AI on the shop floor: A game-changer to achieve economies of speed

Today, we are experiencing a shift in manufacturing. Companies are shifting their focus from achieving economies of scale to economies of speed to achieve competitive advantages. However, realizing this is becoming increasingly difficult. It is due to growing product varieties, process complexities, stochastic machine behaviors on the shop floor, and supply chain disruptions. Engineers should act to the challenges much faster than ever before. They have to analyze several process variables and factors to quickly and confidently make a decision. Here is where the unique capabilities of AI can tremendously help companies realize the desired level of economies of speed. Using AI, engineers can analyze millions of lines of shop floor data from hundreds of process variables to make faster, better, and confident decisions. For example, AI can help shop floor engineers quickly find the root causes of the problems and prescribe appropriate actions. It, in turn, reduce the response time and leads to more effective decisions and actions. AI can even predict the future factory dynamics based on several variables and prescribe concrete proactive actions to engineers. Imagine a scenario where the engineer, before entering the factory in the morning, already knows what problems will occur during the day and what actions they need to be prepared to take. Do they feel so empowered? They will. It is because of two reasons. First, they will not be surprised by the upcoming problems and will be ready to act proactively. Second, they can be confident that the actions will effectively solve the problem without any ambiguity. We can help factories to make that happen in reality using AI. We estimate the benefits of using AI solutions is 30-50% shop floor productivity gains which directly adds to the bottom line.

How can AI be used on the shop floor?

Consider a shop floor in a structure of three levels. The bottom is the process focus, the middle is the machine focus, and the top is the factory focus. We can use AI in all three levels to optimize factory performance in a fraction of a time. The process level is where the tool in the machine interacts with the raw material to produce a product. At this level, we can use AI to control and optimize the process by studying the relationship between raw materials and the tool. For example, in a CNC machine, AI can achieve adaptive machining capabilities, i.e., adapting the tool parameters (such as cutting force and feed rate) to the variations in the incoming raw materials to optimize the performance. It can increase the tool life and reduce power consumption. At the machine level, we can use AI to optimize the performance by studying its overall behavior using data collected from different sensors such as temperature, power, pressure, etc. For example, AI can help achieve condition-based and predictive maintenance of the machine, thus increasing its technical availability. At the factory level, we can study the interactions between different shop floor elements such as machines, buffer, and material handling equipment from a flow perspective. The aim is to create a smoother, faster, and swift flow of products in the factory. We can use AI to optimize factory planning, scheduling, resource allocation, bottleneck elimination, and early identification of quality defects to optimize factory performance. A good tip here is to think about automating lean-based flow improvements using AI.

We develop AI solutions for different use cases based on customer needs. We use market-leading tools and platforms for doing this. We develop customized architectures that can extract the data automatically from AWS, Azure, Snowflake, database, or data lake. Then, we run the AI algorithms in the cloud and provide real-time insights and visualizations (e.g., using Power BI, QlikSense, Tableau) and actions to shop floor engineers (either at the back-office computers or mobile phones). We customize our AI solutions to enable shop floor engineers at all skill levels to consume AI insights to support faster and more effective decision-making. Thus, the companies can bring AI closer to the manufacturing engineers who needed it the most in day-to-day decision-making.

How to identify relevant use cases?

Identifying the manufacturing use cases at the process, machine, and factory levels is not a one-man decision. We need both AI and manufacturing engineers’ inputs to select feasible and impactful use cases. It is teamwork. AI engineers have a good understanding of AI’s strengths, weaknesses, and limitations. Manufacturing engineers have extensive knowledge of their manufacturing processes and factory dynamics. They can select high-impact manufacturing problems that benefit from having an AI-based solution. Sitting across a table, both can discuss the manufacturing problems and AI suitability to solve these problems effectively. For example, there might be instances where manufacturing engineers select a high-impact problem. But AI might not be a potential solution to the problem due to the technical difficulties encountered in computing and its implementation. Alternatively, there can be situations where there are medium impact problems, which may benefit from an AI solution in the long run. We can identify those possibilities and limitations only through a cross-functional team of manufacturing and AI engineers. Moreover, such a cross-functional team setup can ensure that the deep insights generated through AI translate into a measurable impact on the shop floor.

Which use case to choose first for AI implementation?

