No description available.

Evaluating the social impacts of engineered products, or effects products have on the daily lives of individuals, is critical to ensuring that products are having positive impacts while avoiding negative impacts and to learning how to improve product designs for a more positive social impact. One approach to quantifying a product's social impact is to use social impact indicators that combine user data in a meaningful way to give insight into the current social condition of an individual or population. However, determining social impact indicators relative to engineered products and individuals in developing countries can be difficult when there is a large geographical distance between the users of a product and those designing them and since many conventional methods of user data collection require direct human interaction with or observation of users of a product. This means user data may only be collected at a single instance in time and infrequently due to the large human resources and cost associated with obtaining them. Alternatively, internet-connected, remote data collection devices paired with deep learning models can provide an effective way to use in-situ sensors to collect data required to calculate social impact indicators remotely, continuously, and less expensively than other methods. This research has identified key principles that can enable researchers, designers, and practitioners to avoid pitfalls and challenges that could be encountered at various stages of the process of using remote sensor devices and deep learning to evaluate social impact indicators of products in developing countries. Chapter 2 introduces a framework that outlines how low-fidelity user data often obtainable using remote sensors or digital technology can be collected and correlated with high-fidelity, infrequently collected user data to enable continuous, remote monitoring of engineered products using deep learning. An example application of this framework demonstrates how it can be used to collect data for calculating several social impact indicators related to water hand pumps in Uganda during a 4 day study. Chapter 3 builds on the framework established in Chapter 2 to provide principles for enabling insights when engaging in long-term deployment of using in-situ sensors and deep learning to monitor the social impact indicators of products in developing countries. These principles were identified while using this approach to monitor the social impact indicators of a water hand pump in Uganda over a 5 month data collection period. Chapter 4 provides principles for successfully developing remote data collection devices used to collect user data for determining social impact indicators. A design tool called the "Social Impact Sensor Canvas" is provided to guide device development along with a discussion of the key decisions, critical questions, common options, and considerations that should be addressed during each stage of device development to increase the likelihood of success. Lastly, Chapter 5 discusses the conclusions made possible through this research along with proposed future work.