Optimized production planning with AI: ISRINGHAUSEN relies on intelligent factory control
As a manufacturer of seating systems, ISRINGHAUSEN faces the challenge of mapping a large number of variants on an assembly line. Since the production sequence of the variants has a high influence on productivity and the workload of the workers, optimizing the sequence is of great importance. As this is a complex optimization problem with many influencing variables, it cannot be solved economically using conventional approaches. However, there is potential in the use of an AI agent which, when properly trained, generates solutions for different optimization variables. The it’s OWL project ‘Sustainable and Human-centered Production Planning and Control Based on Reinforcement Learning Techniques (SUPPORT)’ addresses this issue and focuses on production planning with artificial intelligence. In collaboration with Bielefeld University of Applied Sciences and Fraunhofer IOSB-INA, ISRINGHAUSEN is developing an AI agent based on reinforcement learning technology that uses various data to determine the production sequence of vehicle seats.
Challenges of production sequences with high variant diversity
Vehicle manufacturers offer their customers a wide range of configurations. This also includes the configuration of the seats, which means that ISRINGHAUSEN has to produce a large number of variants on the production lines. The variant-forming options have different effects on production times in some cases. The optimum production sequence essentially depends on the workstation-specific production times, so that the focus is on the variance in production times.
In the production line, the individual work content is divided between several workstations, whereby the work content per station varies greatly depending on the seat variant to be produced. To level out the workstation-specific production times, a buffer area is provided after each workstation. In this way, production can be mapped in continuous flow despite different production times per workstation.
Planning the production sequence with AI agents
The AI agent is trained with data from previous production and learns to plan the production sequence taking various optimization variables into account. These optimization variables are, for example, productivity, line utilization and worker load. To plan the production sequence, the production orders for different seats are bundled into a batch. The various production times are then assigned to the workstations for each seat. Based on this information, the optimum production sequence is determined using various predefined rules. In the future, the AI agent will support factory control.
Data basis for training the AI agent
As part of the SUPPORT project, ISRINGHAUSEN, in collaboration with Fraunhofer IOSB-INA and Bielefeld University of Applied Sciences, is developing an AI agent that is able to ensure maximum productivity in line production with a high number of variants using intelligent production control. At the same time, the aim is to reduce the workload on employees by equalizing complex seat variants in the production sequence and balancing the workload per workstation. Various influencing variables, such as the complexity of the requested variants, the delivery date, material availability, employee qualifications, the day of the week and the corresponding time of day, are taken into account and used as the basis for the AI agent. A central data model of human-centered parameters has already been developed in the SUPPORT project. It combines all available data sources and has an interface to enable live data access. Building on this, the next step is to train the RL agent with data.
