Category: ypec-2024

Categories
Undergraduate ypec-2024

UG16 – Fittix

Fittix is all-in-one mobile app leveraging cutting-edge AI technology to tackle poor fit and style while generating enormous data to aspire new designs. Fittix’s AI technology utilise the concept of computer vision with training on 15000 local body image to allow accurate calculation of the customers’ body size with just two 2D photos (front and side body). The all-in-one app offers seamless, personalized shopping experience with our recommendation algorithm to get the best-fit style with the shortest time. The algorithm based on tags, ranking, weather conditions, seasonal trends and the purchase history of customers to identify the common design, brands, colour mix and purchase reasons (e.g. vacation, working or socialising). The company also promotes sustainable fashion and reduce global textile waste through serious clothes selection (second-hand, recycled materials clothes and outlet) by partnering with global NGOs and brands. With significant market opportunities and dramatic competition. We differentiate ourselves with the unique experiential technology and most care, customised customers app journey. The company’s business model includes partial features subscription fees, sales commissions, advertising revenue, and data insight services, maximising profit and allow easy access to new technologies

Categories
Undergraduate ypec-2024

UG23 – Sustainable Energy: Energy storage and thermal management of solar cells via Phase Change Materials

JarvisKit is a personalized AI assistant that seamlessly integrates into the user’s life, being accessible across multiple devices and communication channels like mobile application, web platform, and smart glasses. Additionally, it can process multiple input modalities: text, audio, and images. By harnessing the power of cutting-edge multimodal language and vision models, including Gemini, GPT4V, and Grok-1.5, JarvisKit enables the creation of sophisticated AI agents that possess a deep contextual understanding of the real world, empowering individuals with instant access to realtime information and personalized chat. To ensure flexibility and adaptability, JarvisKit will offer two deployment options: a local version that leverages smaller models like idefics2 and octopus v3, running on portable devices, allowing for complete autonomy and data privacy, and a cloud-based version that taps into the capabilities of larger AI models like GPT4V, Grok-1.5V and Gemini Pro 1.5, providing unparalleled processing power and scalability. This innovative platform has far-reaching implications for various industries, including education and creative fields, where it can unlock new possibilities and increase productivity.

Categories
Open ypec-2024

ON05 – Equipment Maintenance Model using Artificial Intelligence

The Electrical and Mechanical Services Department has successfully replaced the chiller plant at Tai Lung Veterinary Laboratory as part of its drive to achieve carbon neutrality. It was proposed that Artificial Intelligence (AI) models could be employed to optimise the control logic of the chiller plant to enhance its efficiency. Instead of relying on mathematical models based on the physical principles, the optimisation approach is constructed using data-driven models developed from previous operational data collected from the chiller plant. The performance of the optimisation algorithm was assessed through the utilisation of data from 2023. Statistical analysis was implemented to determine the extent to which the chiller plant improved, with respect to both energy consumption and the coefficient of performance. The chiller plant optimisation with partially observable reinforcement learning (RL) algorithm has been proposed. This technique involves the development of three machine learning models including forecast cooling demand, predict cooling load, and predict energy consumption. The resultant cooling load prediction model and energy consumption prediction model establish a simulation environment for RL agents to explore and it allows them to learn and adapt the dynamic control policies while fulfilling the cooling load requirements.

Categories
Secondary ypec-2024

SS03 – Motion Guardian

According to the Department of Health data, falls, sprains, and collisions are the leading causes of injuries in all types of incidents and household accidents. These issues are particularly pronounced among the elderly and disabled population. Our objective is to provide an accurate, convenient, and cost-effective solution to minimize the occurrence and severity of such accidents, ensuring prompt assistance and safeguarding the well-being of those affected. Furthermore, we aim to identify commercial prospects for our invention, expanding its reach to benefit individuals across various socio-economic strata. To address these objectives, we have developed Motion Guardian, a system that leverages visual machine-learning technology known as Mediapipe for motion detection. Employing the KNN algorithm and incorporating a false alarm prevention mechanism, our solution offers users a combination of precision, convenience, and stability. The versatility of our system allows for easy installation on cameras and mobile devices, with low implementation costs, presenting substantial business opportunities in future-oriented establishments such as senior centers and healthcare facilities.

Categories
Undergraduate ypec-2024

UG28 – MetaMorph: A Personalized 3D Avatar Style Customization Solution Based on Neural Radiance Fields

MetaMorph is a custom website for 3D style transfer based on neural radiance fields, designed to offer users a convenient, low-cost, and personalized virtual avatar experience. By inputting one or more video segments of a person, MetaMorph can reconstruct high-precision 3D models from 2D videos. Users can then customize their virtual avatars in several ways: 1) selecting from pre-designed style templates; 2) using text prompts to drive the customization of avatars in specific styles; or 3) uploading videos of target style models to generate avatars in the desired style. This enables a variety of style transfer effects, ultimately producing a unique virtual avatar that meets personalized needs. MetaMorph’s technology holds significant potential for widespread applications, with its innovative features being highly sought after in fields such as gaming, animation, and virtual reality, providing users with a rich and immersive experience.

