Grant: Discovery Grants Program - Individual

Advances at different levels of monitoring systems from our team and others have led to ubiquitous communicating sensing devices. Still, fabrication costs, dimensions, and energy efficiency are significant barriers to the adoption of sensing devices for smart monitoring systems across critical application domains. Moreover, the valuable data gathered by these sensors can be harnessed more effectively through automatic anomaly detection (AD). AD identifies deviations from normal behavior, with applications ranging from predicting equipment failures to health monitoring and cyberattack detection. The challenges in developing high-performance efficient AD systems are primarily driven by the limited availability of labeled anomaly data and the necessity for these systems to effectively generalize to rare, novel events that are infrequently observed in most applications. Building on our recent successes, the long-term objective of this research program is to enable AD systems to be efficiently deployed in real-world applications, while being highly scalable and adaptable to new settings or evolving environments. This program will focus on two specific and complementary challenges: improved-sensing communicating devices and the learning models to be used on the data they collect. Since antennas are critical design elements in sensing networks-especially with miniaturization/shape constraints or when battery operation is not sustainable-our team will target the advancement of MIMO antenna designs using additive manufacturing and machine-learning-aided design for compact sensing devices to gather data in AD systems. We will also propose strategies to train reliable predictive models despite limited application-specific data. We will adopt an integrated design approach that targets the advancement of data-efficient learning models based on self-supervised pretraining, as well as scalable continual learning after deployment. Furthermore, we will integrate these advancements into a complete AD system for experimental deployment in an industrial use-case. The outcomes of this research program will lead to efficient networks of wireless sensors to detect anomalies with higher accuracy and minimal training burden. AD is applicable to numerous domains, such as targeted preventive maintenance, failure prevention of machinery and infrastructure, intrusion detection and preventive health treatment before serious or fatal pathologies develop. The strategies at the heart of this program are targeted for success based on our interdisciplinary expertise in both hardware and machine learning development in a wide range of applications. The program will contribute to Canada's competitiveness in advanced manufacturing, security and health of Canadians, and civil infrastructure monitoring-particularly in the context of climate change.