EN

Case Study - Linde

Published on March 17, 2021

Case Study

Creating efficiencies with artificial intelligence at Linde

“This project is a great example of how collaboration speeds up value generation. We get the machine learning support from appliedAI, the technical process knowledge from different departments of Linde Engineering and the operational knowledge from Linde Gas. Each party brings their best knowhow.” - Dr. Dexin Luo, Head of Artificial Intelligence Solutions & Technologies at Linde

.

In 2017, Linde, the world’s market leader in the field of industrial gases, joined the appliedAI initiative as a partner in order to accelerate the adoption of AI. Linde is currently working toward implementing AI solutions throughout its business units to increase productivity and competitiveness across its value chains and to master the journey to AI maturity. Linde and appliedAI have conducted various cooperative projects since Linde became an appliedAI partner. Some of these projects include ecosystem activities like AI delegation trips to China and Canada, as well as educational training workshops for employees. Additionally, Linde has also utilized appliedAI’s engineering expertise in projects implementing AI applications.

So, how does Linde apply AI to improve the efficiency of its industrial plants, and how does appliedAI support the optimization of plant operations? The following case study provides insight into Linde’s collaborative activities with appliedAI in the area of artificial intelligence for plant operations.



Timeline - Linde and appliedAI partnership

Timeline: Linde and appliedAI partnership
Source: appliedAI


1) Initial Situation and Problem Statement

Since its first AI trial project in 2016, Linde has been actively developing and implementing AI solutions across its value chain, from sales and supply chain to operations. Linde also engages in longer-term AI research topics, such as optimizing the efficiency of their thousand-plus production plants around the globe.

Linde has already operationalized a portfolio of AI projects while focusing on professionalizing AI product strategy and product management. The company has a central AI-Team that develops innovative AI solutions and provides AI trainings to the company as well as digital teams embedded in business units with AI development capabilities. Linde has chosen to leverage AI to gain a competitive advantage by increasing efficiency.

To improve its AI maturity, Linde identified the following measures, such as educating and enabling employees as well as the optimization of plant operations, that need to be taken:
.

1. Educate employees on AI

The first challenge Linde faced was to educate their domain experts and operational employees on artificial intelligence. There has been a lack of experts who understand AI technology and plant operations, which is why there is currently a need for domain experts who are able to lay out a concrete plan regarding where AI could be used to help improve plant efficiency. Raising awareness about the AI activities within the company was important to build an understanding of AI and help speed up the AI implementation process.
.

2. Optimize plant operations and automate control systems

Linde utilizes advanced plant operation technologies, and a majority of Linde’s production plants are remotely controlled by Remote Operating Centers (ROCs). A joint team of Linde’s domain experts and appliedAI engineers identified additional areas for improvement in plant control to increase plant efficiency. In response to business requests, Linde decided to take measures to push the limits of optimizing plants even further with the help of artificial intelligence.

The plant control system consists of low-level controls and high-level controls. The task of the low-level controls is to guarantee the correct settings for individual valves as well as actuators. This task is managed by a Distributed Control System (DCS). The high-level control mechanisms, on the other hand, are tasked with controlling the plant as a whole. The controllers need to be tuned regularly, and from time to time the plant load needs to be changed, a task which still needs to be conducted manually by employees.

Setting up new plant components manually as well as maintaining and returning control algorithms of the plants on a regular basis is an especially taxing pain point, because it requires several weeks of work. Although the tasks of the high-level controls still need to be conducted manually by employees to ensure that the plants work in unison and optimally, these tasks could be handled by AI, which would work more efficiently and precisely.

Since the control algorithms within ROCs are not self-learning, they require troves of data to predict a plant’s behavior. This data is also crucial for applying a specific strategy to control the valves to make sure that the process values are close to target values, in order to gain optimal operational results. Although this procedure works fine, it can be further improved to become more efficient and sustainable. AI has the potential to provide an adaptive system that learns plant behavior and reacts even better in dynamic situations. By using an AI plant control system for forecasting real plant behavior and directly reacting to it, Linde would be able to save significant amounts of energy and would require less work while gaining better and more energy-efficient results.

