Integration of On-Machine Surface Metrology and Machining Process Prediction

As the demand for manufactured parts increase tremendously due to the short life span of products as well as mass production in this era; one of the most challenging problems faced production and quality departments is to ensure that the quality of manufactured parts is not compromised over quantity.

In order to avoid excess downtime during quality assessments, on-line or on-machine inspection is becoming preferred over off-line inspection. For dimensional inspection, on-machine probing can be adopted to reduce the need for manual or CMM based inspection. For typical surfaces produced on CNC machine tools used in advanced manufacturing industries, there is currently no on-machine measurement solution. To ensure high surface quality as well as high throughput on a typical shop floor; this thesis presents two approaches to enhance the chances of right-first-time manufacturing.
The first is a new methodology that exploits the characteristics of a 2D laser line scanner for fast, on-machine areal surface measurement at the micro-scale level. A commercial low-cost laser triangulation instrument is improved and utilised to obtain a high resolution and wider measurement area suitable for the identification of a range of areal roughness parameters on machined parts produced using a typical milling process.

Crucially, the traceability for the new methodology is also established. This includes reports of the measurement uncertainties associated with each of its metrological characteristics (MCs) and the combined uncertainty of each of its axes. The MCs that were considered include measurement noise, residual flatness, linearity deviation, perpendicularity deviation and amplification coefficient. A comparison of the novel technique with conventional lab-based surface metrology instrument shows a high correlation (99 %) between parameters as well as its ability to identify key features on a machined part such as scratches. The inspected surfaces had Sa (0.3 − 6 μm) which covers most of the advanced manufacturing industries such as aerospace, automotive and civil nuclear.

The second approach involves the development of a novel method for enhancing surface quality and machining parameter prediction on the manufacturing shop floor. A proposed artificial neural networks (ANN) model is utilised to predict the quality of the surface of manufactured parts based on the machining conditions as a preliminary or ‘pre-process’ tool before manufacturing. Pilot investigations leading to the selection of the appropriate parameters other than only Sa to be used as the output of the model is conducted. Also, other investigations that improve the robustness of the developed model to ensure that accurate prediction can be made were conducted.

Finally, the proposed model is shown to compare favourably to alternative models in literature with different network architectures, training and learning algorithms. The model was validated for a range of conditions and was confirmed to have an accuracy of Sq (91%), Sal (93%) and Sa (91%) even for process parameters that were outside of the range used for training the model.