Digital Twins and AI for manufacturers

Manufacturing AI Business Operations Digital Transformation AI Data Analytics AI Use Case Industry 4.0

Published:
Est. reading time: 6 minutes
Author: Solange Borrego

Digital Twins are virtual replicas of real-world systems enabling low-cost modelling of the factory floor to help optimise processes. Combined with AI the Digital Twin can support improved forecasting, dynamic optimisation, and more.

Solange Borrego

Quality Engineer with 5 years of experience in manufacturing, pivoting to Data Science and AI.

More

Manufacturers are increasingly adopting digital twins and artificial intelligence in order to be more competitive. Digital twins provide a 3D digital representation of the real-world physical assets, such as machinery, vehicles or buildings. At first glance this may not seem too different from what companies have been doing for years with CAD drawings, but there are great advantages that come with using these models: they can be used to analyse everything from machine performance to resource allocation; they allow engineers and designers to collaborate on creating new products without having to meet face-to-face; and they help companies make better decisions about where capital investments should go.

What is a Digital Twin and how does it work for manufacturers?

A Digital Twin is, simply put, a virtual and exact replica of a physical process, device or system that helps an organisation make model-driven decisions. The Digital Twin is a combination of models and simulation with actual, real information, like big data or sensors.

According to MarketsandMarkets, there will be a 38% growth in the Digital Twins global market until 2023, and manufacturing is not an exception. For example, there could be a Digital Twin of the entire shop-floor, or of a specific machine or component, and this means that it’s possible to, using just a computer, simulate certain output for any of these specific parts, or the entire plant, and obtain good results by leveraging real-time and historical data.

Digital Twins is not a new concept, being used by engineers in designing processes and prototypes, to getting the correct product specifications and the materials to be used. They are also important to validate issues that may arise due to regulation, quality, and durability, and enable the organisation to model, understand and choose the best approach for the physical product or process.

Because of the rapid evolution of previous technologies, as well as new ones appearing more and more often, there is a new focus on Digital Twins, and the gains that could come to the manufacturers who implement them in their processes or shop-floor. Manufacturers can use Digital Twins to make a virtual replica of their shop-floor and products, reducing time and cost associated with physical testing (which can involve production installation, assembly, downtime, etc.).

Since sensor data and other types of production/product measurements are more and more common, it makes sense to use this information for better Digital Twins, balancing real-world data with theoretical models. However, at this stage, Digital Twins by themselves do not allow for deep predictive analysis – this is where AI comes into play.

How can AI support Digital Twins and the manufacturing processes?

The newest application of Digital Twins leverages neural networks and machine learning in AI that use production data, such as the ones given by sensors, to obtain insights about the process without direct testing in the production shop-floor, which also means it won’t be necessary to rely only on theoretical models. As such, there is a strong relationship between AI and Digital Twins, since the latter’s plethora of data can feed and train the AI models in order for them to make accurate predictions.

For instance, a neural network could find nonlinear relationships between unconventional data types, identifying new correlations between data sets that were not thought of only by applying the theoretical models, since a neural network does not distinguish between production variables, such as temperature or pressure, and learns solely based on the data that is given to it.

Why are Digital Twins and AI important now?

There are many improvements that go hand-in-hand with the relevance of Digital Twins and AI in manufacturing right now:

1. The facilitated collection of production data

There are many options on the market for IoT sensors that are getting cheaper and appearing in many different types, which means it’s cheaper than ever to capture production data from the physical twin.

2. AI human-computer solutions are improving

Workers are starting to use augmented reality, virtual assistants, and chatbots, making it easier to engage with Digital Twins and AI.

3. Using analytics can get valuable insights

The ever-growing cloud services and new knowledge on machine learning are improving the insights that can be extracted from modelling and simulating based on production data.

The benefits of using Digital Twins and AI in manufacturing

One of the major benefits of using Digital Twins and AI in manufacturing is improved uptime, since it’s possible to better predict future failure and to maintain the equipment running smoothly with the forecast from the Digital Twins and AI. There are also improvements related to planning and design processes, which lead to major cost reduction, since it’s only necessary to simulate a given scenario using Digital Twins and AI.

Another area with major improvements is, of course, maintenance. A Digital Twin leveraging AI can predict when a certain equipment will fail, allowing to schedule predictive maintenances that are not only getting input from OEM manuals. As well as the reduction in downtime, this can reduce maintenance costs substantially.

There are also other areas that can have new approaches, such as staff training; it’s possible to train virtual workers that are in high-risk functions using the Digital Twin, similar to how it’s done with pilots via flight simulators. Since these technologies also improve the work on the shop-floor, it frees up people who are very knowledgeable from less-valued work to continue improving the plant to streamline further processes.

Challenges that may arise from implementing these technologies

Since Digital Twins are virtual replicas of an actual process or product, it’s not always possible to get the perfect production data, especially with chemical and biological reactions which have variables that are difficult or costly to measure in real-time. This means that it’s necessary to look at by-products or different measurements to get proxy-data (e.g., from light or heat) to get some data that can be used in Digital Twins.

Even when this data is available, it’s also important to verify the quality of this production data. If there are major gaps or unavailability, it may be difficult to test models with it. It’s also very important to double check the outputs from machine learning models with the actual physical process to ensure the predicted models make sense in the real world.

Examples of successful companies who have implemented these technologies

General Electric has used Digital Twin tech across different sectors to save $1.5 billion worldwide. A heavy industry manufacturing plant identified a problem with their CNC twin-spindle lathe, saving nearly $100 thousand.

Chevron has also used this technology in their plants to reduce issues in Supply Chain and to monitor equipment in real-time.

Big firms such as IBM, Siemens and Microsoft are also developing Digital Twins solutions.

Further reading