¶ UC5 - AI techniques at different layers in Machine Tool Domain
This use case applies AI across different organizational and technical layers in precision machining of large-size parts. The goal is to improve machining efficiency, product quality, process robustness, and resource use by helping operators understand the current state of CNC operations, estimate remaining time, detect abnormal machine behavior, and receive AI-supported recommendations. In parallel, the use case addresses interoperability between data layers in machine-tool companies, including production, logistics, scheduling, and management systems. The expected outcome is AI-based process optimization, predictive maintenance, output quality measurement, improved human-machine interaction, fewer faulty parts, reduced rework, and better energy and resource efficiency.
| # |
Technology Name |
Description |
| 1 |
Automatic tool breakage detection in milling operations |
The system integrates advanced AI and signal-processing techniques to enable robust, scalable, and real-time detection of tool breakage in milling operations. The system is deployed at the edge using a microservice architecture. |
| KPI # |
Related Technology |
KPI Target |
| 1 |
Regression and classification algorithms for tool breakage and tool wear |
Tool failure detection model accuracy above 95%. |
| 2 |
Tool breakage detection application |
Tool failure detection in a real environment within 3 seconds. |
| 3 |
Motif Discovery Singularity Toolkit |
50% reduction in data segmentation time compared to the current manual segmentation procedure of part programs. |
| 4 |
FFT feature extraction for tool breakage detection |
High-frequency data processing and feature extraction at 1-second frequency. |
| KPI # |
Baseline |
| 1 |
No system for tool breakage detection. |
| 2 |
No baseline value provided. |
| 3 |
No baseline value provided. |
| 4 |
No baseline value provided. |
| KPI # |
Validation Method |
| 1 |
Evaluate the capability of the tool breakage detection algorithms under lab conditions, using two sets of experiments: induced tool failures and experiments extended until the tool reaches end of life. |
| 2 |
Deploy the detection system in a shopfloor machine environment and use Savvy platform data to validate performance within the 3-second detection target. |
| 3 |
Test the Motif Discovery Singularity Toolkit web application and collect user feedback on segmentation quality and ease of use. Compare the average time required for manual segmentation with the time required when using the toolkit, based on expert estimation. |
| 4 |
No validation method provided. |
- 25% improvement in classification precision.
- 30% improvement in time spent on quality control.
- Reduced re-configuration time of the AI-based system by approximately 30–50%.
- 20% reduction of required data for scaling new ML models to new machines.
- Self-adjusted/adapted manufacturing processes above 15%.
The use case contributes to improved classification precision through FFT-based harmonics analysis for high-quality feature extraction from vibration signals, automatic segmentation using the Motif Discovery Singularity Toolkit, and the testing of several ML algorithms for comparison. The real-time detection of tool breakage reduces the need for manual inspection and supports progress toward unattended machining, contributing to reduced time spent on quality control. Re-configuration time is addressed through dockerized microservices, fast deployment and updates, and Arrowhead integration. The modular architecture, including FFT extraction embedded in the PAC system, 1-second FFT feature reading at the Savvy Edge, and deployment of the tool failure detection application, supports scaling ML models to new machines with less required data. Fine-tuning capabilities based on automatic threshold analysis, an AI-Gym training environment, and configurable notification and alerting mechanisms support self-adjusted and adapted manufacturing processes.
| AI Toolbox Tool |
Use within the Use Case |
| FFT feature extraction for tool breakage detection |
Used to process high-frequency vibration signals from milling operations and extract frequency-domain indicators for tool breakage detection. The tool supported 1-second FFT-based feature generation for the edge-deployed detection pipeline. |
| Time-Series Anomaly Detection Tool |
Used to detect abnormal temporal patterns in machine and vibration signals during milling operations. It supported the identification of tool failure events and abnormal machine behavior in both lab and shopfloor validation settings. |
| Tabular Prediction Tool |
Used to train and evaluate regression and classification models for tool breakage and tool wear detection based on extracted process and signal features. It supported the target of achieving tool failure detection accuracy above 95%. |
| Device-Level Quality Detection Tool |
Used to connect machine-signal-related features with device- or process-level quality issues. It supported the analysis of how tool breakage, tool wear, and abnormal machining behavior affect machining quality and production robustness. |
| Explainability Tool |
Used to interpret the tool failure detection models and support operator understanding of the most relevant signal features behind breakage or wear predictions. This helped make AI-supported recommendations more transparent in the machine tool domain. |