Modelling and in-process monitoring of abrasive belt grinding process

Automation and self-monitoring implementation of manufacturing processes will support the development of interoperable ecosystem relevant to the Industry 4.0 concept. Among many industrial cases, abrasive belt grinding is a tertiary machining process used to achieve desired surface quality and to machine off features such as burrs and weld seams. Manufacturers are in the need of an ability to closely monitor and optimise the performance of abrasive belt grinding processes to meet tight tolerances. The abrasive belt grinding process is highly nonlinear due to the complexity of the underlying physical mechanisms, some of which remain unknown. Existing research in the literature on in-situ tool wear prediction were primarily focused on hard tools, but limited effort can be found on that of compliant belt tools. Although many advanced machining cells are equipped with belt grinder and robotic manipulators for surface finishing, industries still rely on skilled operators to manually remove weld seams using belt sanders. Self monitoring of such a dynamic process in industrial robot cell environment is essential in having a fully automated system. This research aims to model the robotic abrasive belt grinding process in dry conditions appropriate for monitoring purposes. The first part of this thesis discusses the influence of the process parameters on material removal and surface quality in abrasive belt grinding process. Interpretation of Taguchi's Design of Experiments (DoE) experimental results revealed that abrasive grain distribution on backing material has significant influence on material removal and surface quality. Subsequently, a systematic approach to mathematically model the belt grinding process using regression techniques based on soft computing is presented. The second part of the thesis deals with real-time monitoring of the belt grinding tool life. Predicting belt tool life helps to determine whether it is under-utilised, overused or it is due for replacement. Unlike other rigid abrasive machining tools, in abrasive belts the grains are not regenerated. The influence of grain wear on material removal mechanisms namely cutting, ploughing and rubbing were investigated with single grit scratch tests and Acoustic Emission (AE) sensor reading analysis. Having understood the effect of abrasive grain wear on belt grinding performance, a methodology to virtually monitor the coated abrasive belt tool life in real time with the help of physical sensors and machine learning classifiers is developed. In the last part of this thesis, an automated weld seam removal method is proposed. The method offers a real time endpoint verification system for weld seam removal using accelerometer, force and vision-based sensors along with machine learning and deep learning algorithms. Expectedly, this will reduce unnecessary costs and also increase the safety level of operators. In general, the proposed modelling and virtual metrology techniques will add values to the entire manufacturing process, in particular to those involving abrasive belt grinding, and will comply to Industry 4.0 objectives.Doctor of Philosoph