ECE 1 Integration of miniature marine robots for environmental sensing

Save this PDF as:
 WORD  PNG  TXT  JPG

Size: px
Start display at page:

Download "ECE 1 Integration of miniature marine robots for environmental sensing"

Transcription

1 Summer Engineering Research Internship for US Students (SERIUS) Host Departments: Department of Electrical & Computer Engineering / Tropical Marine Science Institute, Acoustic Research Laboratory ( / ( ECE 1 Integration of miniature marine robots for environmental sensing Recent technology development has enabled a swath of small marine robots. Surface vehicles are the simplest form of such a vehicle. But despite their advantage of requiring minimal operation logistics, they are limited in their endurance and the number of sensors they can carry. We are interested in putting together a fleet of heterogeneous surface vehicles for cooperative operations. It would allow us to survey a reasonably large area within their individual endurance limits while each of them could carry different sensors to support the missions. The student will help us to set up the hardware and software of the surface vehicles towards this objective. The student would have the choice of working on either Teledyne Z boat, Clearpath robotics Kingfisher, or a miniature surface robot developed in house. The student would also have the option of working together with another group of students who will be setting up an aerial drone for join operations with the marine robots. We encourage motivated students that are passionate in robotics, interested in hardware and fluent in programming to join us in this exciting endeavour. Computing, electronics and computer engineering A report on the system architecture, set up procedure and field testing result of the robots.

2 No. of participants able to host 1 s Assoc Prof Mandar CHITRE Mr KOAY Teong Beng Familiarization in object oriented programming, C++, Java, Familiarization with Linux environment, ROS

3 ECE 2 Automatic feature extraction from acoustic recordings to detect events of interest Long term acoustic monitoring in typically involves large amount of acoustic data to scan through to look for events of interest. These recordings mostly contain high energy, impulsive ambient noise, hence detecting or identifying acoustic events of interest within these recordings is a challenging research question. This project involves extraction of features/descriptors from long term acoustic recordings obtained in Singapore waters. The features can be used to indicate events of interest that are scattered sparsely throughout the recordings. We currently possess a large dataset of such recordings, and are trying to identify methods to detect such events. Feature extraction can be done through several approaches. One promising technique to achieve this is by using machine learning. Machine learning has been shown to be very effective in many problems including acoustics, image and speech processing. The project would explore the use of unsupervised techniques like principal component analysis or autoencoders to compress the bulky acoustic data into a smaller number of features. These features can then be used to identify events of interest within the recordings. Computing, electronics and computer engineering A report on the performance of an algorithm in acoustic event detection in noisy ambient acoustics. No. of participants able to host 1 s Assoc Prof Mandar CHITRE Mr KOAY Teong Beng Dr Hari VISHNU Familiarization on Python Signal processing background is beneficial Machine learning experience is a plus

4 Host Department: Department of Electrical & Computer Engineering ( ECE 3 Analog spintronic devices Spintronics is a device technology that utilizes electron spin as information carriers, which uses the magnetoresistance effect in magnetic tunnel junctions as a means of converting between voltage and spin information. Recent experiments have demonstrated a material stack in which oxygen ions can be electrically injected and expelled from a magnetic layer, which changes the magnetic properties of the magnetic layer. This pens up a new avenue to realizing a new type of analog spintronic device. In this project, we will use micromagnetic simulations to explore different device structures that leverage electrical control of magnetic properties to implement novel electronic device behavior. Students are expected to develop methodologies to evaluate the use of their proposed device structures in new circuits, and validate their ideas. to achieve at the end of 8 week attachment Physics, Electrical Engineering, Materials Science & Engineering Preliminary simulation results in understanding the use of electricalcontrol of magnetic properties to implement analog spintronic devices. Asst Prof Kelvin FONG Xuanyao Knowledge of Python and MATLAB Familiarity with analog and digital circuit concepts (e.g., opamps, logic gates, etc.)