2) Approach and Methodology

After identifying pain points and areas needing improvement, appliedAI and Linde started taking several important measures. One of the first actions that appliedAI took to solve the pain points mentioned above was to help Linde to educate its employees in the area of AI through a number of workshops and training opportunities. Additionally, appliedAI’s team of highly experienced engineers and specialists worked closely with Linde to help develop an AI-controlled plant using reinforcement learning. To maintain the momentum of these initiatives, appliedAI and Linde created an AI-driven “Lego AI wonderland”, which communicates the AI vision and raises awareness of the AI activities along Linde’s business units to the public and employees. The following sections offer an insight into the mentioned cooperative projects conducted to help Linde reach its goals.


2.1 Workshops and AI tools for Linde

To enable employees to identify and initiate AI use cases, employees from Linde participated in several AI workshops as well as in Ecosystem meetups from appliedAI. These courses by appliedAI are designed to support companies in educating their employees and enabling organizations to master the journey to AI maturity and to accelerate the adoption of AI. Linde employees participated in AI introductory courses, AI strategy courses and project management workshops.

.

Participants during an appliedAI workshop

aAI people working in workshop
Source: appliedAI


So far, Linde employees have participated in AI Introduction workshops, which took place throughout their partnership with appliedAI. The AI Intro workshop provided an intensive and hands-on introduction to the domain of artificial intelligence from multiple perspectives. The goal of this workshop was to establish among Linde’s employees a common understanding of what AI is about. It also enabled participants to determine where AI technology could be applied in Linde’s daily work and equipping them with basic knowledge in the area of AI. By attending the workshop, the employees were able to examine AI from a technical perspective and were better equipped to have meaningful conversations about the application of AI at Linde.

In addition, Linde employees attended AI Strategy workshops, which aim to enable companies to develop their AI strategy. The Strategy workshops demonstrated the opportunities and challenges associated with the adoption of AI technologies and Linde employees got to know the elements of comprehensive AI strategies designed to help Linde concretize and further expand their AI vision. They were also equipped with relevant prerequisites for the successful application of AI in their company and learned how to evaluate the current AI-related maturity of companies.

Participation in AI Project Management workshops helped Linde’s employees to gain a deeper understanding of how to manage AI projects. During the workshops, the participants acquired a comprehensive overview of the complete life cycle of AI projects and the common obstacles during each project phase. The workshops helped empower employees to apply best-practice methods and process models that help manage AI projects, thus equipping them with the knowledge to:

  • identify the key role profiles for various project phases

  • avoid typical pitfalls during the implementation of AI projects

  • discern the diverse software and hardware requirements for different AI project types

Linde also participated in appliedAI’s AI delegation to China in 2018 as well as its delegation to Canada and Singapore in 2019, which aimed to build up connections to AI companies and startups within the Canadian AI ecosystem and to represent the AI ecosystem from a German perspective.

With help of the knowledge gained in the workshops, and in combination with their internal expertise, the employees were able to identify and create high-value AI use cases for Linde. While doing so, Linde used the “AI Use Case Cards” developed by appliedAI’s strategy team. The cards are meant to help appliedAI partners to run use-case workshops and create AI use cases on their own. They helped Linde to conduct its own AI use-case ideation.

.

appliedAI Use Case Cards

use case cards
Source: appliedAI


.


2.2 Linde Connected AI Wonderland

In addition, a miniature version of the complex AI activities at Linde was built with Lego® by appliedAI to raise awareness of the AI vision and activities at Linde within the company and among the public. This model, which demonstrates Linde’s value chain and the AI activity along the processes by making Linde’s digital vision tangible, is on public display in Linde’s Munich office. This miniaturized version of a Linde plant and the logistics to key customers was built using a simple IoT setup (with Raspberry Pi and Arduino) and shows some of the “in production” use cases of AI at Linde, such as a camera system, monitoring gas cylinders and a prediction system that forecasts the demand and the needed production level of the plant at any point in time.

Linde Connected AI Wonderland by appliedAI
Source: appliedAI


2.3 AIPlantControl

In order to master the journey to AI transformation, Linde also partnered with appliedAI to optimize plant operations by implementing AI into the control systems of industrial plants.

