Abstract
Increased adoption of "digital manufacturing" and "industry 4.0" initiatives to transform traditional shop floors into agile factories can generate large pool of sensor data and machine signals. Devices such as Industrial Internet of Things (IIoT) and other sensors maybe used to capture data before, during and after manufacture for condition monitoring and process control. However, the acquisition of such voluminous amount of data may pose significant challenges to manufacturers, from operational, technical and financial perspectives. With analytics and storage as the focal point of collecting these data, relevant data can be stored for further analysis to extract vital information for visualisation and/or informed operational decision making. There are numerous advantages in the implementation of cloud computing services for storage and processing of large volumes of data and several large cloud service providers are targeting manufacturers with custom-built platforms e.g. Siemens MindSphere to improve productivity. The integration of smart devices to manufacturing systems will create a consortium comprising of computing nodes which provides flexibility in sensor data management on an edge-cloud network. As such, data may be processed and/or stored on different nodes of the network, thereby, reducing the volume of data/computational task offloaded to the cloud. Firstly, the thesis identified and investigated extensively the factors (criteria) that may affect decision-making in sensor data management such as cost, technical performance and compliance with legal regulations, alongside their sub-criteria.Secondly, the thesis developed a decision-making model intended to aid manufacturers decide where to store and process the captured sensor data while optimising these factors. A deterministic decision-making model was developed using a fuzzified hybrid Multi-Criteria Decision-Making (MCDM) method; Fuzzy Analytic Hierarchy Process (FAHP) to assign weights to the criteria and Fuzzy Technique for Order of Preference by Similarity to Ideal Solution (FTOPSIS) to rank the alternatives and deciding where each data stream should be stored and/or processed on a distributed IIoT architecture.
Finally, case studies were developed to represent real-world applications from industrial partners. Results show that the model achieved comparable decisions to experts’ recommendations when designing a processing and storage solution. Sensitivity analysis was then conducted to show how the model responds to a change in criteria weight, to ensure the model is not static. What-if analysis were conducted to ascertain how a change in performance rating for some criteria may result in a change in the final decision on where to store and process data, such as increase in amount of data captured per second and change in the market price of data storage etc,. Having shown the model to be applicable and adaptable, further development is recommended to introduce more factors to support distributed decision-making across the nodes of an edge-cloud network.
Date of Award | 14 Oct 2022 |
---|---|
Original language | English |
Supervisor | Andrew Longstaff (Main Supervisor) & Simon Parkinson (Co-Supervisor) |