STEM Research Symposium

AI Explainability through Explicit, Implicit and Relative Reasoning

Joshua Stapleton, Kenny Gao, and Dan Grissom, Ph.D.

Thanks to modern advances in computation, artificial intelligence is being used to solve increasingly-complex problems by generating increasingly-complex machine learning models. While machine learning is used throughout industry to provide significant value to both business owners and customers, solutions often come in the form of “black box” predictors with little transparency, not due to malicious intentions, but due to a lack of understanding. Ubiquitous usage of machine learning solutions throughout society demands novel support structures for explainability, such that the biases being used in any machine learning model can be revealed by a third party with little-to-no input from the model’s creators; to this front, this work proposes a novel, ethically-minded and human-focused approach to provide AI explainability through explicit, implicit and relative reasoning.

Gas Storage Monitoring System

Natalie Palos, Dallas Gray, Will Cook, Sarah Hernandez, Leslie Torres, James Yeh, PhD

This research project utilizes Internet of Things and LoRa modulation techniques to streamline the process of pinpointing the location and severity of gas leaks, specifically in gas pipeline transportation. This project consists of a network of nodes configured in a daisy-chain topology which communicate using LoRa and a cloud based database where sensor readings are stored. According to Debra Wunch et. all in their article “Quantifying the loss of processed natural gas within California's South Coast Air Basin using long-term measurements of ethane and methane,” the South Coast Air Basin, with a population of about 18 million people, emits approximately 413,000 tonnes of methane and 23,000 tonnes of ethane annually. These numbers are severely exacerbated by gas leaks which this project seeks to address. Each node is composed of a gas sensor(s), battery, solar panel and the WiFi and LoRa capable LoRa32 microcontroller. These nodes are to be placed along the gas pipeline at a specific distance level from the central data collection node, and will only forward data from another node if the data comes from a node with a higher distance level. Collision avoidance and collision detection algorithms with appropriate back-off will be utilized to prevent any data from being lost. Once the data reaches the central data collection node, that node will then store the data in the cloud hosted Firebase database. The Gas Monitoring System will provide enough information to determine when and where there is a gas leak, and the utilization of this project will streamline the process of locating and stopping gas leaks and will ultimately prevent the adverse effects of undetected gas leaks.

Gesture-Controlled Wireless Inventory Management System

Will Cook and Hsi-Jen (James) Yeh, Ph.D.,

Azusa Pacific University, Department of Engineering & Computer Science

This research project uses Internet-of-Things hardware and custom software to create a gesture-controlled wireless inventory management system. The two main products from the project are a wearable, gesture-controlled barcode scanner for recording and accounting for items in the inventory, and a cross-platform desktop application for the management of the inventory, both of which can be used in stores, warehouses and laboratories. The barcode scanner consists of two pieces of hardware connected via a cable. The main unit, a self-contained Wi-Fi microcontroller known as an M5Stick-C, is worn on the wrist like a watch, and the camera module, an M5Stick-V, is worn around two fingers. The camera module captures 2-3 frames per second and runs a basic algorithm to detect barcodes and QR codes. If a code is detected, the camera module decodes it into a string of characters, then sends the characters to the main unit. The main unit displays the decoded string of characters to the wearer, and then prompts the wearer to do something with the barcode. Depending on how the wearer moves his or her wrist, different actions are performed, ranging from adding a new item to the lab database to updating the quantity of an item in the database. The client application is built using JavaScript with React and Electron JS, which allows it to run on all three major desktop operating systems: Windows, macOS, and Linux. It allows a user to add, modify, and delete items in the lab inventory, which is cloud hosted with Firebase. The user can request a restock of any item directly from the app. The user can also search the inventory database by item name or barcode number. The app can also “reverse lookup” barcodes, which allows the user to auto-populate the item’s name, price, and description, as well as a photo of the item.

Singular-Axis Attitude Control System

Isaac Benitez and Aisha Chen, Ph.D.

Azusa Pacific University, Department of Engineering and Computer Science

Most Cubesat satellites require an attitude control system to orient and maintain its orientation. There are different ways to approach the problem but one of the most common ways to solve this engineering problem is by using reaction wheels. Off the shelf components for Cubesats can range well into the thousands of dollars. The purpose of this research is to lay down a small foundation for which future APU Engineering Senior Design Project students can build upon. The system employs a single reaction wheel system that will orient the cubesat prototype according to user input. The input will be received wirelessly to simulate a ground station-to-satellite communication system.