My name is KuangPing Tseng, graduated from NCKU. My proficiencies encompass Virtual Metrology, Yield Management, Machine Learning, Deep Learning, Statistical Analysis, and various other data science techniques, particularly as applied to intelligent manufacturing systems. Currently, I am actively seeking opportunities in the realm of Data Science and the development of Machine Learning algorithms.
Gender : Male Date of Birth : 1994/10/10 New Taipei City , Taiwan
Development of FAC CIM Big Data-related services and reports for facility management. (R, Python)
Integration of facility and FAB MES information to achieve collaborative operational automation.
Maintenance of other FAC automation-related reports and technical support for issue reporting. (C#)
Support for FAB CIM technology development. (C#)
National Cheng Kung University (NCKU) Institute of Manufacturing Information and Systems - Master's Degree Candidate 2017.07 - 2019.09
Development and full-factory implementation of Automatic Virtual Metrology (AVM) models for semiconductor and machine tool industries. (Matlab, Machine Learning)
Development and implementation of automation programs and services for machine tools. (C#)
Development of AI models for machine tools and design of application services. (Matlab, Python)
Development and implementation of algorithms related to smart manufacturing or statistical process control.
PROGRAMMING SKILLS
Python
Machine Learning Development Deep Learning (Tensorflow) Data Visualization
Signal processing
Django, Flask
Matlab
Signal processing
Algorithm Development
Data Visualization
C#
PLC Connection & Automation
MVC
R Language
Statistical Analysis
Data Visualization
Data Pre-processing
Others
Docker
SPSS Minitab SQL
Git(CI/CD)
PROJECT
Non-Intrusive load monitoring(NILM)
Non-Intrusive Load Monitoring System Design (NILM) Reduce the cost of installing sensors in various circuits/appliances in past energy management systems, NILM utilizes only a single main meter coupled with relevant machine learning (DAE, CNN-DAE) and statistical analysis method. It can estimate the power consumption and states of various common home appliances, achieving a low-cost, efficient alternative for home energy management systems.
Develop and Deployment API & Models of Energy Management System Developed a model upload&refresh system and using Docker Container to deploy backend APIs, including: A. Reading fine-tuned AI models, and provide model prediction&result periodically for energy services. B. Conduct periodic data statistics of connected devices to ensure the proper collection of user device data.
C. For various services, retrieving model computation data from the database, coding Django or Flask APIs to provide customized user reports for the energy management system web report.
Through verification with over 8000 households by Taiwan Power Company, the implementation of NILM in home energy management systems can help individual users save a minimum of 10% on their electricity consumption.
Residential Electricity Abnormal Detection System
Using machine learning and statistical trend analysis algorithms, this system establishes the normal electricity consumption patterns of users and monitors their daily power loads. It can detect the following behaviors:
Analyzing users' long-term electricity consumption trends to identify trend inflection points and historical highs/lows.
Alerting users to relevant abnormal electricity consumption behavior loads and abnormal periods.
The developed Abnormal Detection algorithm has been integrated and applied in the following platforms:
Collaborating with various hardware vendors, we establish MQTT or API servers, write data streaming scripts using AWS Glue, and ensure real-time uploading of user sensor information to the database. This process involves unifying heterogeneous hardware data formats and performing data normalization, achieving a unified and instantaneous data streaming service.
FAC ICCI(Intelligent Control Chart Integrated) Platform in TSMC
I implemented advanced statistical process control techniques inspired by FAB production lines, introducing them to both 300mm and 200mm FAB facilities across Taiwan. This implementation covered a wide range of monitoring elements, with around 1000 or more sensors per FAB, including aspects such as air conditioning, cooling water, gas pipelines, and power supply system. This approach was a significant improvement over the previous static Out-of-Control (OOC) monitoring method.
ICCI monitoring strategy incorporated the following control features:
Recording High/Low values
Notation
Detection Mismatches
Identifying Mean Shifts and Variance Shifts
Early detection of Out-of-Control (OOC) situations
These indicators were utilized to provide early warnings, preventing any negative impact on the manufacturing process.
