One of my favorite projects is also a wonderful use case to analyze if Industrial Cloud is feasible. With my background in the automotive industry and industrial automation, it should be no surprise that this relates to car part manufacturing. After joining AWS re:invent as analyst where I focused on Industrial Machine Learning and Cloud in Manufacturing, I decide to revisit this project and give you an update on this USE CASE.
After a very successful pilot project to optimize the process of filling casting machines with liquid aluminum, the team was eager to bring the solution to other facilities. And with that goal in mind, the team also realized that it was necessary to automate the learning process. Different facilities have different machines and different products, so a “copy paste” operation would not make sense. And there are many more variations that influence the process, like for example different ladles to fill the holding furnaces.
While analyzing the already available data from the different facilities, the team kept discovering variations which they would have addressed with assumptions and rules-of-thumb, just like they did before this project started. They discovered for example that small variations in the alloying of aluminum have significant influence on the temperature curve. Since most facilities get their aluminum supplies from different locations, these variations turned out to be very important.
During the pilot phase, the team had learned that the type of the ladle and the aging of the insolating lining have impact on the temperature development of the aluminum during the filling process. When comparing they available information, they also discovered that different types of forklifts and loading installations influenced the duration of the loading process and with that, the temperature development. All in all, there were too many variables to pursue a “manual learning process”.
With their service provider the decision was quickly made to opt for a cloud-based Machine Learning model to analyze all the variables and results from all facilities. Being an AWS partner, the service provider selected Machine Learning on AWS as platform to analyze the data and generate rules that feed the on-prem MES and Production Planning functions. With their ERP already run from the cloud, and all plant-floor data available through APIs, all required data could be extracted directly, a big advantage over the original pilot project. And a clear confirmation that the team had taken the lessons learned from the pilot project very serious.
As Digital Transformation advocates, we always highlight that it is not just about technology. Mindsets and skillsets need to evolve with the transformation. As coach for the original pilot project, I was delighted to see that this was happening throughout the organization. Let me give you a great example.
Filling the holding furnaces is only possible when the furnace is not pressurized. This gives the forklift driver a fixed window which is determined by the recipe for the product that is casted on the machine. When the furnace is not pressurized, the machine is either opening to extract the product, or closing to start a new cycle, and any delay would cause the casting mould to cool down and an interruption of the cycle. And that is something that needs to be avoided!
This rule was set as a fixed rule extracted directly from the recipe, and accepted by everyone as valid, until something unexpected happened. During a test phase for an improved recipe, the cycle was shortened by 30 seconds, but because this was just a test phase, this change was not entered into the recipe for the product. Much to everyone’s surprise, the ML analyzes indicated that this offset of 30 seconds did not influence the mould temperature as much as everyone expected, and the positive impact of additional time to fill the holding furnace to the optimal level outweighed the negative impact of interrupting the optimal casting cycle.
2 assumptions everyone had taken as essential flew out the window with one unexpected variation in the process! A few years ago, it would have been unthinkable that the team would have taken this as an important lesson, but not anymore. They instantly started to evaluate how this could impact the project and which benefits could be achieved. Some careful tests revealed that this isn’t a straightforward variation, like most others were not straightforward.
The team had gained so much confidence in the Machine Learning solution that they decided to gradually release the fixed rule for the filling time window and let the ML engine figure it out. As a Digital Transformation Coach, I recognize two very important changes:
- The team has learned to enjoy letting go and learn from every experience.
- The team had also learned to navigate with the system and create a learning environment for themselves and the technology.
This USE CASE Industrial ML and Cloud in Manufacturing based on AWS re:Invent shows that Industrial Cloud is feasible with the right mindset and partners. That also means that we have to be willing and able to accept the challenges and opportunities of cloud and digital transformation in industrial environments. A good way to start is the online business education module Cloud & IoT in Manufacturing – Embrace the potential of Cloud and IoT without exposing your organization to uncontrolled risks.