Modern value chains are the critical backbone of the world economy. Value chains are complex networks of activities and interactions both within and across organizations of various types. Such activities are essential for creating the various goods, services and products necessary for the sustenance of our daily lives. Some examples are producing raw materials required for industrial manufacturing, designing of engineering products, operations and maintenance of high-value equipment, providing various end-user or customer services, among others. Although these are distinctly different activities, delivering these reliably require the effective collaboration between organizations. One of the main challenges in such collaborative work is the impact of unforeseen events that can disrupt any activity within a value chain. Against this background, digital twins (DT) offer a solution to help organizations make decisions under uncertainties to better manage value chains. DTs are models of real-world systems that can be used to simulate their behaviour for generating real-time or right-time insights about their operations. However, building effective DTs of real-world entities require replicating their behaviour reliably by capturing their parameters and constraints in detail. Doing so, however, is highly challenging because value chain entities are characterised by complex and unique properties that can vary subtly even across related entities within the same family (for example, power plants can use different technologies – steam, natural gas, or waste – for producing electricity and the DTs for the various types of power plants will be very different). Hence, building DTs by enumerating the physical system properties can be difficult to replicate (e.g., even across different but related entities as exemplified before) and scale (e.g., from simpler to larger/more complex entities).
Against this background, this PhD project aims to develop an alternative method to building digital twins to address the limitations of current DT methods. In this context, rapid instrumentation of industrial systems through IoT sensors has made operational data more readily available. Furthermore, recent progress in advanced artificial intelligence algorithms such as deep learning has created promising opportunities to build complex models of real-world systems using a data-driven approach.
More specifically, the PhD project will research current methods and limitations of developing empirical models of value-chain systems using sensor data. It will then design methods for developing behaviour models of multi-level value-chain systems by leveraging multi-class deep learning algorithms or other related frameworks, applied to sensor data. The methodology will be demonstrated by developing representative models of real-world value-chains (for example, those taken from the energy or manufacturing domains, for which adequate open-source data sets can be obtained). It is expected that the research output will contribute towards developing more resilient cyber-physical systems by enabling the design of more robust digital twins which in turn can improve value-chain decision making under uncertainty.