Metal spinning is a sheet forming process to produce axisymmetric products, but its commercial operation still depends on a dedicated mandrel which determines the shape of the product, and skilled craftsmen to control the working tool. In Flexible Asymmetric Spinning (FAS) the mandrel is replaced with three numerically controlled internal rollers, thereby removing the setup time and cost associated with producing the dedicated mandrel. However, if FAS could also be automated, the setup time and cost could be reduced further. This thesis focuses on three elements which need to be in place for the automation of FAS: the automation of the internal rollers; compensation for springback; and toolpath design to prevent failure.

Typically, automation requires a process model. To automate the internal rollers, a process model which predicts the effect of the internal roller position on the workpiece shape would be required – but as FAS is a novel process, no such models exist. To compensate for springback, a model of workpiece shape is required. To prevent failure, a model of the two modes of failure – wrinkling and tearing – is needed. For offline automation, these should be accurate models – but accurate models of both workpiece shape and failure are too slow to make this feasible. For online automation, fast, approximate models can be used – measurements of the product can be fed back in order to compensate for the model errors. However, a literature review showed that no models exist for workpiece shape or failure which are both fast enough for online use, and detailed enough to give information on how tool actions should be changed. This is why FAS has not yet been automated.

In previous work, the internal rollers were positioned through trial-and-error and only a straightwalled cup was successfully produced. In this work, a laser line scanner is installed to measure the workpiece shape online, and one of the internal rollers is positioned at the point where the workpiece just begins to diverge from the target shape. This prevents overlap with the target shape, and allows a greater range of products to be made.

Springback is typically prevented in conventional spinning by pressing the material hard against the mandrel. This is not possible in FAS due to the force limits on the internal tools. However, in FAS it is possible to move the working roller inside the target shape to compensate for springback. The laser line scanner is used to measure springback and calibrate a simple elastic cantilever model of springback online. By using this model to calculate how far to move the tool inside the target shape, springback errors are reduced by 75%.

Two approaches to toolpath design to avoid failure are investigated: Firstly, a finite horizon control system – where failure is checked for only for a short time into the future – is tentatively demonstrated using a slow but accurate finite element (FE) model, but this is too slow for industrial use. However, with a faster, linear-elastic model, the control system is too conservative and fails to produce the final product. Secondly, an empirical approach is investigated: a series of trials are carried out with a parameterised toolpath. The result is a tentative set of rules for toolpath design which may provide the basis for a future control system.

Overall, this thesis makes steps towards the automation of internal rollers, compensation of springback, and design of toolpaths to prevent failure in FAS. With further work to extend the control system developed here to automate all three internal rollers and to verify the robustness of the springback compensation system, any conventional spinning machine could potentially be replaced by an FAS machine – with the toolpath of the working roller designed manually, as it currently is in conventional spinning. Yet the tentative sets of rules on toolpath design also open the door to a potential automatic toolpath generation system, and further work should begin by testing the robustness of these rules with changes in material and geometry. Then, with some likely extensions, they could be embedded into a working control system to fully automate FAS.