Licentiate Thesis Defense-Sania Partovian

Abstract:

Connected Internet of Things devices are becoming more powerful, yet the challenge of effectively managing them to prevent failures remains ongoing. In some scenarios, such as with devices produced by a single company, it is possible to use the same fault detection provided by the manufacturer and follow the instructions for resolving the threat. In other cases, such as with devices produced by different companies, the heterogeneity of devices necessitates a more detailed and complex approach to fault detection. Log files, which are records of events or processes generated by a device’s software or hardware, are crucial for monitoring device behavior.

It is important to consider the diversity of log files and data to detect threats, identify their root causes, and provide effective solutions. In the realm of troubleshooting interconnected Internet of Things devices, current solutions predominantly address homogeneous device environments, which limit their scalability and adaptability to diverse device types and configurations. Instead, a more flexible approach is needed; one that can accommodate a variety of connected devices while minimizing reliance on specific company instructions. One such method is smart troubleshooting, which involves a 4-stepcycle which includes prevention, detection and diagnosis, recovery, and evolution of threats.

Given these premises, the ultimate goal of this research is to define a smart troubleshooting approach based on log files and product information. By leveraging a generalized methodology, this approach seeks to enhance the management of internet of things systems in complex, multi-manufacturer environments.

This thesis focuses on a systematic review of log files and the state of the art in troubleshooting methodologies. During the research, the scarcity of publicly available log files for troubleshooting purposes was identified. Consequently, a method was proposed for generating synthetic log files using generative adversarial networks. The proposed methodology leverages these log files along with product information to enhance smart troubleshooting.

To validate the approach and gather industry feedback, questionnaires and interviews were conducted. Following this, machine learning algorithms will be employed to implement and refine the proposed method. By leveraging a generalized methodology, this approach seeks to improve the management and fault detection of internet of things systems in complex, heterogeneous environments.