Throughout the project, Linde’s corporate digital team, “Linde.Digital”, has spared no effort to reach its goal and is applying AI in collaboration with a team of experienced AI engineers from appliedAI. What is special about this proof of concept is that the team directly created the control system in a real-world environment by working with a real production plant instead of a simulator, in order to prove the potential value of the technology.

.

Linde and appliedAI developing AIPlantControl together

AIPlantControl Employees working
Source: appliedAI/Linde (Christian Oehse [foreground] and Nicolas Blum [background])

During the project, the team developed a reinforcement learning algorithm to build an AIPlantControl system, which supports the operational side of Linde’s business in order to optimize plant operations and gain efficiency by using intelligent AI-driven controllers.

To realize the prototype of AIPlantControl, the team applied reinforcement learning to the existing control algorithms to use AI for setting the plants’ operating parameters for valves and compressor power levels in order to gain efficient operation by automatic intelligent controls. As a result, the system learns the behavior of Linde’s processing plants and recognizes how to set the dials in order to reach the highest efficiency possible, while lowering maintenance effort.

In addition to taking this measure toward increased work efficiency, the team trained the control system to imitate and adapt the physical behavior of Linde’s plants in order to reach higher control precision, which leads to maximum energy efficiency in times of rising usage of renewable energy. This has the effect of lowering the plants’ energy consumption during stable operations, because automatic fine-tuning by the AI controller adjusts the plants to their optimal setting.

To give a more precise example, in the AIPlantControl project, the team applied the purely data-driven, self-learning process controller to the first air separation plant (ASU) from Linde. The attached graphic shows the huge impact that reinforcement learning can have on the control quality and efficiency of processing plants and demonstrates the optimization of Linde’s plant operation and the increased control precision gained by the AIPlantControl system. It shows the differences in control quality between using AI for optimization (right-hand side) and using an air separation plant without applied artificial intelligence. (cf. Blum, B./Zapp, G./Oehse, C./Dr. Rehfeldt, S./Prof. Dr. Klein, H.: Untersuchung eines rein datengetriebenen, selbstlernenden Prozessreglers im Produktivbetrieb am Beispiel einer Luftzerlegungsanlage. ed. appliedAI / Linde GmbH / Technische Universität München, 2020)


Control quality comparison during a load change

Linde optimization 1
Linde optimization 2
Source: appliedAI/Linde GmbH/Technische Universität München (2020)


3) Outcome and next steps

Looking back, we can see that the joint appliedAI-Linde projects are already bearing fruit. Educating employees in the area of AI and equipping them with AI tools has supported Linde in expanding its AI team and empowering them to further implement AI in its business units.

By applying AI and using reinforcement learning, the team mastered the development of Linde’s AIPlantControl system, which is able to help optimize plant operations.

Through the implementation of an AI-based control system, the effort required for tuning the controller was decreased, while the control precision of the system was further improved. This enhancement could be achieved only by exploiting AI’s self-learning capabilities. The improved control behavior results in a more optimized plant behavior, leading to savings in the operational and energy costs of the plant.

Given these outstanding results from the first self-learning processing plant and the high potential to gain efficiency across the board, Linde is planning to roll out another three AI-based, self-learning plants in Central Europe in the future. There are also plans to expand AI implementation in Linde’s business even further in order to become even more mature in the area of AI and benefit even further from the potential of AI for optimizing plant operations to stay competitive.

In addition, the team is planning to use reinforcement learning in the future to teach the AI control system to adapt the production processes of Linde’s processing plants and to adapt to the natural fluctuations in supply on the energy markets. The AIPlantControl in this case acts as an AI-based learning controller, which compensates fluctuations in the grid by using more electricity during times when it is affordable and in great supply and using less when there is high demand and is more costly. By doing so, Linde will gain even more efficiency and further accelerate its journey to AI maturity.


More information about all general AI activities from Linde along the value chain can be found here.

Additional relevant content for you

2021 German AI Startup Landscape

Additional relevant content for you

Case Study: Village Data Analytics with TFE Energy