Proceedings of the XMO Industrial Seminar 2026: Excellence in Manufacturing and Operations
Keywords
Sound monitoring, Convolutional neural network, Machine learning
Tracks
DIGITAL MANUFACTURING
DOI
10.5703/1288284318685
Abstract
Operational state monitoring is critical for assessing machine utilization and reducing downtime in smart manufacturing. However, many Wire Electrical Discharge Machining (Wire EDM) systems remain legacy machines with limited access to controller-level data, making non-invasive sensing approaches necessary. This study proposes a lightweight two-stage sound-based operational state monitoring framework using an Internal Sound Sensor (ISS) and a 1D convolutional neural network (CNN). The ISS, built around a stethoscope-inspired bell structure, captures machine-specific acoustic signatures while suppressing ambient machine-shop noise and spark-related acoustic interference. The proposed framework classifies machine states in two stages: Stage 1 identifies coarse states such as beep, machine-off and machine-on while Stage 2 recognizes co-occurring component-level activities such as wire, coolant, and table operations. Experimental results demonstrate that Stage 1 achieved an accuracy of 0.867 and a macro F1-score of 0.909, while Stage 2 achieved an exact match accuracy of 0.838 and a macro F1-score of 0.934. Latent-space analysis using t-SNE further showed clear separation among operational conditions, indicating that the proposed model effectively learns discriminative acoustic representations from ISS signals. Finally, the framework was integrated into a real-time dashboard to verify its applicability in online wire-EDM monitoring.
Real-time two-stage machine sound monitoring for wire-EDM state recognition
Operational state monitoring is critical for assessing machine utilization and reducing downtime in smart manufacturing. However, many Wire Electrical Discharge Machining (Wire EDM) systems remain legacy machines with limited access to controller-level data, making non-invasive sensing approaches necessary. This study proposes a lightweight two-stage sound-based operational state monitoring framework using an Internal Sound Sensor (ISS) and a 1D convolutional neural network (CNN). The ISS, built around a stethoscope-inspired bell structure, captures machine-specific acoustic signatures while suppressing ambient machine-shop noise and spark-related acoustic interference. The proposed framework classifies machine states in two stages: Stage 1 identifies coarse states such as beep, machine-off and machine-on while Stage 2 recognizes co-occurring component-level activities such as wire, coolant, and table operations. Experimental results demonstrate that Stage 1 achieved an accuracy of 0.867 and a macro F1-score of 0.909, while Stage 2 achieved an exact match accuracy of 0.838 and a macro F1-score of 0.934. Latent-space analysis using t-SNE further showed clear separation among operational conditions, indicating that the proposed model effectively learns discriminative acoustic representations from ISS signals. Finally, the framework was integrated into a real-time dashboard to verify its applicability in online wire-EDM monitoring.