MEC202

Stress and irregular lifestyles often aggravate skin problems such as acne, pigmentation, and signs of premature aging. Light therapy masks provide a convenient and efficient solution that allows them to easily perform skin care in their daily lives.

This project aims to

• • Apply deep learning methods to automatically analyze the user’s skin condition through face images.

• • Develop portable phototherapy mask which integrates multiple phototherapy modes such as red light, yellow light and purple light to deal with various skin problems.

• • Recommends personalized phototherapy plans.

The prototype should use multi-wavelength LEDs, large dynamic range, and variable spectrum to achieve multi-mode phototherapy. Meanwhile, automatic analysis of skin conditions is performed to meet personalized skin treatment needs. This intelligent personalized recommendation function is a significant innovation of this design, which can adjust the treatment plan according to the actual needs of the user, thereby improving the effect and efficiency of phototherapy.

Digital healthcare has improved the efficiency and quality of medical services by integrating sensing technology, thereby enhancing the medical service in specific fields. The application of digital healthcare in fracture treatment has yet to be fully explored. Many patients with fractures may suffer from sequelae such as wrist pain and limited mobility due to lack of follow-up, failure of the splint, and other reasons.

This project aims to design a digital Traditional Chinese Medicine (TCM) splint for treating distal radius fracture, which applies sensing technology in TCM orthopedic methods to address existing shortcomings, such as the inability to accurately monitor the splint’s tightness and the skin’s condition by integrating multi-sensors to monitor the pressure, temperature, and humidity of the radius fracture site in realtime. Experimental data should be collected to prove the effectiveness of the proposed digital splint.

During stroke rehabilitation or other neurorehabilitation therapy, hand function training is usually the most challenging stage and the last stage to regain the independence of living. This project will focus on the design of a hand rehabilitation robot targeting at training the hand stretching and gripping functions, independent moving abilities of fingers, separation motion between fingers. Both passive and active versions of such robots can be considered. For a passive version, stretching function is necessary as the patients in the early and middle rehabilitation stages usually suffer from hypertonia in their hands. They can hardly open their palms and fingers independently. For an active version, the robot will be powered by electrical motors or other actuators, and multiple functions as mentioned above should be considered. Besides, grip sensor, haptic sensor, bend sensor, vibration sensor and other interactive sensors can be integrated to the hand robot for evaluating the rehabilitation process, optimizing real-time assistance and providing multiple sources of feedback to patients. Furthermore, this robot can be integrated into a wrist rehabilitation platform, which will provide bilateral wrist rehabilitation with VR games. The combination will provide visible training guidance and facilitate the recovery of ADL function.

Training of activities of daily life (ADL) will be much helpful to functional recovery of stroke patients. However, a large amount of the patients can hardly complete training tasks without the help of therapist or assistive devices. A portable and wearable upper limb robot can assist stoke patients with such ADL rehabilitation training by providing adjustable assistance on specific joints like shoulder. The target of this project is to design various VR games that will work together with a shoulder rehabilitation robot. The games should be designed according to ADL training content and encourage the use of different joints including shoulder, elbow, and wrist. Both single arm training and double arm training can be considered. Besides, the function of evaluation should be considered like evaluating the moving ranges of arm or joints, maximum moving speed. The games can be designed independently with software Unity and will run on a real upper limb rehabilitation robot or even be tested in clinical trials.

With the advancement of global energy transition and the increasing target share of renewable energy, electrochemical energy storage technology has become increasingly important in power systems. Lithium-ion batteries, due to their high energy density, long lifespan, and environmental friendliness, have become the core technology of energy storage power stations. However, the safety and operational reliability of lithium-ion batteries face numerous challenges, such as overcharging, over-discharging, cell inconsistencies, and thermal runaway, which can lead to severe safety risks and performance degradation. Therefore, developing efficient fault diagnosis and early warning methods is of great significance for ensuring the safe and stable operation of energy storage stations.

This project focuses on the development of a lithium battery fault prediction and early warning system based on deep learning. By integrating lithium battery operational data with advanced fault diagnosis algorithms, the project aims to improve the accuracy and real-time performance of fault detection. Specifically, the project combines deep learning algorithms and Battery Management Systems (BMS) to predict and detect anomalies in operational indicators such as voltage and temperature. Through temporal models, recurrent neural networks (RNN), and further comparisons, a generative algorithm is developed. The goal is to build a fast and reliable fault early warning system, providing technical support for the efficient operation of energy storage power stations and offering theoretical guidance for the design of future energy storage systems.

With the global energy transition and the rapid development of renewable energy, lithium-ion batteries have become a core technology in energy storage systems and electric vehicles due to their high energy density, long lifespan, and low self-discharge rate. However, as lithium-ion batteries are highly complex, nonlinear, and time-varying systems, accurately estimating their State of Charge (SOC) faces numerous challenges, such as battery aging, temperature variations, and complex dynamic operating conditions. These factors impose higher demands on the reliability and safety of Battery Management Systems (BMS), making the development of more precise, efficient, and robust SOC estimation algorithms of great significance.