5 ECE 4 Characterizing flexible transition metal oxide devices Flexible electronics device technologies are revolutionizing our imagination for future electronics, especially for smart display and health monitoring systems. In this regard, flexible transition metal oxide device technology is a promising candidate that can implement i) scalable, high speed (ns order) and ultralow energy (fj order) nonvolatile multi level memories in simple structures; ii) thin film transistors that are needed for CMOS style logic gates; and iii) new device structures that are well suited for computing schemes (such as neuromorphic computing) that leverage the unique analog characteristics of the device technology. In this project, students will gain first hand experience working with researchers who are fabricating device structures based on transition metal oxide technology. Students attached to the project team are expected to develop and validate device characterization methodologies that can enable useful device models for studying the underlying device physics and their implications on the electrical behavior of the device. Physics, Electrical Engineering, Materials Science & Engineering Characterize a transition metal oxide device for device modeling Asst Prof Kelvin FONG Xuanyao Familiarity with oscilloscopes, waveform generators and signal (eg programming analyzers skills, prerequisites, reading list, etc)

6 ECE 5 Exploring transition metal oxide devices Transition metal oxide device technology is being investigated as an enabler of flexible electronics. Researchers are actively developing techniques to implement, in transition metal oxide device technology, a) thin film transistors that can be used to build CMOS style logic gates; b) scalable, high speed and ultralow energy non volatile multi level memories; and c) new device structures that are well suited for computing schemes (such as neuromorphic computing) that leverage the unique analog characteristics of the device technology. In this project, students will aid the development of transition metal oxide technology. Students attached to the project team are expected to develop and validate computer models of transition metal oxide devices for studying the underlying device physics and their implications on the electrical behavior of the device. Physics, Electrical Engineering, Materials Science & Engineering Develop a physical model for a transition metal oxide device and run detailed device simulations to evaluate the device operation Asst Prof Kelvin FONG Xuanyao Familiarity with COMSOL and MATLAB Knowledge of finite element analysis will be helpful

7 ECE 6 of multi port non volatile embedded memories Non volatile memory technologies such as spin transfer torque magnetic RAM (STT MRAM) have the capability for ultrafast write operations as fast as SRAM. However, to achieve such fast write speeds either lead to extremely high write energy consumption or breakdown and failure of the memory device. Hence, the write performance is sacrificed to maintain device reliability and keep write energy within acceptable bounds. However, multi port designs avoid issues with write operations blocking accesses to the rest of the memory array. Thus, the memory can service access requests at much faster speeds than the write performance. In this project, students are expected to learn about the operation of STT MRAM as well as future genres of MRAM (such as spin orbit torque MRAM and voltagecontrolled MRAM). We will then develop multi ported designs based on these memory technologies, and evaluate their energy consumption, performance, and suitability for embedded memory applications Electrical Engineering, Computer Engineering Proposed multi port designs of spin orbit torque and voltage controlled MRAMs. Asst Prof Kelvin FONG Xuanyao Familiarity with SPICE simulations, and analog and digital circuit concepts (e.g., opamps, logic gates, etc.)

8 ECE 7 Neural networks for device compact modeling Future electronic systems are likely to be based on a hybrid of device technologies. Device compact models are crucial in helping electronics designers identify the most promising applications for the corresponding device technology. Existing physical device compact models are insufficient for studying circuits and subsystems implemented using many emerging device technologies due to their complexity and difficulty in calibrating to experimentally measured device characterization data. Recently, neural networks (NNs) has emerged a powerful tool that can learn nonlinear input output relationships. Thus, NNs offer a promising solution for rapid prototyping of device compact models from any device characterization data. Since the NNs hides the complicated device physics, NN based device compact models can enable fast simulations of circuits and subsystems that are implemented from the corresponding devices. Students attached to this project can be expected to develop of NNs algorithms to model electrical behavior of advanced transistor devices, and incorporate the model into SPICE simulator for circuit design and simulation. Physics, Electrical Engineering, Computer Science & Engineering Develop a method to train a neural network to model the device characteristics of an electronic device Asst Prof Kelvin FONG Xuanyao Familiarity with MATLAB and Python Knowledge of neural networks will be helpful

