Sustainable infrastructures

Explaining the science

Sensing, processing and actuation, are the three main steps adopted by recent cyber-physical systems (CPS), where data are usually collected with Internet of Things (IoT) devices. These devices are typically constrained to limited resources, such as small battery and low computation capacity. State-of-the-art machine learning (ML) and artificial intelligence (AI) can aid complex CPS problems, but are heavily dependent on the quality of data they receive.

Fault discovery, diagnosis and recovery for IoT networks

To ensure the quality of data collected by ubiquitous sensing networks, where the devices therein are typically fault prone and the communications between devices are unreliable, an online system is needed which can help to identify and diagnose failures and recover from a faulty state. Different mathematical tools such as statistics, machine learning, and graph theory can be exploited to build such systems. However, these IoT networks are typically constrained by limited resources such as battery and computational capacity. Therefore, lightweight approaches are required to replace traditional methodologies.

Mining time series streams in real time

Most data collected in CPSs are in the form of time series, which is a series of data points indexed (or listed or graphed) in time order. However this time series data usually includes noise and trivial information which may cause bias in post analysis even with state-of-the-art ML and AI solutions. To improve the quality of analytic results, meaningful information can be extracted from their raw format by using analytic tools such as motif discovery, feature extraction and visualisation. By extending these approaches, new solutions, which can handle data in an adaptive manner, are proposed to support mining tasks on time series streams.