Project SUPPORT: New form of artificial intelligence optimizes production planning

Production planning and control (PPC) in companies influences profitability as well as material flows, machine assignments and the actual deployment of employees. Current methods for planning optimization reach their limits due to the high complexity and usually only focus on increasing the productivity of a production process.

The aim of the project is to simplify complex production planning. This is to be achieved through reinforcement learning. Reinforcement learning is a form of machine learning and thus belongs to artificial intelligence (AI). The advantage of reinforcement learning is that it can find solutions even for very complex problems. For this purpose, the AI is not shown which action is the best in which situation, but it receives a reward at a certain time, which can also be negative. However, a simulation model is needed for training. Since manual creation is costly, the model is to be created automatically in SUPPORT. In addition to the previous optimization variables, the workload of the employees should also be taken into account, so that an increase in productivity takes place in harmony with the employees. This should make it possible to solve complex PPS tasks efficiently and sustainably with little effort.

Problem description 

Production planning and control is the central instance for influencing the operational processes in manufacturing companies. It influences both the profitability and thus the economic sustainability of the operation as well as the material flows, the machine assignments and utilization as well as the concrete deployment of all employees involved in production. This multitude of elements to be controlled determines the complexity of the planning task. Added to this are market-related challenges such as a high number of product variants, short product life cycles, shortened planning horizons and a lack of skilled personnel. Current methods for planning optimization reach their limits at a high level of complexity and mostly focus one-dimensionally on increasing the productivity of a manufacturing process. In particular, social and human aspects, which are also influenced by the production plan and the control rules, are not taken into account for reasons of complexity alone. Furthermore, there are high efforts in the creation of production plans as well as a lack of adaptability and reactivity in case of changed or new production environments as well as in case of disturbances and external influences.

Objective & approach

In this project, the challenges of production planning are to be overcome by performing highly complex production planning and control tasks with reinforcement learning (RL) methods from the subject area of machine learning (ML) and artificial intelligence (AI). Thereby, a high quality of optimization is to be achieved, while simultaneously mapping economic and ecological as well as human-centered target variables. Due to the particular importance of highly qualified employees in Germany as a high-cost location, physical and psychological factors influencing the well-being and long-term performance of operational employees in the production process will be directly included in the planning process.

The project consists of the development of two core components. First, the data available in a production facility is to be modeled using digital twins and then used to create an automated simulation model. Subsequently, these simulation models are to be used to train RL agents and thereby optimize production planning. The preliminary work of the application companies in the areas of production optimization, production simulation and digital twins will be taken up and integrated into the components to be developed.

Project result & added value

For a reusability of the results, the data modeling is done by means of digital twins and the data exchange to existing IT systems is done by standardized interfaces. In addition, a toolbox will be created that will enable business end users to review the generated simulations and production plans and adapt them according to their expertise. In this way, companies can be provided with a low-threshold entry into AI-supported production planning. The added value lies in high productivity and relief for production planners and employees.

The practical implementation of this approach is taking place in two pilot projects, which encompass different production environments: On the one hand, just-in-time production will be considered, which requires a high time availability of the ordered products. On the other hand, the methods will be applied to a make-to-stock production so that the broadest possible transferability to other companies and industries is ensured.

Profile