Other Projects
Research Paper on Predictive Modeling & Correlation Heat-mapping with Big Data Analytics: Invited and Presented at ICBDE 2024 in Oxford, UK.
Abstract:
Bankruptcy presents a significant challenge for businesses worldwide, indicating financial distress and potential failure. Accurate prediction of company bankruptcy is crucial for stakeholders, including investors, creditors, and management, as it enables them to assess a company's financial health and stability. This research paper aims to predict company bankruptcy likelihood using financial performance metrics. The methodology involves leveraging Big Data Analytics with Jupyter Notebook to develop a correlation heat map and to train a predictive model. The dataset used is sourced from the Taiwan Economic Journals, comprising over 600,000 data points collected from 1999 to 2009. It includes nearly 100 input features, such as return on total asset growth, pre-tax income/net sales, equity/total assets, equity turnover, and current ratio. Analysis of the correlation heat map revealed the following noteworthy findings. The metrics with the strongest positive correlation to bankruptcy are debt ratio, liability to assets, and borrowing dependency. Conversely, net income to total assets and net worth/assets exhibit the strongest negative correlation. By utilizing these financial performance metrics and employing advanced analytics techniques, this research contributes to effective bankruptcy prediction and allows stakeholders to make informed decisions. Understanding the factors that contribute to bankruptcy risk can help companies monitor their financial stability and take appropriate measures to mitigate potential failure.
Arduino Based Autonomous Mars Agro Rover
Problem Statement: Design and implement an autonomous dual-system rover equipped with soil analysis and precision farming capabilities for agricultural exploration and maintenance within a controlled Mars biodome.
3 Primary Subsystems: Drivetrain consisting of 4x BO Motors, 2x Ultrasonic Sensors for obstacle and plant detection, 1x Maker ESP32 Board as the acting micron controller board, 1x 12V LiPo battery, 1x Mini-breadboard, 1x LCD Screen. Arm System consisting of a Servo-driven linear actuator, 3D-printed Rack & Pinion, 3D Printed Robotic Arm, Soil Moisture Sensor. Pump System consisting of a Tank for storing nitrogen-infused fertilizer, Nozzle & Water Pump for controlled water release and a Weight Sensor to measure Tank Fullness.
Original Design Blueprint + Decision Matrix (Engineering Notebook)
Early Prototype Testing
Microcontroller Research (Engineering Notebook)
Gantt Chart (Engineering Notebook)
VEX-integrated PID Project
Objective
Develop a complex task for a VEX V5 robot that utilizes a PID (Proportional-Integral-Derivative) control model to perform precise movements or actions. This project will demonstrate the application of calculus, computer science, and physics through robotic automation.
The Robot
We chose to refurbish a retired VEX robot from the Over Under season to be the subject of our PID control model. To make this possible, we had to attach a couple sensors to provide us with data. We also built and attached a robotic arm to the robot in order to determine the effectiveness of a PID on a hang mechanism. Of the numerous VEX sensors available, we decided on the optical encoder, the vision sensor, the distance sensor, the optical sensor and the Inertial Sensor.
Manually Produced CAD file of Robot
Detailed Projet Description
This project requires us to design an algorithm for a VEX V5 robot that uses a PID control system to achieve precise movements or actions. A PID controller will be programmed to continuously adjust the robot's behavior based on real-time sensor feedback using proportional, integral, and derivative calculations. We will develop and fine-tune the PID loop code, leveraging sensor data from the VEX robot. Computer science concepts are integral in successfully bringing this theoretical control algorithm to real-world application in robotics. Calculus will be applied to optimize tuning parameter values and analyze the system's mathematical behavior over time. Concepts from physics like forces, torques, and motions must also be integrated to model and predict the robot's performance.
Training the Vision Sensor
Approximation of error function
After cleansing the data provided by the trained vision sensor, the data was then put into Excel, and a 5th degree polynomial approximation was used. However, when converting that equation to use to calculate the integral and derivative terms, it was found the Excel equation was actually incorrect when graphed on Desmos. As such, the procedure of increasing decimal places reported by Excel was used to arrive at a accurate representation in Desmos.
Conclusions
Our graphical visualizations told us that the original gains (integral not pictured) are the most optimal for this application, with the most weighting being towards the proportional component to increase the speed of the response.
Errors with increased D gains: The accumulated error is quite large because the process variable only slowly returns to the setpoint (the x axis). This is because the dampening effect of the D term (the derivative of e(t) when e(t) is decreasing towards the x axis is negative, and as such, the overall output of the PID controller is decreased, making the correction take a longer time.
Errors with increased P gains: The oscillations are quite significant because the effect of the P gain outweighs the dampening effect of the D gain. As such, the system continues to provide a very large corrective action that it cannot dampen quick enough, causing oscillations.