Dr Rajesh Ransing of the Civil and Computational Engineering Research centre at the
Ransing claimed that the software learns from the information given from past manufacturing components in a way that both human experts and neural networks cannot. He said that the problem with experts is that their knowledge is based on rules, which cannot capture all the combinations, while neural networks have a self-learning ability but fail because they require a large number of examples and cannot extrapolate the information.
Traditionally, if a problem is encountered during casting, rejection analysis is carried out and three or four corrective actions take place to resolve it. If this does not work, further action is taken in a hit-and-miss process that is at odds with the overall objective of a manufacturer searching for a short lead-time.
‘I feel our new technique is best because you are using historical data rather than gut feeling. With this software you can actually identify a cause-and-effect relationship,’ said Ransing.
Casting is one of the most widely used manufacturing processes in numerous industrial sectors, including aerospace, automotive and power generation, with an industry turnover of £2.2bn.
But according to David Critchley, manager of the Cast Metals Federation (CMF),
Critchley said that against this backdrop of strong foreign competition, a tool such as X1Recall could be critically important to the
‘This industry is less automated than most, so is probably 10 years behind the automotive industry, Added to this, many experts are retiring and we don’t have much young blood coming through. This means there is a knowledge gap, which is a big problem,’ said Ransing.
With a £425,000 grant from the Welsh Development Agency, Ransing is using this year to undertake industrial collaborative research to further the commercial viability of his patent.
To this end, Ransing and the CMF have brought together a consortium including Rolls- Royce and seven foundries — which use pressure die casting, investment casting and sand casting — to test X1Recall.
‘For them it will not only help analyse the causes for rejection but also help them optimise their process because the way the software works is unique compared to a neural network,’ said Ransing. ‘It learns from examples, can explain its diagnosis, and can also identify cause and effect relationships.’
Ransing said that the system also learns combinations from previous knowledge; so if a new component is added to the process, the system would detail the optimum process parameters.
Last month Rolls-Royce began testing the software in the manufacture of its turbines, and Ransing is confident that by the beginning of next year he will have a full commercial project.