Production Optimization Algorithms Recognize the Causes of Errors in the Manufacturing System and Learn from Them
Fraunhofer researchers have developed an analysis tool that detects faults in fast-cycle production systems and improves troubleshooting. The software has already proven itself in various applications.
The pharmaceutical and consumer goods industries, run capital-intensive manufacturing facilities and rely on continuously maximizing productivity. Otherwise there is a risk of cost pressure and financing gaps. However, in practice: “The more complex the plant, the lower the productivity." This is how Felix Müller, Group Leader Autonomous Production Optimization at Fraunhofer IPA, sums it up. Furthermore, many production plants comprise a large number of stations and process jobs so quickly that the causes of faults cannot be seen with the naked eye.
Self-Learning Algorithms Evaluate Data Synchronously
Using "Smart System Optimization", Müller and his team have developed an analysis tool that continuously detects errors and their causes in fast-acting, interlinked production systems: A powerful connector accesses the data in the machine control system at high frequency via the respective manufacturer protocol. The result is a continuous database, which is evaluated by several self-learning algorithms synchronously. They identify where faults during production arise, how they are related and what their priorities are in relieving them. In this way, defects that lead to the failure of the entire system can be repaired or even predicted more quickly.
Operator Assistance System Continuously Learns New Things
However, it is not always clear what to do if an error is going to occur. Additionally, follow-up messages from the system make the situation even more confusing for the operator. For this reason, Müller and his team have developed Shannon, an intelligent operator assistance system for complex production lines based on "Smart System Optimization". Until now, it was often up to the machine operators to decide what to do to correct a fault. Now, however, the affected machine sends them a detailed step-by-step guide to the smartphone or tablet. The database and the links between faults and solutions are constantly being expanded during plant operation.
This gives system operators the opportunity to create their own instructions, for example on troubleshooting measures. These instructions can include text, photos or videos. Furthermore, the plant operator can give feedback on the information provided, which is used to improve it. Plant operators are also actively encouraged to contribute knowledge, for example in the description of detected but unknown events. In this way a clearly understandable and continuously linked knowledge base comprising errors, events and solutions can be established. Shannon is currently used as a tablet and smartphone app in several factories, where it has significantly reduced the time it takes to troubleshoot problems.
Automated Machine Benchmarking Increases Efficiency
Automated machine benchmarking is also possible with "Smart System Optimization": In many production halls there are dozens of identical or similar machines and always carry out the same processing cycle. Examples include injection molding, die casting, blow molding and thermoforming machines. Although they all have the same structure, some work slower than others. This is usually due to wear of certain components, varying sensor behavior or different tool settings as well as material fluctuations.
In machine benchmarking, the entire process is first defined on one machine and divided into individual steps. The high-frequency connector then generates a database at the machine controller that evaluates a machine learning algorithm package. This happens simultaneously at all connected machines and is merged virtually to form an ideal process sequence. Based on this, the tool immediately recognizes when a machine is running slower than planned and links this to a technical cause. Users can thus not only eliminate faults before they occur, but also obtain an optimized cycle time for the connected machines by combining the best individual steps. Depending on the machine, this led to cycle time reductions of between 2 and 18 % in the previous prototype applications. So even the fastest machine to date can become even faster. The application has now been converted into a continuously learning software called Darwin.
Researchers Set up a Start-Up
Darwin, the intelligent machine benchmarking system, has already been used by several automotive suppliers and an injection molding machine manufacturer - also across plants. Shannon is already being used by large automotive and pharmaceutical companies. Reason enough for Felix Müller and his two co-founders Thomas Hilzbrich and Pablo Mayer to become start their own business with "Smart System Optimization". Its start-up company, Plus 10 GmbH, currently has offices in Stuttgart and Augsburg and recently commenced operations.
This article was first published by konstruktionspraxis.
Original by Gudrun Zehrer / Translation by Alexander Stark