Hi. I'm Jiatong Zhang.

With the goal of improving troubleshooting efficiency in distribution center automation systems, our project focuses on incorporating machine learning algorithms into the troubleshooting process. The aim is to quickly classify the type of problem (hardware, sensor, or wiring/power) within the category of electrical issues. This allows engineers to address the problem using the most effective method.

To gather representative data, our team member gained access to the database of an industry-leading company. We tested both deep learning and non-deep learning approaches. The selected models for testing included Support Vector Machine (SVM), Random Forest, Decision Tree, Logistic Regression, K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP).

Among these models, the Random Forest model demonstrated the highest accuracy for problem classification, achieving a training accuracy of 99.58% and a test accuracy of 79%. Notably, the model performed exceptionally well in classifying Electrical Hardware issues, with a precision of 79% and a recall of 85%.

I designed and implemented a graphical user interface (GUI) to improve accessibility and convenience of testing and practical usage. It will provide a prediction based on the suspected features that the user types in. The prediction is based on the Random Forest model.

The project aimed to design a bike rack for a concept car of an automotive company. The design needed to accommodate two bikes, allow easy attachment/detachment, save storage space, and cater to various bike sizes and frame styles. Importantly, the design had to demonstrate Design for Manufacturing (DFM) and Design for Assembly (DFA) techniques.

We began with ideation and conceptual sketches, creating a versatile design that incorporated a rotatable U-part and bottom structure, allowing for easy storage and trunk access. A key feature was the height adjuster, enabling adaptability to different bike sizes and car models. The rack was designed with several components including the structure U part, rack stand, height adjuster, and receiver tube connector, all made of steel for stability. Additionally, the bike holder with elastic rubber straps was made from plastic and rubber to secure the bikes safely.

We utilized Fusion 360 for static stress analysis. Our research determined that the average force exerted by a standard mountain bike was around 32 lbf, which was applied directly on the bike holder in our model. Initial results indicated that the design needed to be improved to provide a higher safety factor when holding two bikes.

After identifying the areas of improvement, we made significant modifications to the design. This involved integrating Aluminum 6061 instead of steel for non-structural components. This choice not only strengthened our design but also reduced the mass from 13.28 kg to 10.22 kg, creating a more efficient bike rack.

The project resulted in a functional, easy-to-use, and adaptable bike rack design that satisfies customer needs. The rack was safe, secure, and efficient, featuring a design ready for manufacturing and assembly.

This project focuses on the development of a mass estimator for drone payloads and its integration into the controller of a CrazyFlie quadcopter. The objective is to address the challenges posed by variable system dynamics resulting from changes in payload weight and mid-flight delivery mechanisms.

Our team successfully designed and implemented a mass estimator using the recursive least squares (RLS) algorithm. The estimator demonstrated accurate detection of paperclip weights with an error rate of less than 15% in offline tests. By incorporating the estimated mass into a PID controller, the project achieved improved trajectory tracking compared to a static PID controller.

In this project, I actively contributed to various tasks, including flying the quadcopter with a joystick, implementing code flashing, finding a suitable mounting mechanism for the magnet, executing autonomous trajectories, collaborating on data logging, reconfiguring the paperclip mounting structure, and setting up autonomous trajectories with data logging for analysis. I worked closely with the team to refine the controller and estimator designs based on the collected data.

The successful outcomes of this project emphasize the significance of the mass estimator in addressing dynamic changes in drone systems during flight. The methodologies and findings can be applied to enhance the performance of existing payload delivery quadcopters.

Implementing traditional sensors on soft robots presents significant challenges. In this project, our team introduce a novel approach by developing a soft robot that seamlessly integrates both sensing and actuation capabilities. Our project involved designing a 3D printed mold, fabricating a pneumatic actuated gripper, and implementing flex sensors to enable real-time measurement of bending angle.

To validate the accuracy and feasibility of our approach, we compared the data collected from Arduino, which reads the flex sensor measurements, with measurements made in AutoCAD. The results demonstrate that the implementation of flex sensors on soft grippers achieves a high degree of feasibility and accuracy.

By incorporating flex sensors, we enable the acquisition of real-time feedback data, which opens up possibilities for closed-loop control over soft robotic grippers. This advancement has significant potential for enhancing the manipulation capabilities of soft robots in various applications.

The project was initiated by a cycling enthusiast with the vision to develop a dual crank bicycle that could be operated in a prone position. The primary aim was to introduce this new riding style into the market at a low cost and as an accessory to standard bikes.

The team developed a unique hand-driven clutch for standard mountain bikes, designed in SOLIDWORKS. The clutch was created to transfer input power effectively and operate as an engage/disengage mechanism. A steering lock was also designed to allow riders to lock the steering handle when using the hand crank, maintaining a degree of freedom for balance. Furthermore, a protective case was added to shield riders from potential injury caused by the additional chain connecting the clutch to the sprocket.

ANSYS was utilized for Finite Element Analysis (FEA), focusing on stress and fatigue failure of the components. Particular attention was paid to the clutch's block tooth and wall sections, as these were identified as potential points of failure. The selection of springs was based on desired spring constants and fatigue requirements.

While the project was halted due to pandemic-related restrictions, the groundwork laid by this team sets the foundation for future teams to improve the manufacturability and marketability of the design. The design's potential for increased power efficiency and solving balance issues will be addressed in future development stages.

The primary goal of this project was to design a robotic cleaner capable of navigating its environment autonomously. The robot needed to avoid obstacles and perform cleaning operations.

We designed the robot structure inspired by 'Cog', a mobile robot we previously worked on. The robot was programmed to wander randomly, capable of moving forward, backward, and turning to either side. Inputs to the robot were taken from Sharp GP2Y0A21YK0F IR Distance Sensor and HC SR04 Ultrasonic Distance Sensor. Outputs included controls for the Gear DC Motor and Mega 2560 Starter Kit Fan Motor, and debugging LEDs. As a rudimentary implementation of the cleaning function, we used the Fan Motor as a brush substitute.

The robot was successful in wandering randomly and its sensors effectively detected distances. The obstacle avoidance function was operational, though not as planned, primarily due to issues with updating the sensor using Timer1 interrupt. Furthermore, due to limitations with the Gear DC Motor and the delay() function used for controls, Timer1 was not fully operational, suggesting a need for a servo-based improvement.

The aim of this project was to develop multiple navigation competencies for a mobile robot named 'Cog'. Leveraging wireless communication and graphical user interfaces, we implemented path planning, path following, localization, and mapping operations using MATLAB and Cog.

The team successfully implemented wireless communication between Cog and MATLAB, designing a user interface to visualize transmitted data. We employed both topological and metric navigation methodologies to maneuver Cog from a starting point to a goal position. Localization tasks were performed using infrared (IR) feedback to identify Cog's location on a predefined map. Lastly, a topological map was created by guiding Cog throughout the environment and collecting IR feedback data.

The project demonstrated that Cog could competently execute wireless communication, GUI operations, path planning and following, and localization and mapping. We evaluated and confirmed the effective application of our programming logics and solutions.

This challenging project provided a rewarding experience for me, particularly witnessing the successful implementation of desired behaviors in Cog. Through this, I gained not only programming skills, but also problem-solving capabilities, and a deeper understanding of robotic theories.