Embedded Files
The left image shows an acoustic sensor mounted next to the spindle of a CNC milling machine, capturing cutting sound in real time during machining. The right image shows how this acoustic data is used — first to train a classification model (Manufacturing AI), and then to deploy that model for real-time field monitoring (Edge AI Deployment). The top section illustrates the flow from raw acoustic signal → MFCC feature extraction → 1D CNN classifier, producing a normal/abnormal classification. The bottom section shows the trained model deployed on a compact Jetson Nano-based control box, which listens to machining sound on-site in real time and immediately reports the result through green/red LED indicators.
Applied Technologies
1. Manufacturing AI — Acoustic-based Normal/Defect Detection
Sound signals generated during machining are converted into MFCC (Mel-Frequency Cepstral Coefficients) features and classified using a 1D CNN model to distinguish normal operation from abnormal or defective conditions. Because the process relies solely on acoustic signals — without additional vision or vibration sensors — it can be readily retrofitted onto existing equipment.
2. Edge AI — Jetson Nano-based Real-Time Tool Wear Detection
The trained classification model is deployed on a compact Jetson Nano-based control box, enabling on-device inference directly in the field without any cloud connection. An acoustic sensor mounted near the CNC spindle continuously captures machining sound, which is classified on the spot as normal or abnormal, with the result output through GPIO-connected LEDs (green/red). This allows operators to immediately recognize tool wear or abnormal conditions without needing a separate monitor.
Manufacuring AI
Edge AI
Page updated
Google Sites
Report abuse