Toward compact, low-latency, and reliable monitoring pipelines

The Design of Minimal Deep Learning Systems for Online Tool Condition Monitoring in Stacked Drilling

Jiduo Zhang¹, Dongze He^{2, 3, *}, Robert Heinemann¹, Otto Jan Bakker¹

¹ Department of Mechanical and Aerospace Engineering, The University of Manchester
² Henry Royce Institute, Department of Materials, The University of Manchester
³ National X-ray Computed Tomography

Overview

This study establishes a practical route for designing minimal deep learning systems in stacked-drilling monitoring. Instead of enlarging model size, it identifies the minimum signal conditions needed to preserve high and stable incidence-recognition performance while reducing sensing, transmission, and computation costs.

Major Contributions

• A minimum-sufficient signal design framework is proposed to link sampling duration, frequency, and phase with deep-learning reliability.
• Boundary conditions for minimum sufficient duration (MSD) and minimum sufficient frequency (MSF) are derived and validated from harmonic structure and spindle rate.
• Equivalent-accuracy behavior under translation-invariant CNN recognition is established to support immediate response without major accuracy loss.
• Engineering-oriented guidelines are provided for compact and low-latency online tool condition monitoring deployment.

Key Findings

• Critical process incidences remain accurately identifiable with shorter signals once minimum sufficient conditions are satisfied.
• Sampling frequency and duration dominate model performance, while phase influence is limited under stationary and homologous segment assumptions.
• Spindle-related harmonic information is the key carrier of discriminative features.
• Minimal-signal design can significantly reduce data volume and system overhead while preserving classification quality.

Practical Implications

For industrial deployment, this approach supports faster online monitoring with lower data and compute requirements while maintaining decision quality for adaptive drilling.