9 ECE 8 Neural networks for model parameter extraction Compact models for semiconductor devices and VLSI circuits have been widely studied in the microelectronics industry. For the BSIM transistor model, more than a hundred device model parameters need to be extracted from device characterization data to successfully calibrate the model. In the procedure adopted by industry for compact model parameter extraction, several well known deficiencies, such as poor convergence and difficulty with simultaneous multi objective optimization have been encountered. Recently, neural networks (NNs) has demonstrated the ability to efficiently learn the relationships between sets of data. Thus, NNs is promising alternative to enable fast compact model parameter extraction from device characterization data. Students attached to this project can be expected to develop NN algorithms for compact model parameter extraction from simulated device characterization data based on BSIM4/EKV, and to compare with the results obtained by using the standard extraction procedures. Physics, Electrical Engineering, Computer Science & Engineering Develop a method to train a neural network to extract compact model parameters from the device characteristics of an electronic device Asst Prof Kelvin FONG Xuanyao Familiarity with MATLAB and Python Knowledge of neural networks will be helpful

10 ECE 9 Skyrmionic devices for future computing schemes Skyrmion is a new spintronic phenomenon that can implement futuristic computing schemes such as the bio inspired neuromorphic computing. In this project, we will use micromagnetic simulations to explore different device structures that leverage skyrmions to implement novel electronic device behavior. Students are expected to develop methodologies to evaluate the use of their proposed device structures in new circuits, and validate their ideas. Physics, Electrical Engineering, Materials Science & Engineering Preliminary simulation results in understanding the use of skyrmionic devices as artificial synapses in the neuromorphic computing framework. Asst Prof Kelvin FONG Xuanyao Knowledge of Python and MATLAB Familiarity with analog and digital circuit concepts (e.g., opamps, logic gates, etc.)

11 ECE 10 Decoding the Brain Recent developments in neural recording technologies have made it possible to record from hundreds of individual neurons in the brain. This is a major advance that allows the use of brain signals to control prostheses with large degrees of freedom. It also enables investigators to study the neural code used by populations of neurons to represent and process information in the brain. In this project, we will analyze data recorded from the frontal cortex of awake, behaving monkeys to understand how populations of neurons in different areas respond in a working memory task. We will investigate different neural codes (Bayesian probability, information theory, partial directed coherence, etc.) to understand how information is processed and transformed from one area to another. Students will get to learn to work with large neural data sets, correlate neural data with the behavior of animals, program in Matlab, and perform large scale data analysis on a High Performance Computing cluster. Electrical and Computer Engineering, Bioengineering/Biomedical Engineering, Computer Science, Neuroscience, Psychology Write Matlab/Python code to analyze data and visualize results. Asst Prof Shih Cheng YEN Familiarity with Matlab, data acquisition, signal processing, and statistics.

12 ECE 11 Wireless Bioelectronic Devices Miniaturized wireless bioelectronics promise new ways to treat disorders through precise modulation of physiological parameters. This project focuses broadly on the design, assembly, and testing of a miniaturized wireless device. Participants may focus on one or more of the following: (i) design of a miniaturized wireless stimulator, (ii) electromagnetic modeling of wireless powering, and (iii) experimental characterization of devices. Electrical Engineering, Biomedical Engineering By the end of the attachment, students are expected to: Obtain practical and theoretical experience in bioelectronic wireless technologies an experimental setup to test devices Contribute to ongoing research projects in the group Asst Prof John Ho Interested applicants are invited to look through papers recently published by the group:

13 ECE 12 Mobile Vital Signs Monitoring System One of the limitations of current vital signs monitoring system is that the user is usually limited to movement within a room or the gateway is on body. In this project, we will look at different ways to develop a truly mobile vital signs monitoring system. You will join a team of engineers and fellow students to work on different parts of the project, database, server, mobile apps, localization algorithms, etc depending on your interest. Signal Processing, Computer Engineering 1. Good literature review of the problem. 2. Data collection and analysis. 3. Detailed documentation and report. Assoc Prof Arthur Tay Comfortable with programming, embedded systems.

