Virtual Model Building for Multi-Axis Machine Tools Using Field Data
MetadataShow full item record
Accurate machine dynamic models are the foundation of many advanced machining technologies such as virtual process planning and machine condition monitoring. Viewing recent designs of modern high-performance machine tools, to enhance the machine versatility and productivity, the machine axis configuration is becoming more complex and diversified, and direct drive motors are more commonly used. Due to the above trends, coupled and nonlinear multibody dynamics in machine tools are gaining more attention. Also, vibration due to limited structural rigidity is an important issue that must be considered simultaneously. Hence, this research aims at building high-fidelity machine dynamic models that are capable of predicting the dynamic responses, such as the tracking error and motor current signals, considering a wide range of dynamic effects such as structural flexibility, inter-axis coupling, and posture-dependency. Building machine dynamic models via conventional bottom-up approaches may require extensive investigation on every single component. Such approaches are time-consuming or sometimes infeasible for the machine end-users. Alternatively, as the recent trend of Industry 4.0, utilizing data via Computer Numerical Controls (CNCs) and/or non-intrusive sensors to build the machine model is rather favorable for industrial implementation. Thus, the methods proposed in this thesis are top-down model building approaches, utilizing available data from CNCs and/or other auxiliary sensors. The achieved contributions and results of this thesis are summarized below. As the first contribution, a new modeling and identification technique targeting a closed-loop control system of coupled rigid multi-axis feed drives has been developed. A multi-axis closed-loop control system, including the controller and the electromechanical plant, is described by a multiple-input multiple-output (MIMO) linear time-invariant (LTI) system, coupled with a generalized disturbance input that represents all the nonlinear dynamics. Then, the parameters of the open-loop and closed-loop dynamic models are respectively identified by a strategy that combines linear Least Squares (LS) and constrained global optimization. This strategy strikes a balance between model accuracy and computational efficiency. This proposed method was validated using an industrial 5-axis laser drilling machine and an experimental feed drive, achieving 2.38% and 5.26% root mean square (RMS) prediction error, respectively. Inter-axis coupling effects, i.e., the motion of one axis causing the dynamic responses of another axis, are correctly predicted. Also, the tracking error induced by motor ripple and nonlinear friction is correctly predicted as well. As the second contribution, the above proposed methodology is extended to also consider structural flexibility, which is a crucial behavior of large-sized industrial 5-axis machine tools. More importantly, structural vibration is nonlinear and posture-dependent due to the nature of a multibody system. In this thesis, prominent cases of flexibility-induced vibrations in a linear feed drive are studied and modeled by lumped mass-spring-damper system. Then, a flexible linear drive coupled with a rotary drive is systematically analyzed. It is found that the case with internal structural vibration between the linear and rotary drives requires an additional motion sensor for the proposed model identification method. This particular case is studied with an experimental setup. The thesis presents a method to reconstruct such missing internal structural vibration using the data from the embedded encoders as well as a low-cost micro-electromechanical system (MEMS) inertial measurement unit (IMU) mounted on the machine table. It is achieved by first synchronizing the data, aligning inertial frames, and calibrating mounting misalignments. Finally, the unknown internal vibration is reconstructed by comparing the rigid and flexible machine kinematic models. Due to the measurement limitation of MEMS IMUs and geometric assembly error, the reconstructed angle is unfortunately inaccurate. Nevertheless, the vibratory angular velocity and acceleration are consistently reconstructed, which is sufficient for the identification with reasonable model simplification. Finally, the reconstructed internal vibration along with the gathered servo data are used to identify the proposed machine dynamic model. Due to the separation of linear and nonlinear dynamics, the vibratory dynamics can be simply considered by adding complex pole pairs into the MIMO LTI system. Experimental validation shows that the identified model is able to predict the dynamic responses of the tracking error and motor force/torque to the input command trajectory and external disturbances, with 2% ~ 6% RMS error. Especially, the vibratory inter-axis coupling effect and posture-dependent effect are accurately depicted. Overall, this thesis presents a dynamic model-building approach for multi-axis feed drive assemblies. The proposed model is general and can be configured according to the kinematic configuration. The model-building approach only requires the data from the servo system or auxiliary motion sensors, e.g., an IMU, which is non-intrusive and in favor of industrial implementation. Future research includes further investigation of the IMU measurement, geometric error identification, validation using more complicated feed drive system, and applications to the planning and monitoring of 5-axis machining process.
Cite this version of the work
Chia-Pei Wang (2023). Virtual Model Building for Multi-Axis Machine Tools Using Field Data. UWSpace. http://hdl.handle.net/10012/19129