This article is a transcript from a presentation our CEO, Markus Guerster gave regarding Generative AI in manufacturing, more specifically related to membrane filtration equipment. You can see the full video for the discussion below.
Hello everyone, it’s a real pleasure to be here today, especially following some truly insightful talks. My name is Markus, I’m the founder and CEO of Mount Blanc AI, and I will be discussing today about Generative AI for membrane filtration equipment and beyond.
Rather than delving into the deep technical details of membranes, my focus will be on AI and its potential implications for this industry. Those who attended Steve Beckman’s talk yesterday will find the perfect segue into my discussion today.
Understanding Generative AI
Before diving into the core of my presentation, let me ask you a few questions. Who has heard of ChatGPT? I hope everyone is raising their hands now. Now, who has used it at least once? And who uses it regularly – say, once per week or even once per day?
ChatGPT, a manifestation of Generative AI, has become a commonly used term in such a short time. If we look at the statistics and compare it to platforms like Spotify and Instagram, it’s impressive. Spotify took around half a year to get one million users, Instagram was twice as fast, and ChatGPT reached one million users in just five days.
But Generative AI is not just about ChatGPT. It falls under the umbrella of narrow AI with limited memory and, as the name suggests, it generates content such as text, images, music, voice, videos, and much more. However, it’s crucial to note that the AI revolution doesn’t mean replacing humans. On the contrary, I view AI as a tool to empower humans to do more.
AI as a Tool for Empowerment
Consider the relationship between a gardener and his tools. The invention of tools like a shovel didn’t replace the gardener. Instead, it made his work more efficient, enabling him to achieve results that would be impossible with bare hands. AI is very similar to this. It is a tool that can handle large amounts of data, automate repetitive tasks, and even learn from experience.
Just as a shovel cannot decide where to dig, AI relies on human guidance. Humans set the goals, provide the data, interpret the results, and make the final decisions. AI may change the nature of some jobs, just as the shovel changed the nature of gardening, but it doesn’t replace the need for human judgment, creativity, understanding, and empathy. Like a gardening tool, AI is opening up new possibilities for innovation and growth.
The Future of AI
AI is here to stay. It’s not a new idea, having started in the 1960s, but we are now in a new wave of AI that’s been growing for about 15 years. Many factors are propelling this advancement: availability of vast amounts of data, huge computational resources, advancements in machine learning, and global investment from governments, public sector, universities, and the private sector.
Generative AI in the Industrial Context
A recent study by IoT Analytics that analyzed the topics CEOs discuss shows that AI is on top of the minds of executives. Generative AI, chat GPT, conversational AI, and Industry 4.0 are topics gaining increasing importance.
But you might be wondering, what can AI actually do for you? Well, the solution to leveraging the power of AI is the marriage between large public generative AI models and industrial data. This union opens up several potential applications in four major categories: Sales and Marketing, Product Development, Supply Chain Optimization, and Operations.
These areas provide a good starting point if you are considering experimenting with AI in your organization. It’s clear that the era of Generative AI is upon us and it’s critical that we understand its potential and how to leverage it to our advantage.
Furthermore, generative AI can be employed for efficient troubleshooting, functioning as an assistant to help us find the root cause of a problem. Once the issue is identified, it can be addressed to prevent future failures.
To put these ideas into perspective, consider a case study involving a dairy farm operation. This farm, handling 100 to 2 million liters per day, utilizes membrane filtration equipment to separate casein and whey proteins. The farm faced business challenges concerning resource allocation and continuous validation of its Cleaning-In-Place (CIP) process.
What we did was connect to their S7 PLCs and stream all of the 200+ data points into our AI algorithms, processing over a billion data points. After running these algorithms, we visualized and aggregated the results to inform the operators’ decisions. The outcomes were promising—we validated over a thousand cleaning cycles, identified areas where resources could be saved, and optimized batch durations.
So, you may wonder, how can you apply AI in your organization? If you haven’t considered it yet, I encourage you to give it some thought. I’ve divided the key considerations into three categories:
- Generate Value for the Business: AI is an innovative, cutting-edge technology. You might not be able to write a business plan before you start experimenting with it. The good news is that you can start small, with low costs, and iterate fast. By this approach, AI’s value will spread within your organization from one low-hanging fruit to the next.
- Acquire Knowledge: While GPT is easy to use, combining it with your industrial data can be complex. The technology is changing on a daily basis. Thus, bringing in missing knowledge, attending training sessions, and having vendors compete for your business can be beneficial.
- Ensure Security, Privacy, and Trust: It’s important to make sure everyone understands the do’s and don’ts. Picking the right vendor is also crucial, as different vendors have varying approaches to privacy and security.
Lastly, don’t stop experimenting. The cost of doing so is very low, and you might be surprised at the innovative solutions your team can come up with when given the freedom and the right tools.
