Reducing the uncertainty of thermal model calibration using on-machine probing and data fusion

Various sources of error hinder the possibility of achieving tight accuracy requirements for high-value manufacturing processes. These are often classified as: pseudo-static geometric errors; non-rigid body errors; thermal errors; and dynamic errors. It is comparatively complicated to obtain an accurate error map for the thermal errors because they are influenced by various factors with different materials, time constants, asymmetric heating sources and machining process, environmental effects, etc. Their transient nature and complex interaction mean that they are relatively difficult to compensate using pre-calibration methods. For error correction, the magnitude and sign of the error must first be measured or estimated. Pre-calibrated thermal compensation has been shown to be an effective means of improving accuracy. However, the time required to acquire the calibration data is prohibitive, reducing the uptake of this technology in industrial applications. Furthermore, changing conditions of the machine or factory environment are not adequately accommodated by pre-calibrated compensation, leading to degradation in performance. The supplementary use of on-machine probing, which is often installed for process control, can help to achieve better results. During the probing operation, the probe is carried by the machine tool axes. Therefore, the measurement data that it takes inevitably includes both the probing errors and those originating from the inaccuracies of a machine tool as well as any deviation in the part or artefact being measured. Each of these error sources must be understood and evaluated to be able to establish a measurement with a stated uncertainty. This is a vital preliminary step to ensure that the calibration parameters of the thermal model are not contaminated by other effects. This thesis investigates the various sources of measurement uncertainties for probing on a CNC machine tool and quantify their effects in the particular case where the on-machine probing is used to calibrate the thermal error model. Thermal errors constitute the largest uncertainty source for on-machine probing. The maximum observed thermal displacement error was approximately 220 μm for both X and Z-axis heating test at 100 % speed. To reduce the influence of this uncertainty source, sensor data fusion model using artificial neural network and principal component analysis was developed. The output of this model showed better than 90 % correlation to the measured thermal displacement. This data fusion model was developed for the temperature and FBG sensors. To facilitate the integration of the sensor and to ease the communication with machine tool controller, a modular machine tool structural monitoring system using LabVIEW environment was developed. Finally, to improve the performance of the data fusion model in order to reduce the thermal uncertainty, a novel photo-microsensor based sensing head for displacement measurement is presented and analysed in detail. This prototype sensor has measurement range of 20 μm and resolution of 21 nm.