Project: OptimizationChat

In the context of industrial production, there are a variety of optimization problems that can be solved with the help of established solvers. Examples include process parameter optimization, cost-optimized sequence planning, route optimization, resource planning, packing problems, and various problems in the areas of process control and warehouse management. The prerequisite for using mathematical solvers to solve such optimization problems is a formal mathematical representation of the planning processes in the company and a reduction of the calculation problem to a standard problem such as integer linear programming (ILP). However, the description of planning scenarios as calculation problems, with a clearly defined input and a question, is rarely available. Rather, it is the case that the operational processes are available in the form of expert knowledge among specialists. In consultation with optimization experts, it is often possible to identify the calculation problems behind the operational processes. This identification usually takes place in the form of dialogues in which constraints and optimization goals are defined jointly. Furthermore, in these dialogues, the practical feasibility of the modeling is evaluated and iteratively adjusted for sample instances.

The basic idea behind the OptimizationChat project is to automate the process described above with the help of generative artificial intelligence. The project aims to develop the technology for a chat system that derives mathematical formalizations that can be solved with established solvers from a natural language dialogue with the user. The chat system thus takes on the role of the optimization expert from the dialogue described above. The goal is for this chat system to model a calculation problem from an informal process description of operational procedures without requiring the user to have mathematical knowledge. In addition to purely modeling the optimization problem, the chat system should also take into account the problem knowledge required for the solution from a database. Furthermore, it should have an explanatory component that clearly describes the connection between mathematical modeling and industrial practice. A particular challenge in the project is the focus on problems from everyday industrial life and the associated processing of natural language inputs in industry-related language. Instead of simply converting (mathematical) continuous text into a formal representation, the aim is to develop a chat system that is adapted to the vocabulary of German industry. This should enable SMEs in particular, which generally have little or no expertise in mathematical optimization methods, to identify the numerous optimization problems in their companies and convert them into mathematically solvable formalisms. Furthermore, the project aims to derive concrete new business models from the technical possibilities gained.

Fraunhofer IOSB-INA

Uni Bielefeld

Databay

Digital Twin Factory

OPTANO

CLAAS (associated)

HEPU (associated)

IWT (associated)

Remmert (associated)

Sollich (associated)