The optimization of a CNC machine tool’s production performance can be approached under the design of an intelligent, self-tuning machine that implements process control/monitoring algorithms, as well as simulations of the machine tool and machining process. However, the robustness of such strategies often relies on the machine tool’s structural dynamics model which is process and position-dependent. A machine tool can exhibit a speed/load-dependent spindle vibration mode which reflects a deflection of the spindle-holder-tool assembly, and positiondependent modes of larger structural components. This thesis presents the identification of inprocess machine tool dynamics using operational modal analysis and machine learning principles. The process-dependent force coefficients which characterize the milling force model are also identified.

The in-process spindle mode is identified using a modification of operational modal analysis which adopts vibration transmissibility as the relative vibrations between multiple sensors. By assuming a consistent mode shape about the external housing that locates the vibration sensors, the proposed method identifies the spindle mode under operating conditions. An alternative strategy is also posed by applying transmissibility to complement the classical inverse stability solution which tunes a modal model of the spindle mode to realize accurate process stability predictions.

The machine tool’s position-dependent dynamics is predicted using the progressive neural network, a transfer learning technique that integrates two networks in the simulation and realworld domains. The simulation domain network is posed under a dynamics simulation model that characterizes the general position-dependency of the machine, and such knowledge is transferred to the real-world domain network to improve its dynamics prediction accuracy under a limited number of experimental data.

The process-dependent force coefficients which form the milling force model are identified using a least-squares method posed under averaged milling forces measured in toolpaths with varying radial immersions or feed rates. The edge force coefficients which reflect the ploughing component of the milling forces are also demonstrated to be viable tool health indicators due to their strong correlation to tool wear.

All proposed methodologies are experimentally validated, and the identification of these process-dependent characteristics can be adopted to improve the fidelity of simulations and process control/monitoring algorithms.