Starting from a factory level can be a good choice. It can provide engineers a top-down perspective of the factory and reveal the critical processes. Moreover, such a top-down approach will help identify the use cases at the machine level and subsequently at the process level. Now there can be a situation where there can be too many factory-level use cases. And the challenge is to prioritize and select the first use case. There are many approaches to prioritize the use cases and start initiating AI projects. We have a unique data-driven approach to helping clients to prioritize the use cases. In this approach, we map all the shop floor problems (in terms of use case) in two dimensions: frequency and impact. Frequency refers to how many times the problems occur in a given time frame. The impact dimension refers to the value impact obtained by solving the problem (e.g., impact on time, productivity, cost, etc.). Once we create a map, it becomes easy to identify the problem sets and create a road map to identify and prioritize the use cases. The first choice is the problems in the high frequency and high impact area of the map for initiating AI projects. We then identify the different tasks in the workflow used to solve the problem. Following that, we assess which decisions are taken by experience and explore how those can be automated or augment those decisions using AI. One additional benefit of having such a map is that one can eliminate the problems in the low frequency and low impact portion of the map to initiate AI projects. These problems may not benefit significantly from having an AI solution, given the complexity, efforts required, and development costs.

What are some recommendations to implement AI successfully?

There are four practical recommendations for manufacturing companies that are starting their journey in AI. These will be helpful to set the right expectations and mindset before a company takes its first AI project.

Recommendations:

• Chose AI for high-impact high frequent problems
• Problem Scoping >> Data sets
• Augmentation is necessary
• Starting small is beneficial

Understanding that AI is not a silver bullet: Every shop floor problem does not require AI. Sometimes simple solutions do work. Moreover, one has to understand that every manufacturing problem can be unique. Each one of them might need a unique AI solution. For example, an AI for predictive maintenance for a roller bearing in a CNC machine might have a different architecture from an AI for a roller bearing in an induction motor. It is because of the differences in geometry of the machines and different machine dynamics. So, the same problem in two different setups can have two unique AI solutions. Our unique problem scoping framework will help to select the right set of problems.

Problem >>> data sets: A common question we always encounter from manufacturing companies is, “I have big data sets. I want insights from the data”. This practice might not be impactful. Having a data focus might lose their attention in solving high frequency and high impact problems. So, selecting the right problem should always be the first step. Manufacturing engineers should spend significant time in this step. Figuring out the problem and working backward towards the AI solution is the most optimal approach. We bring extensive manufacturing and AI implementation experience to help clients appropriately conceptualize and scope the shop floor problem.

Augmentation: Manufacturing engineers should always augment AI results. AI tends to learn the patterns from historical data and provides probabilistic-based predictions of the future. In this process, there are chances that AI might have missed learning certain aspects from the data, and the forecasting may not always be accurate. In one of the use cases, we developed an AI to predict the bottlenecks on the shop floor. The accuracy of the AI was 90% on average after several iterations. Now the question is, how can the accuracy gap of 10% be filled? That is where manufacturing engineer’s domain knowledge gained through years of experience is valuable. We help realize augmentation using our unique humans-in-the-loop AI framework.

Start small: The initial AI projects can be small. Having a wide-scoped project can be complex and can take significant time to complete. As a result, there can be a delay in realizing the benefits of the AI solution. It is good to have a small project (a good tip here is to break down a big project into several small ones) at the start. In that project, develop and quickly implement the AI solution, and realize the benefits. Then, one can gradually increase the scope by including other complexities. Such practices will help companies to learn about what it takes to build, implement and maintain the AI over time. Also, it gives enough time for manufacturing engineers to get experience on how to consume AI insights and feel empowered for a continued AI journey. We help clients in all stages of the AI lifecycle and scale them effectively and efficiently on the shop floor.

I would love to hear your thoughts and dreams about how you want to implement AI on the shop floor. Feel free to reach out to me and, we can jointly realize your AI dreams.

About Author

---

Mukund Subramaniyan

Mukund works in the manufacturing sector, mainly advising clients on topics related to digital manufacturing, data analytics, artificial intelligence, and operations transformation. He has extensive experience working with major automotive companies in India and Sweden including original equipment manufacturers (OEMs) and suppliers. His educational and professional background integrates technical expertise as computer scientists with the manufacturing expertise of engineers. This ensures that the deep insights generated through AI translate into real measurable impact in an organization. He is passionate about transforming manufacturing operations using IIoT, AI, data, insights, and actions.

Contact details: mukund.subramaniyan@capgemini.com

Linkedin: https://www.linkedin.com/in/mukundsubramaniyan/

Twitter: https://twitter.com/mukundgnana03