Integration of Facility and FAB MES Informationin TSMC
Integrated facility and FAB MES information to achieve collaborative operational automation, including:
Automated scheduling and planning of Facility System Preventive Maintenance (PM) by incorporating production line MES equipment data, mitigating impact on online production.
Detection of patterns generated by various maintenance or routine operations in the facility system, along with anomaly detection (e.g., abnormalities caused by component replacements).
Maintenance of automated reports related to Facility Automation Control (FAC) and provision of technical support for incident reporting.
AVM (Automatic Virtual Metrology) Implement for Aerospace Industry and Semiconductor
AVM Introduction
When the product has not been or can not be actually measured for final inspection, the quality of the product is estimated (Y) by using the parameters of the process (X). The online and immediate prediction is made, so we able to achieve the full inspection of all products.
AVM Algorithm 4 modules
Numerical Prediction Module: This module comprises two algorithms, which are Artificial Neural Networks(ANN) and Partial Least Squares (PLS). It aims to predict the quality by given process parameters using these algorithms.
Process Parameter Anomaly Detection: This functionality identifies abnormal process parameters and presents their trends to users for reference.
Reliance Index for Predictions: A Reliance Index is introduced to assess the confidence in predictions. It triggers an alert when there is a significant disparity between the predictions generated by the two algorithms.
Global Similarity Index: The Global Similarity Index is a comprehensive metric considering all process parameters. It detects significant deviations and alerts the user when substantial shifts occur.
CNN(Convolutional Neural Networks) has a powerful effect in the field of image recognition, and then extends many network architectures, such as R-CNN, AlexNet, GoogleNet, etc. This study introduces CNN from image recognition into the field of signal processing and identification, and changes the network input from the picture RGB value to the raw data of the accelerometer (vibration sensor), then identifies and classifies it.
The input signal of this study is the vibration signal of the lathe rotation with a wheel; the output is the wheel dynamic balance prediction value.
Contribution
1.This study modifies the structure of the CNN Model so that it can be applied to signal identification and denoising.
2.Use experimental design and data augmentation methods to overcome the problem of insuficient samples and meet the requirements of deep learning.
3.Improve the traditional MSE Loss Function to make it suitable for angle(radius) prediction and improve the accuracy of the model.
4.This research has been submitted to the IEEE Robotics and Automation Letters
Publications
1. Tieng Hao ; Yu-Yong Li ; Kuang-Ping Tseng ; Haw-Ching Yang ; Fan-Tien Cheng, "An Automated Dynamic-Balancing-Inspection Scheme for Wheel Machining," in IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 2224-2231, April 2020.doi: 10.1109/LRA.2020.2970953
2. 曾廣平(Kuang-Ping Tseng)；劉書安。NILM非侵入式負載監控技術的現況與挑戰。中國電機工程學會(The Chinese Institute of Electrical Engineering) 電工通訊季刊 2023年第二季（六月號）
Patents
APPARATUS AND METHOD FOR ANALYZING ELECTRICAL LOAD, AND APPARATUS FOR MODELING ELECTRICAL LOAD
Patent Application Numbers: 111142783, 2022-186489, 17/989,721
Applied Countries: Taiwan (Republic of China), Japan, United States
Application Date: November 9, 2022
Status: Under Examination
METHOD AND SYSTEM FOR IDENTIFYING OPERATING STATUS OF ELECTRICAL APPLIANCE BASED ON NON-INTRUSIVE LOAD MONITORING Application Numbers: 111134648, 2022-161880, 17/974,494
Applied Countries: Taiwan (Republic of China), Japan, United States
Application Date: September 14, 2022
Status: Under Examination
Career Award
"Presidential Hackathon Excellent Team Award", 2020 (https://official.presidential-hackathon.net/history/2020/2020_01.html)
"Future Digital Awards for Smart Cities & IoT Innovation Carbon Reduction Innovation of the Year", 2023 (https://www.juniperresearch.com/future-digital-awards/smart-cities-iot-innovation)
"Future Digital Awards for Smart Cities & IoT Innovation Urban Smart Grid Innovation", 2023(https://www.juniperresearch.com/future-digital-awards/smart-cities-iot-innovation)
