Completed project 07/2023 - 06/2025

Current progress in the field of artificial intelligence is mainly being made in video/image processing and language models. In machining production, on the other hand, the use of AI has so far remained the exception. This is due to the high complexity and variability of the processes in conjunction with the challenge of obtaining high-quality data in the first place. Many machine tools are not designed to collect data. Retrofitting is usually expensive, technically complex and disrupts the production process, which often makes economic implementation unattractive. Data generation itself is also costly.  It requires long machine runtimes, qualified personnel, and high material consumption due to tools and workpieces. Additionally, the amount of data generated is small and characterized by high information redundancy, making it difficult to interpret and transfer to other processes. New technologies that are robust, adaptable and transferable are needed so that data processing methods can be used sensibly in production. They must be able to work with data that is collected directly in the process and at a reasonable cost. This is precisely where the project "Intelligent software services based on the iTENDO sensor-based tool holder" comes in.

The iTENDO research project is developing technologies that can evaluate vibration data from SCHUNK's sensor-supported tool holder. The aim is to reliably predict central process conditions in machining production, such as tool wear or chattering, and thus create the basis for intelligent process monitoring.

Physical models for describing manufacturing processes require numerous experimentally determined parameters that have to be redetermined for each machine and each process. This makes them difficult to transfer and expensive to implement. This is why iTENDO pursues a data-based approach.

For this to remain economical, a data acquisition system is needed that can be easily and cost-effectively integrated into existing machines. The iTENDO tool holder meets precisely these requirements. It is used like a conventional holder, but also provides 1D acceleration data directly on the tool. The result is a practical and scalable solution for data acquisition. This data is analyzed using various methods. Methods from signal processing such as the Fourier transformation, the Welch method and the short time Fourier transformation are used. The extracted information is further processed using either neural networks such as convolutional neural networks or specially developed algorithms. Among other things, these calculate energy distribution, entropy, anomalies or the complexity of the signals. In the next step, the process states are predicted with the help of classification and regression models. Recurrent networks such as long short-term memory models are also used here. At the same time, the amount of computing power and training data required to enable reliable predictions and make the solution efficient and profitable for industrial use is investigated.