AI-Cognition-Supporting Assistance System for In-Process Control in Manufacturing

Project Description

AI-supported Image Processing and Assistance Functions

PROJECT GOALS

The goal is to develop an AI-cognition-supporting assistance system for in-process control (KIKA-IPK), which enables a more resource-efficient process and material configuration through self-learning image feature correlations with process properties. Here, the experience knowledge of the machine operator is modeled through machine learning techniques to connect visual quality characteristics on one side and process properties on the other. As a project outcome, an assistance system is aimed at that allows for a more resource-efficient target configuration of process parameters by mapping optical quality characteristics of the product and its process sizes in an AI model during manufacturing.

INNOVATION & METHODOLOGY

Within the framework of the R&D project, methods are developed that enable backward inference from visual product features to process characteristics that are hardly measurable. This makes it possible to regulate the process in such a targeted way that quality deviations during manufacturing are compensated for and efficiently adjusted to new product features. For this purpose, the image, process, and material data streams as well as user feedback during manufacturing are analyzed through the interface of the “AI-cognition-supporting assistance system” (KIKA) and the results are transmitted comprehensibly to the participants and the machine control in real-time. The AI services are integrated into two scenarios for additive manufacturing, 3D metal printing with steel, and the personalized medication printing, demonstrating the resource efficiency potential in industrial applications.

OUR CONTRIBUTION

The technological goal of Gestalt Robotics is primarily to expand the technology portfolio to include services of “Active Learning” involving user feedback. In this way, a technological bridge is created between existing application areas of AI-supported image processing and new fields of application in intelligent assistance systems. Additionally, the industrial application of exploratory learning methods, e.g., Reinforcement Learning, can be piloted within the project framework. Concrete outcomes include a recommendation system to support the machine operator in the visual product characterization using ML methods during manufacturing and corresponding measures for compensating quality deviations as specific instructions for the machine operator and the control system.