John J. Shi's Home Page

{home}
General: {resume} {publications & presentations}
Experience: {software & web} {teaching}
Research channels: {knowledge engineering} {product process} {supply chain} {health care & biomedical}

Selected Efforts in Knowledge based Engineering

Data Cleansing – Dealing with Missing Data

Date: 01/2003 ~ 08/2003
Employer: University of Cincinnati
Position: Research Associate
Managed two master students to study on the problem of missing attributes in engineering datasets. Various methods such as mean imputation, regression imputation, principal components analysis, neural networks, clustering, and their combinations were explored.

Credit Card Application Approval Data Analysis

Date: 11/2002 ~ 06/2003
Employer: University of Cincinnati
Position: Research Associate
Two datasets were provided by an England research institute. We built neural network model to help the prediction of the card application approval. The 70% accuracy could be arrived. Then the data analysis steps such as discretization, dimensionality reduction, and rule extraction were carried out. With only two parameters left and 8 IF-THEN rules, we built a prediction model with 68% accuracy. Another dataset provided the institute reached similar results.
http://www.eng.uc.edu/icams/research/credit/

Rule Extraction in Ergonomics Questionnaire Data

Date: 03/2003 ~
Employer: University of Cincinnati
Position: Research Associate
30k data samples were given to extract useful information. Techniques to deal with duplication and inconsistency, high amount and dimensionality in the data set were studied. Self-developed AKD Data mining software tool was used to extract If-Then rules from the data. These rules were used to train and diagnose the lifting task workers, which finally could reduce the working hardness/effort of workers, e.g., alleviating lower back pain.
http://www.eng.uc.edu/icams/research/Effort%20Case/

Condition-based Maintenance

Date: 02/2003 ~
Employer: University of Cincinnati
Position: Research Associate
Developed methodology of AMFM (Adaptive Mamdani Fuzzy Model) based on neuro-fuzzy technology. This method could integrate fuzzy rules in a neural network; parameters of the network could adapt to new environment. The software was developed with Visual C++, which is now a off-line system, and would be upgraded into a real-time monitoring system. It had become a powerful adaptive knowledge system, used in several monitoring applications of product manufacturing and food producing.
http://www.eng.uc.edu/icams/research/cbm/

Adaptive Knowledge Discovery

Date: 11/2002 ~ 01/2003
Employer: University of Cincinnati
Position: Research Associate
This ambitious project of “Adaptive Knowledge Discovery” was proposed by Professor Samuel H. Huang and mainly managed by Mr. J. Shi. Three students are involved to design, develop, and test a creative and stable software tool, which could finally enable the knowledge discovery as soon as the industry case data is given. The components in format of DLL are being developed using Visual C++, including normalization, discretization (Chi2), dimensionality reduction (based on statistics), neural network modeling, rule extraction from decision tree and clustering, AMFM rule tuning (Adaptive Mamdani Fuzzy Model), as well as rule base pruning.
http://www.eng.uc.edu/icams/research/akd/

Lorain Tubular Yield Prediction

Date: 10/2002 ~ 01/2003
Employer: University of Cincinnati
Position: Research Associate
Developed a analytic model and a software tool to predict the “yield” from the mill and generate simple rules for better understanding of the process involved: managed a research student to collect data from Lorain Tubular; used statistical analysis such as hypothesis testing of different “plugs” and regression analysis of “yield”; ranked parameters by significance based on the statistic results; and built a neural network model for prediction.
http://www.eng.uc.edu/icams/research/lt/

Marzetti Response Surface

Date: 07/2002 ~ 09/2002
Employer: University of Cincinnati
Position: Research Associate
Developed a computing model for predicting microbial spoilage in dressings and sauces, which could help the company view the impacts of attributes such as pH, acid/moisture ratio, % protein, % sugar, and % salt on the potential spoiling time (weeks). Behind the computer model (a program called response surface), it was a neural network developed by Visual C++.
http://www.eng.uc.edu/icams/research/rs/

Kraft Oscar Mayer Sensor Control Feasibility Study

Date: 07/2002 ~ 09/2002
Employer: University of Cincinnati
Position: Research Associate
Developed a computer program - Curve Fit, which used a curve based on 3-7 parameters to fit the data measured in a Microwave Spectrometry instrument. The parameters calculated by the program would be used to predict the fat, moisture and protein percentages.
http://www.eng.uc.edu/icams/research/kraft/

Clustering-based Neural Network Construction

Date: 05/2002 ~ 06/2002
Employer: University of Cincinnati
Position: Research Associate
This project was to use several clustering techniques such as subtractive clustering and k-nearest neighbor to classify datasets from engineering and service industrial applications. The neural network was used to carry out function approximation and forecasting based on the clustering centers found. I managed a PhD student to develop a Visual C++ program to implement these methodologies.

Quadrilateral Mesh Generation

Date: 02/2002 ~ 10/2002
Employer: University of Cincinnati
Position: Research Associate
The project of “Automated FEA/CFD Hexahedral Mesh Generation Using an Integrated Neural Network/Rule-based Method” was carried out in Intelligent CAM Laboratory with the collaboration of Parker Hannifin Corporation, sponsored by Ohio Aerospace Institute. Expert knowledge in hexahedral mesh generation was extracted from experienced users, using an innovative neural network/rule-based approach. The knowledge extracted was documented, verified, and then computerized to develop a prototype software tool. The tool (QMesh) was used in place of human experts to automatically generate block topology for a given geometry, which is then interfaced with a commercial grid generation software tool (GridPro/az3000) to generate and optimize the final hexahedral/quadrilateral meshes.
http://www.eng.uc.edu/icams/research/mesh/ with access code “ucicams”

Neural Network Architecture Optimization

Date: 02/2002 ~ 10/2002
Employer: University of Cincinnati
Position: Research Associate
This project of “Neural Network Architecture Optimization” tried to facilitate the automation of neural network modeling procedure. One big issue was studied: How to determine the number of neurons in the middle layers of a neural network? Both network growing and pruning technologies were testified in a few cases. A clustering-based network architecture optimization approach (NNOpt) was proposed based on the research. Then, associated program was developed using Visual C++ 6.0. The program contains classes of BP network, clustering, data processing, statistics, etc.
http://www.eng.uc.edu/icams/research/akd/Documents/NNOpt%20report.doc

 

Copyright Jun Shi, 2005. Any comments could be sent to johnjshi@gmail.com.