Fraunhofer Institute for Production Technology IPTResearch project "AIDpro"
In modern production environments, artificial intelligence (AI) is playing an increasingly central role. One example of the use of AI in the production environment is “predictive maintenance”. Here, AI algorithms analyze sensor data to detect early signs of machine wear or failure. However, the successful application of AI depends to a large extent on the quality of the underlying data. Production data can be affected by a variety of factors, including sensor errors, data loss, manipulated data or machine defects. In addition, uncertainties arise from variable manufacturing processes, such as seasonality, contamination or wear, which are insufficiently taken into account when training AI models. One of the major benefits of AI is the support it provides for decision-making based on data. However, the decisions that are now based on unreliable data pose significant risks for products, machines and employees. This is where the research project “Anomaly Detection in Production – Data Validation for Production Processes (AIDpro)” comes in, developing a comprehensive solution for automated data validation and anomaly detection in process data streams.
AI in production – but only with assured data quality
The aim of the project is to create an innovative data validation system that ensures a reliable database for AI-supported processes. The system not only enables automated validation of production data, but also continuous monitoring of the application phase of AI solutions. It analyzes and evaluates data quality by performing completeness and confidence checks using expert knowledge. This involves a rule-based check to determine whether an individual data point from a data set corresponds to a predefined data schema. This is complemented by statistical analysis methods and outlier detection supported by deep learning. The system will also be able to detect a change in the data that indicates a change in production conditions, known as data drift. This information can be used, for example, to retrain the AI model with the changed production conditions in order to map these changes.
A modular structure allows the system to be flexibly integrated into various production processes. The development is based on specific industrial use cases to ensure a practical and adaptable solution. Combining state-of-the-art AI technologies with in-depth production knowledge creates a powerful monitoring and warning system that detects deviations in production processes in real time, thus enabling countermeasures to be taken at an early stage.
Predictive optimization through continuous monitoring
AIDpro provides a reliable basis for the safe and efficient use of AI in production. Decisions can be based on verified and validated data, minimizing risks and increasing process stability. Continuous monitoring of process data streams ensures that anomalous data points and trends can be immediately detected and dealt with accordingly. Companies therefore benefit from continuous monitoring of their process data while also laying an important foundation for the reliable industrial application of AI in their production environment.
In addition, the project contributes to long-term process optimization by tapping into potential for “predictive quality” and “predictive maintenance”. This not only enables higher product quality, but also more efficient and resource-saving production. Thus, the system should also take into account and promote sustainable and resilient production.
Project partners
- Fraunhofer-Institut für Produktionstechnologie IPT, Aachen
- Fraunhofer-Institut für Angewandte und Integrierte Sicherheit AISEC, Garching b. Munich
Project committee
- Aventus GmbH & Co. KG, Warendorf
- Aviat Networks LTD., Munich
- Bitmotec GmbH, Hemmingen
- elunic AG, Munich
- Giesecke+Devrient GmbH, Munich
- KSB SE & Co. KGaA, Frankenthal
- logsight.ai GmbH, Berlin
- Sikora AG, Bremen
Funding information
The research project “Anomaly Detection in Production – Data Validation for Production Processes (AIDpro)” is funded by the Federal Ministry for Economic Affairs and Climate Action as part of the “Industrielle Gemeinschaftsförderung” program based on a resolution of the German Bundestag.