As we near the end of this discussion, I want to leave you with my personal vision—or dream, if you will—of the future of AI in an industrial context. When I finished my PhD and before I began working in industry, I had a rather naive assumption. I assumed that data would just be there, in excellent quality and well organized, with 90% of the time spent thinking about the algorithm itself.
But the reality of working in industry quickly disabused me of that notion. As you probably already know, data is often a complete mess. You have your Manufacturing Execution Systems (MES), your Enterprise Resource Planning (ERP) systems, and a host of other boxes filled with data. It’s a real issue if you’re trying to solve a problem like a customer complaint and need to backtrace information. You find yourself calling the ERP person, noting down information, calling another person to look into a different system, and days or weeks pass by. Often, the information you need is just not available.
This situation is reminiscent of an episode of Asterix and Obelix where they’re trying to get Form A38. They go from one place to another, trying to get signatures, new forms—it’s essentially the same as the ERP, MES, PLC, CMMS story. These are silos of disconnected systems.
My vision, then, is to have a unified lens. A system where you simply pick up your phone and ask a question—much like I did with GPT earlier. You ask, “How’s my membrane machine doing today? Are there any issues?” And instead of receiving a reply of “I don’t know,” you get a useful response because the system has looked up information across all databases.
Imagine this scenario: “Yes, your machine is doing well today. It just finished a 19-hour batch. It’s currently in cleaning mode, and it will take a couple of minutes longer than usual. Nothing out of the ordinary. Would you like to be notified if something goes wrong?” Now, that would be powerful!
Going back to the gardener analogy, the gardener is still a human being, interacting with the phone, asking questions, and interpreting results. However, the AI is the tool, doing the heavy lifting in the background. It fetches the data, looks it up across all systems, combines it, aggregates it, and presents it to you in a way that makes sense.
I’ll conclude my presentation at this point, and I’m happy to answer any questions. Feel free to scan the QR code to connect with me on LinkedIn. I’d be more than happy to continue this conversation or send you a copy of this slide deck.
Q: Do you have an idea of creating a membrane GPT that combines trainings from suppliers, OEMs, and chemical suppliers to help build skills for employees?
Markus: While nobody is actively working on this specific concept, we have experimented with the idea. By ingesting relevant documents into a private database structure, a membrane GPT could teach individuals about membrane filtration machines, incorporating knowledge from suppliers, OEMs, and data collected. It’s an intriguing proposition that can enhance employee skills and foundational knowledge.
Q: How can we define parameters and gather data trends to answer questions about the performance of a membrane system? Are you collaborating with industry leaders in data collection and analysis?
Markus: Collaboration is crucial in defining parameters and obtaining data trends for accurate analysis. Domain experts, such as process engineers, play a vital role in establishing key parameters for a healthy machine. As for data collection, it requires teamwork between AI systems and industry leaders to ensure accurate evaluations and insights. By combining machine learning capabilities with expert knowledge, we can effectively assess membrane system performance.
Q: How far along are we in creating self-learning AI systems that can teach us about specific industrial processes and provide insights?
Markus: We have made significant progress in developing self-learning AI systems, particularly in processing and analyzing textual data. However, achieving a comprehensive understanding of complex industrial processes and creating AI systems that can teach humans about them is an ongoing journey. Currently, we are at a level similar to that of a junior or senior process engineer, but we continue to advance and refine AI technologies to unlock their full potential.
Q: Is security the advantage of decentralization, and how can we ensure data safety in an AI-driven environment?
Markus: Decentralization does offer security advantages as it requires breaching multiple programs to access all the data. However, in centralized systems, data safety is paramount. Implementing robust encryption, access controls, and authentication mechanisms are essential. By working closely with cybersecurity experts and adopting industry best practices, we can safeguard data, prevent unauthorized access, and protect sensitive information.
Q: How can we prevent AI from generating false or incorrect recommendations that could lead to operational issues?
Markus: Preventing false or incorrect recommendations from AI systems is crucial for reliable decision-making. A feedback loop mechanism can be implemented, where AI-generated recommendations are continuously evaluated and cross-validated against known data or expert insights. Additionally, combining AI-generated recommendations with human expertise ensures more accurate and reliable outcomes.
Q: Are there limits to artificial intelligence hallucinating on processes, and how can we prevent it? How does decentralization impact security?
Markus: AI hallucinations and inaccurate recommendations are challenges that require attention. Techniques such as self-critique can be employed, where AI systems are questioned about their generated answers and prompted to reconsider. While AI won’t run an entire plant or facility autonomously, it can provide recommendations and insights. Decentralization, when coupled with strong security practices, offers protection by reducing the risk of unauthorized access and minimizing single points of failure.