Machining processes form the cornerstone of modern manufacturing industries, yet their inherent complexity arises from nonlinear interactions among multiple physical mechanisms. These interactions impede accurate process prediction, resulting in productivity losses due to uncertainties in machining efficiency, dimensional accuracy, and operational safety. Online machining condition monitoring enables continuous capture of process dynamics and real-time detection of anomalies, particularly when integrated with intelligent algorithms that optimize engineering parameters and enhance equipment safety. However, conventional multi-sensor acquisition systems with high sampling rates generate prohibitive data volumes, incurring substantial transmission, storage, and processing costs. To address these limitations, this PhD research develops a novel On-Rotor Sensing (ORS) methodology which acquires high-fidelity vibration signals directly from rotating components. This approach yields significantly more stable vibration data with enhanced signal-to-noise ratios (SNR). Analysis of milling experimental data demonstrates that, compared to workpiece-mounted wired accelerometers, the ORS system achieves a 92.1% reduction in signal fluctuation amplitude and over 99% improvement in noise suppression. Fundamental differences exist between turning and milling dynamics: turning generates continuous stable forces whereas milling produces intermittent cutting forces governed by time-varying insert engagement variations. To establish high-fidelity dynamic models, this work conducts separate investigations. For turning systems, a finite element model (FEM) incorporating spindle-chuck-workpiece assemblies was developed, with bearing stiffness calibrated through numerical simulation. The resultant stochastic force model demonstrated accuracy in predicting vibration responses during experimental validation. For milling operations, an innovative impact force model was formulated and the first to comprehensively integrate tool runout, wear progression, and breakage effects. Collectively, these physics-based models enable deeper insights into the physical mechanisms governing machining conditions and tool wear progression. Traditional wired sensors face installation limitations and provide limited insights into workpiece quality metrics such as surface roughness. Consequently, the ORS system was implemented on both lathe and vertical milling spindle systems. Comparative studies verified that ORS signals exhibit a correlation with machining conditions than wired accelerometers, capturing vibration characteristics more representative of actual cutting conditions. Experimental validations conducted on industrial machining equipment confirmed the effectiveness of ORS. Milling parameter identification was accomplished using the proposed Zero-Crossing Tacholess Order Tracking (ZC-TLOT) spectrum analysis. For tool wear states classification, the methodology attained 93.4% accuracy utilizing only three time-domain features (Gini index, RMS, mean). This performance surpasses traditional methods (83.6% accuracy) by 9.8%, demonstrating significant robustness under substantial noise interference. These advancements validate ORS as a transformative methodology for real-time condition monitoring. Industrial machining experiment validations confirm enhanced tool states detection sensitivity and superior predictive capability for surface roughness, ultimately reducing quality control costs within automotive component manufacturing.
| Date of Award | 16 Sept 2025 |
|---|
| Original language | English |
|---|
| Sponsors | Beijing Institute of Technology |
|---|
| Supervisor | Fengshou Gu (Main Supervisor) & Helen Miao (Co-Supervisor) |
|---|