This project focuses on the development of SOC prediction algorithms and system for batteries based on deep learning. By introducing advanced time series models such as Long Short-Term Memory (LSTM) networks, the project aims to address issues in traditional SOC estimation methods, including model complexity, low computational efficiency, and significant impacts from temperature and aging. LSTM networks can capture the long-term dependencies in battery data and, combined with dynamic characteristic data of lithium batteries (e.g., voltage, current, and temperature), achieve high-accuracy SOC predictions. Additionally, data preprocessing and model optimization strategies are employed to enhance the generalization ability and noise resistance of the algorithm. The ultimate goal is to build an efficient and reliable SOC estimation system, providing theoretical support and technical assurance for the safe operation and performance optimization of energy storage batteries.

Fire warning and monitoring technologies have become essential components of public safety. However, traditional fire detection technologies (such as smoke detectors and temperature sensors) often suffer from delayed detection, high false alarm rates, and limited applicability, making them insufficient for fire monitoring in today’s complex scenarios. Automatically analyzing fire signals in videos or images is essential.

• To enrich the dataset and improve the performance of fire detection, a Generative Adversarial Networks (GAN)-based fire image synthesis model and system should be developed to address the issue of insufficient fire image samples. By combining smoke and fire characteristics with engineering scene images, the system can generate diverse and realistic fire image samples. It can simulate fire images in scenarios involving various combustible materials in engineering facilities and even generate fire images in non-fire scenarios.
• A GAN-based fire image synthesis model with deep learning object detection models, such as YOLOv8, should be designedto develop a highly generalized fire image detection model. Training the detection model on the synthesized fire samples can ensure better adaptability to diverse fire scenarios and enhance its accuracy in real-world applications.

Fire monitoring in large spaces faces numerous technical challenges, especially in the real-time processing and analysis of multi-channel video streams. These spaces are typically equipped with a large number of high-definition cameras, making the efficient collection and processing of multi-channel video streams a pressing problem to solve. On the other hand, the training of fire detection models requires a large volume of high-quality, accurately labeled samples, and how to quickly label and generate fire samples that meet the requirements of real-world scenarios has become a key point. Therefore, building an integrated system for video stream acquisition, streaming, sample annotation, and fire detection holds significan practical value.

This project focuses on:

• Real-time acquisition of 25-camera video streams (Can start from 1 camera)

To address the issue of high camera resolution, Hikvision SDK will be used for efficient video stream processing. A display interface will be designed to simultaneously showcase video feeds from 25 cameras, with a feature that allows users to enlarge any selected video feed. The interface will also include “Start Collection” and “Stop Collection” buttons to enable users to control the start and stop of video collection.

• Video streaming

The video streams collected from the 25 cameras will be streamed to a designated storage space on the server. This process will require the development of both a streaming client-side program and a server-side program to ensure stable and efficient transmission of the video streams to the server.

• Sample annotation functionality

A manual annotation tool will be developed on the backend server for labeling a portion of the fire sample data. The labeled samples will be used to train a model based on the YOLO series of networks. Once training is completed, the trained model will be applied to detect fire-related events in the collected video. Detected fire images and their corresponding bounding boxes will be saved in YOLO annotation format to facilitate further analysis and usage.

Multi-port converters are widely used in DC microgrids, electric vehicle charging and discharging systems, and renewable energy grid integration due to their high power density, high efficiency, and bidirectional power transfer capability. The flexible management of multiple energy ports, such as integrating photovoltaic (PV) panels, batteries, and DC loads, enhances system integration and improves overall efficiency. However, optimizing the stability of multi-port converters to further enhance system performance remains a key point.

This project aims to design a fault detection circuit for a multi-port converter and compare its efficiency under different parameter configurations. By analyzing the location of faulty switches and their corresponding waveforms, the operational status of the system during faults can be better understood. Based on these insights, this project will propose a fault detection method and a fault-tolerant control strategy. The findings of this project have broad application prospects in the fields of renewable energy systems, energy storage management, and electric vehicles

Project Aim

This project aims to develop a virtual reality (VR) platform tailored for textile designers, allowing them to interact with fabrics and materials in an immersive 3D space. The VR environment will enable designers to explore textiles based on their texture, color, weight, and movement without the need for physical prototypes.

A key feature of the platform is the integration of real textile data from the Materialliance database, enabling realistic simulations of fabric properties such as drape, stretch, and texture. Designers will be able to observe how fabrics behave in different scenarios—how they fold, stretch, or respond to light—providing valuable insights for design decision-making.

The platform will incorporate interactive tools for real-time material manipulation, allowing designers to test how textiles move, drape, and layer in various design contexts. Additionally, it will support collaborative work, enabling multiple designers or students to interact, brainstorm, and experiment with materials in the same virtual environment.

Project process

The development of the VR platform will be student-led, guided by industry experts and university tutors. Students will gain hands-on experience in:

• VR development and interaction design
• Textile simulation and material physics
• Game design principles for engaging user experiences

The platform will undergo iterative user testing with designers, allowing them to explore multiple fabric choices within a virtual workspace. This approach will significantly reduce the time and costs associated with traditional prototyping, accelerating the design and decision-making process.

By bridging advanced digital tools with material innovation, this project seeks to revolutionize textile design workflows, enhance creativity, and promote sustainable practices in the fashion and textile industries.