Categories
Undergraduate ypec-2024

UG29 – Train energy-saving algorithm and assisted driving design based on Large Language Model(LLM)

This project focuses on the development of a train energy saving algorithm optimisation and a Large Language Model (LLM) based assisted driving system. The system aims to optimise the energy consumption of railway trains and improve the driving experience through a combination of energy saving algorithms and LLM-based interaction systems. The energy-saving optimisation algorithm focuses on building an efficient algorithm based on convex optimisation and mixed integer linear programming. Taking into account factors such as train speed, traction and braking forces, the algorithm aims to minimise energy consumption while ensuring safety and meeting schedule requirements, thus maximising energy efficiency. The interactive system, leveraging the natural language understanding capabilities of LLM, enables real-time interaction with train drivers. This system is capable of not only understanding the train driver’s natural language inputs in real-time, including questions, instructions, and feedback, but also analyzing the train’s operational data and external environmental information promptly. Based on these inputs, it can predict potential risks and challenges, generate precise optimization suggestions, and provide drivers with coping strategies and recommendations.

Categories
Undergraduate ypec-2024

UG27 – A Swallowing Monitoring System and Method Based on Surface Electromyography and Strain Sensing

Swallowing detection devices are commonly used in the process of swallowing rehabilitation, which is often required for stroke patients and those who have undergone laryngectomy. This work developed a non-invasive swallowing detection patch, which has lower production costs compared to existing products. This work developed a stretchable surface electromyography (sEMG) electrode using GW-PA hydrogel, optimizing adhesion and mechanical properties by adjusting glycerol content. It monitors bioelectrical signals long-term and serves as a human-machine interface. The sensor patch includes a stretchable sEMG sensor and a high-sensitivity, high-stability flexible strain sensor. The sEMG sensor used GW-PA hydrogel via screen printing and hot pressing techniques. Graphene was prepared on the surface of PI films using laser induction and transferred to a PDMS substrate for a PDMS-LIG composite strain sensor. Then it underwent surface silanization treatment to enable stable adhesion with the sEMG electrode array, integrating multimodal sEMG and strain sensing. Applied to the throat, the patch uses multimodal sensing for real-time monitoring. Integrated with a wireless device, it transmits sEMG and strain data wirelessly with high data consistency. The CNN model achieved an 86% recognition accuracy in swallowing behavior patterns, validating the effectiveness of machine learning for swallowing rehabilitation monitoring.

Categories
Undergraduate ypec-2024

UG31 – A versatile bionic dexterous hand

This project has designed a partially underactuated bionic dexterous hand that achieves joint flexion with multiple degrees of freedom through a wire-driven mechanism and incorporates pressure sensors at the fingertips along with pneumatic corrugated tubes for flexible gripping. The bionic hand serves multiple purposes: firstly, it can be controlled by receiving electromyographic signals from the human body to assist amputees in performing daily activities through the contraction and relaxation of the hand; secondly, it can be manipulated to execute specific movements by using computer vision to record real-time hand videos, extract skeletal points, and calculate finger joint angles with the MediaPipe machine learning model. Our dexterous hand boasts a low production cost, diverse applications, and holds a broad range of application prospects.

Categories
Undergraduate ypec-2024

UG26 – Autonomous navigation of surgical robots through deep reinforcement learning of simulated environment interaction

This project presents advancements in utilizing artificial intelligence (AI) for precise control of robotic systems in surgical procedures, focusing on close-loop control strategies. Close-loop control, employing data-driven feedback mechanisms such as deep reinforcement learning (DRL) and imitation learning (IL), offers adaptability to unstructured environments but requires significant data for training, posing challenges like data scarcity and privacy concerns. We address these challenges by proposing end-to-end training strategies for motion planning, mimicking human behavior through DRL for autonomous exploration and interaction with the environment. A key focus lies in navigating flexible endoscopes within the gastrointestinal tract, vital for diagnostic and therapeutic procedures. By modeling deformations using finite element methods (FEM) and training control policies through on-policy DRL, we enable autonomous navigation in dynamic soft tissue environments. Our contributions include employing model-free DRL for controlling tendon-driven flexible robots in contact scenarios and integrating intuitive hand gesture recognition for intraoperative intervention, enhancing safety and efficiency. This research underscores the potential of AI-driven control systems in revolutionizing robotic surgery, promising precise and adaptable robotic interventions with improved patient outcomes.

Categories
Undergraduate ypec-2024

UG33 – Succeed in College Entrance Examination

Project Background The College Entrance Examination poses significant challenges for young students. Aiming to provide assistance and guidance, we build our website (www.gaokao985.online), holding the motto “Assist every tiny dream”. Introduction to Functions Our website offers seven key functions. Users can filter universities and majors by attributes like province, university types, major types, etc. to find ideal choices. Besides, the major recommendation interface helps users select majors aligned with their future career goals. Additionally, the mark analysis tool aids in evaluating subject scores, helping users draw up a study plan. Users can share messages on our message board to express opinions and support peers. Furthermore, our real-time chat room allows users to exchange opinions and consult experts. Technical Details Firstly, we used HTML and CSS for website structure and design. Besides, we adopted JavaScript to handle form submissions and facilitate data transfer to the server. What’s more, we employed AJAX to achieve asynchronous data exchange on web pages, which allows web pages to exchange and update data with the server without reloading the entire page. We also utilized PHP and SQL to connect to the server and perform database operations. Session management helps tracked user login state and relevant messages.