CMMI-EPSRC: Tackling new simulation and optimization challenges towards self-organizing manufacturing digital twins

The manufacturing sector is currently going through a digital transformation to create smart factories, with the potential to bring higher productivity, resilience and efficiency. One technology that is key to improving the factory performance is a Digital Twin. A digital twin is a digital representation of the factory, which keeps up-to-date by mirroring what happens in the real factory. If the digital twin is based on a computer simulation model, it could be used to predict how the system will operate in the short-term (predicting multiple performance indicators such as throughput and inventory levels). This is turn would allow the digital twin to be used as a decision support tool, offering potential short-term actions to improve system performance. Whilst the technological developments are being made towards a digital twin, there are still a number of barriers that prevent digital twins being implemented by many manufacturers. This research aims to reduce some of these barriers, particularly some of the statistical issues associated with their development and maintenance. The first issue relates to determining whether the digital twin is accurate enough for its predictions and suggested solutions to be trusted. Due to high levels of uncertainty and the ever changing nature of the manufacturing system, this is not a straightforward problem. Our research will develop a method that allows errors in the digital twin predictions to be detected. If errors are found, it is important that a digital twin be calibrated to ensure its predictions are in line with the real system. One way to do this is to change the input parameters of the simulation model. These are usually estimated from factory data, but for many manufacturers, the data required is not available in sufficient quantities to give good estimates. The second research goal is to develop a calibration methodology to edit the input parameters to align the digital twin and the factory. The third research goal is to develop a way of determining which combination of actions should be taken to solve short term problems, such as machine failure. In this case, there are often a set of actions that could be taken (replace a tool, reallocate the workforce, reschedule the work order), sometimes in combination with other actions. The digital twin could be used to predict how each combination of actions will perform, and thus select which is best. However, testing all combinations could take too long, so this research will develop a method to effectively use the digital twin to find the best solution to a problem. To increase the utility of the digital twin, all the validation, calibration and optimisation must be done quickly. Otherwise the digital twin will quickly become out of date and its solutions will come too late to be useful. However, running the simulation many times will take considerable computing time. The final goal of the research is to make use of parallel computing to speed up all the processes developed in the first three goals.