14 ECE 13 Identifying weak points in modern smart city electricity grids Modern smart grids consist of several types of renewable generation, storage solutions and smart appliances. Central to all these technologies, is power electronics. Due to the increased power electronic presence in the distribution system, there arise serious stability concerns. This project explores new methods to identify the vulnerable nodes in such smart distribution systems so that corrective actions can be taken from the utility side. In this project, the student will use MATLAB/Simulink to implement a vulnerability grading algorithm based on simplified sensitivity analysis. The effectiveness of the proposed algorithm will be demonstrated for several practical distribution system test cases through numerical simulations. The performance of the proposed algorithm will then be compared with conventional sensitivity analysis in terms of computation time and accuracy in detecting the most vulnerable nodes. Power systems Control systems 1. Gain familiarity with MATLAB programming 2. Understand power system modeling and learn about stability analysis tools 3. Evaluate the effectiveness of the proposed vulnerability detection scheme No. of participants able to host 1 Asst Prof Jimmy Chih Hsien PENG Familiarity with simulation in MATLAB/Simulink

15 ECE 14 Develop the next generation smart meter Smart meter is a smart grid technology that provides critical information to power utilities and energy consumers. Its metering accuracy impacts the operational security of the grid and energy management of consumers. Conventional smart meters lack the ability to capture phase angle, which is a critical piece of information for the anticipated dynamic demand response programs. These challenges call for the need of a next generation smart meter. The scope of this project is to improve a prototyped smart meter developed by the research group ( The interns are expected to enhance the phasor estimating algorithm and data communication using the existing FPGA design. The collected data will also be creatively visualized on a webpage or mobile app in realtime. Primary tasks include algorithm design, web/app design, and some hardware development. The ultimate goal is to commercialize the product. Power systems Embedded systems 1. Familiar with programming in FPGA 2. Appreciate the challenges in complying with industrial standards No. of participants able to host 1 Asst Prof Jimmy Chih Hsien PENG Have basic experience in using FPGA; think outside the box mindset.

16 ECE 15 Generative Adversarial Network (GAN) for Control Application GAN is a deep neural net architecture comprised of two networks, pitting one against the other. Its purpose is to go beyond the normal classifiers that are bounded by their training data, and generate new information based on existing data. Currently, GAN is mainly used for image processing, but its fundamental approach of matrix manipulation could be exploited for other engineering applications. In this project, the objective is to apply GAN for analyzing linear control systems, such as electrical grid. The expected deliverable is generating a new control framework and study the learning characteristics for networked systems. Students are expected to have prior knowledge in programming, using software such as Python and Java. This project will be extending on existing M.Sc. research, with focus on conducting more case studies to better characterize the GAN performance. Machine learning Data science 1. Gain in depth knowledge of using GAN 2. Exhaustive analysis of the merits and limitations of GAN for control application. No. of participants able to host 1 Asst Prof Jimmy Chih Hsien PENG Familiarity with Python, and have basic knowledge in data science. Should have a good laptop to do data analytics.

17 ECE 16 Building Networked Microgrids using a Real Time Digital Simulator Future power distribution systems will consist of inverter based microgrids. These microgrids are connected to renewable generation, battery, and electrical loads. The integration of multiple microgrids in a reliable manner is a hot topic in the present research field. In literature, studies have been conducted in an ideal simulation environment, and there is a lack of detailed transient studies. In this project, the aim is build an emulation platform that helps to study the power sharing within a multi microgrid system. The platform will be implemented in RTDS (an industrial grade emulator.) This project continues on the existing work done by M.Sc. students. Power systems Computer simulations 1. Appreciate the benefits of microgrids and renewable technologies 2. Gain experience in using RSCAD software No. of participants able to host 1 Asst Prof Jimmy Chih Hsien PENG Basic power engineering knowledge, and have experience in programming with object oriented languages.

