Focal points
Under the direction of the two computer science professors Dr Frank Schirmeier and Dr Ulrich Göhner, the IDF focuses on combining mathematical optimisation methods with artificial intelligence applications in the manufacturing environment. The research topics at the IDF are organised into three clusters Digital production plants and data-centric business models, AI-optimised manufacturing processes and innovative sensor technology and large language models and industrial generative AI.
In-process quality control
The IDF is conducting research (funded by the Bavarian Ministry of Economic Affairs, Regional Development and Energy) into in-process QC (i.e. quality control based on sensor data during the manufacturing process, rather than downstream) in machining processes. This can drastically reduce time-consuming and costly measuring efforts.
Innovative data collection methods
Data scientists and mechanical engineering experts conduct interdisciplinary work hand in hand to channel their combined expertise into the IDF’s models. Developing proprietary data collection methods is often crucial for measuring high-quality target variables for AI models.
Production deviation
By integrating image data into a holistic quality assessment pipeline, optical error images can be evaluated together with data from other sensors, measuring e.g. vibration or force.
Established deep learning architectures are used to evaluate image data, which are combined with innovative methods for reducing dimensions and evaluating the relevant image content.
Measurement Methods and Data Collection
A specially developed measurement setup is used at the IDF to precisely record and track tool wear. This setup enables the examination of wear phenomena in the micrometer range under reproducible conditions.
The setup combines high-resolution telecentric optics with precision rotary and linear stages derived from laser technology. This allows tools to be precisely positioned and repeatedly measured from defined perspectives.
The captured image data is analyzed using modern computer vision techniques. This enables relevant wear features to be automatically detected, quantified, and documented throughout the course of the experiments.
Artificial intelligence in manufacturing can reduce material and energy costs as well as scrap, while ensuring consistently high manufacturing quality.
- Innovative analysis of vibration and force measurement data
- Real-time quality control and stability assessment
- Condition monitoring and anomaly detection
- Optical wear measurement
- Simulation of material removal and cutting forces in machining processes
Investments in new technologies strengthen companies’ innovative capacity and international competitiveness.
- Process Time and Cost Forecasting
- Energy Consumption Forecasting, Adaptive Energy Efficiency Optimization, and Energy Flexibility
- Data-Driven Optimization of Planning Processes
- Data Protection-Compliant Cloud Developments for Process Monitoring
- Intelligent, Self-Learning Automation Systems
Digitalisation and automation
Digitalising and automating business processes present a major challenge, as efficient solutions are required in response to increasing complexity and the shortage of skilled workers.
The IDF is researching how business processes can be optimised using hybrid systems consisting of classic algorithms and AI – focusing heavily on applying spatial augmented reality and integrating robotics to extensively automate processes. The aim is to reduce the amount of expert knowledge required, relieve the pressure on skilled workers and shorten process times.
Innovative business models
As part of the European cloud initiative GAIA‑X (and funded by the Federal Ministry of Research, Technology and Space), research is being conducted at the IDF into innovative business models made possible by cross-company data exchange.
This broadly spread data exchange generates enormous added value, but it can also violate sensitive trade secrets – which is why the IDF is developing innovative data anonymisation and encryption methods.
Detecting anomalies
In time-series data streams, such as those generated by industrial sensors, deviations from normal behaviour are often difficult to detect. Assistance can be offered in these cases by the specialised AI methods developed by the IDF in collaboration with well-established manufacturing companies and implemented on site.
Optimisation
Significant efficiency and cost benefits can be achieved by evaluating the data generated using AI methods in business settings.
For specific problems, however, mathematical optimisation algorithms have been shown to find better solutions; in the best case scenario, even the mathematically demonstrably optimum solution.
The IDF’s experts in mathematics and optimisation combine the most appropriate solution methods for each given application scenario.
Correlation and causality
Examining correlations is an established part of exploratory data analysis ¬– helping to identify anomalies in a data set even prior to applying AI methods.
Specifically in the manufacturing environment, identifying genuine causal relationships can generate significant added value: What effect does process parameter X have on quality criterion Y?
To answer this question, the IDF uses highly innovative methods to model causal relationships and then validates the identified causalities in dialogue with manufacturing experts.
Enhancing CAD with Manufacturing Information
Enriching CAD Files with PMI from the Drawing
Today, manufacturing-related information is often scattered across different sources: geometry in 3D CAD files, tolerances, surface finish, and threads in a technical drawing. Although it is possible to incorporate PMI (Product Manufacturing Information) into the CAD model, this process is still very cumbersome and rarely used. This poses a significant obstacle to analysis by AI as well.
The CADistency project at the IDF addresses this very issue and is researching how PMI can be transferred from a drawing into the corresponding CAD model. To this end, a Transformer AI is being trained on millions of synthetically generated pairs of CAD models and drawings.
LLM Agents for CAD Analysis
Tasks such as work planning and CAM programming, which require handling complex CAD files, still cannot be reliably performed by Large Language Models (LLMs).
To make LLM-based CAD analysis more robust and reliable when scaled to complex CAD files, the KIANA project is evaluating both fine-tuning and agent-based approaches to help contract manufacturers quickly evaluate incoming orders.
Predicting Production Times
Processing CAD Files with 3D Convolutional Neural Networks
Based on a large database of similar components, the IDF trained an AI model that predicts machining times on a CNC machine. This enables faster and more effective work planning.
Depending on the problem at hand, various technical approaches are suitable for processing CAD data using AI, ranging from rule-based extraction of attributes such as area and volume to the representation of CAD models as graphs.
AI-Based CAM Planning
Graph-Based Assignment of Machining Operations
In various ongoing and future projects, the IDF is focusing on the automation of CAM programming—that is, the creation of programs for CNC machines based on a CAD model.
In the CAMistry project, the historical database of CAM data is semi-automatically broken down into reusable feature templates. An AI then assigns suitable templates to machining features in new parts.
Knowledge Integration for Chatbots
Customer chatbot developed at IDF with a dashboard for inquiries
Their ability to analyze large amounts of text in a short amount of time while responding to questions in natural language makes LLM-powered chatbots an excellent tool for quickly finding specific information in documentation or machine manuals. LLMs can also quickly generate suggestions for communicating with customers, thereby saving time.
Generative AI opens up new possibilities along the entire CAD-CAM process chain—from evaluating incoming orders to work planning and manufacturing.
- Support for design processes using large language networks and deep learning methods
- Research into language AI systems for analyzing big data sets such as user manuals, legal texts, and patents
- Robotics and automation solutions based on vision-language-action models
