolution of digital twins easy, resulting in significant value creation for our clients.


Data Analyst , Viridis

Data Analyst for Viridis, degree in electrical engineering from UFMG. He has a solid understanding of computational intelligence and statistics, acquired during his time at the Computational Intelligence Laboratory of the School of Engineering of UFMG.

Add new comment