18 ECE 17 Quality Estimation in Manufacturing Processes through Semi Supervised Learning As we move towards the future and Industry 4.0, there is a larger need to automate all areas of manufacturing. A major component of that is identifying manufacturing faults during the process itself, and responding to them with utmost precision and urgency. In this project, we will be looking at data from different areas of a product manufacturing line, which currently includes: near infra red video data from a laser welding operation and numerical/categorical data from an injection molding process. The videos show several runs of a welding operation, and it will usually have some reflections issues and other artifacts due the nature of the mounting platform used. The machine learning models need to be developed in such a way as to account for invariance to rotation and scaling, as well as deal with any image artifacts. Similarly, for the injection molding data, data clean up and preprocessing will be required. As the data is sourced from a real world setting, it would be very difficult to use supervised methods of classification due to the unavailability of fully labeled data. Semi supervised learning, as well as on line learning techniques can be used to fill the gap here. Graphbased methods, auto encoders and generative models have proven successful in the past for dealing with the semi supervised learning problems such as image classification, speech recognition etc. These can be further explored for the purposes of this project. Electrical Engineering, Computer Engineering, Computer Science No. of participants able to host 1 1) Data clean up of original numeric data for injection molding and video data for the welding operation. 2) Explore machine learning models, especially those dealing with semisupervised learning, and imbalanced datasets. 3) Apply one or more of the models to create a robust quality estimation tool. Assoc. Prof. Prahlad Vadakkepat

19 1) Experience with/willingness to learn OpenCV and other image processing tools. 2) Basic knowledge of machine learning architectures. 3) Experience with python preferred.

20 ECE 18 Dancing along a song: deep rhythm modelling and its applications for social robots While the current technologies have enabled social robots to intelligently interact with humans, robots still lack the ability to understand advanced concepts that are more abstract and subjective, such as appreciating a piece of music or painting. In this project, the student is to implement a deep learning based algorithm to analyze several nursery rhymes so that the social robot can understand and follow the songs. To demonstrate the effectiveness of the algorithm, movement of the dancing robot is to be designed and the robot will be controlled to dance based on the result of rhythm model. Finally, demonstration of effectiveness of the designed algorithm will be performed on a social robot platform. Electrical engineering Computer engineering (1) Get familiar with Python programming and deep learning techniques (2) Know how to control servo controllers and robots Prof Shuzhi Sam Ge

21 ECE 19 Indoor Navigation System based on Turtlebot TurtleBot is a low cost, personal robot kit with open source software. With TurtleBot, you ll be able to build a robot that can drive around your house, see in 3D, and have enough horsepower to create exciting applications. In this project, you are supposed to build an indoor navigation system. The RGB D sensor and laser range finder are two main information sources onboard. Given the floorplan of the building, the navigation system drives the robot from a start point to a predefined goal in the building. The main challenges include but not limited to: Indoor map representation and storage (both topological and grid) Sensing and information processing Optimal path planning Implementation from theory to reality Electrical engineering Computer engineering An indoor navigation system with a simulative platform and hardware. Prof Shuzhi Sam Ge

22 ECE 20 Social Robot Control Policy Development Based on Fall Detection As for household service robot, fall detection for elders is one of the most significant functionality. It is reported that one out of three older adults (over age 65) falls every year. The elders who fall will suffer serious injuries, such as hip fracture, head traumas. Some may even cause early death. Especially when the elders live alone, late detection of fall will increase the risk of further injury. Thus, a low cost system that can automatically detect falls in home would perfectly address this problem. In this project, students are supposed to build the control strategy based on our previous work. Given the fall occurrence scenario, a learning scheme is going to be designed such as reinforcement learning and etc. to produce an efficient and suitable reaction for social robot control system. A simulation platform will be built to demonstrate the performance of the proposed policy. To complete this project, strong programming capability is required to develop corresponding modules and integrate all of them together. Iterative testing will be carried on the platform to verify the Performance. Electrical engineering Computer engineering Get familiar with C++/C# or Python programming and software Development; Equipped with basic knowledge of machine learning. Prof Shuzhi Sam Ge Students should be with strong programming capability (C++, C#, python) and interest in pattern recognition and robot system.