Deep Learning Integration

University of Waterloo
Waterloo, Canada
Graduate Research Assistant
May 2021 - Present

Research on Deep Learning Assisted Damage Characterization for Fiber Reinforced Composites

Graphical abstract for task 4.

Task 4: Deep learning-based CNN model for the qualitative and quantitative characterization of damage in non-crimp fabric (NCF) composites (completed)

Developed a CNN model to effectively identify different damage modes, locate damage initiation, and track damage progression, enabling detailed assessment of damage mechanisms in the material.
Preprocessed and annotated experimental images into six damage-related classes; split data into training and testing sets.
Employed a U-Net architecture with skip connections, dropout, and weighted Dice loss to train and evaluate the model under various hyperparameters, including batch size, class weights, and the number of epochs.
Integrated an automated crack-counting algorithm to enable quantitative damage assessment, supporting more precise evaluation of structural integrity.
Advanced predictive maintenance strategies for automotive and aerospace applications, enhancing structural health monitoring and improving failure prevention.

Reference paper: Erli Shi, John Montesano, Yu Zeng, Khizar Rouf, “Deep Learning Based CNN Model for Damage Characterization in Non-crimp Fabric Composites” (under review, submitted to Composite Science and Technology).

Graphical abstract for task 3.

Task 3: Integrated digital imaging technique for correlation of damage evolution and stiffness degradation under cyclic loading (completed)

Developed a novel in-situ integrated digital imaging technique to simultaneously characterize damage and monitor stiffness degradation in the material subjected to tension-tension cyclic loading.
Enabled automated identification of damage mechanisms, localization of damage initiation, and quantification of damage propagation (e.g., crack length increase) in a single test specimen.
Addressed a key gap in fatigue testing methodologies by integrating digital image correlation (DIC) with in-situ monitoring for real-time damage assessment.
Generated critical insights into fatigue damage mechanisms in NCF composites used in wind turbine blades and automotive components.
Provided foundational data for developing progressive fatigue damage models, including metrics like the number of cycles required to initiate tow cracks and the evolution of tow crack density and length increase.

Reference paper: Erli Shi, John Montesano, “Integrated digital imaging technique for correlation of damage evolution and stiffness degradation in non-crimp fabric composite materials under cyclic loading” Composite Structures (accepted, waiting for doi. July 23, 2025).

Graphical abstract for task 2.

Task 2: Characterization of damage in NCF composites at low temperature using an in-situ digital imaging technique (completed)

Developed an automated in-situ digital imaging technique to characterize damage in the material subjected to quasi-static tensile loading at room temperature (22°C) and low temperature (-50°C).
Developed a custom algorithm to automatically detect the initiation and growth of main observed damage modes, including 90° tow cracks, matrix cracks, and 0° tow cracks.
Assessed temperature-dependent damage evolution using quantitative metrics, including crack density, crack density rate, and crack length increasement.
Validated damage modes and failure mechanisms through optical microscopy; provided a schematic illustration of the damage evolution process.
Contributed to the development of next-generation thermoplastic composite materials, supporting the design of lightweight, recyclable, durable, and impact-resistant structures for automotive and aerospace applications.

Reference paper: Erli Shi, John Montesano, “Characterization of damage in non-crimp fabric glass fiber-reinforced reactive thermoplastic composites at low temperature using an in-situ digital imaging technique” Composites Part A: Applied Science and Manufacturing, vol. 190, p. 108674, Mar. 2025, doi: 10.1016/j.compositesa.2024.108674.

Graphical abstract for task 1.

Task 1: A comparative study on mechanical properties and failure mechanisms of a NCF/reactive thermoplastic composites at room and low temperatures (completed)

Comparatively characterized the mechanical performance and failure behavior in NCF glass fiber/reactive thermoplastic acrylic composite (GF/acrylic) against a conventional epoxy-based counterpart (GF/epoxy) at 22°C and -50°C.
Conducted longitudinal, transverse, and 10° off-axis tensile tests to evaluate tensile and shear properties; employed DIC and post-test optical microscopy to evaluate damage evolution and failure mechanisms.
Provided essential insights for optimizing acrylic-based composites in cold-environment applications.

Reference paper: Erli Shi, John Montesano, “Effect of low temperature on the mechanical properties and failure characteristics of an infused non-crimp fabric glass fiber-reinforced reactive thermoplastic” (under review, submitted to Composite Science and Technology)

Conestoga College
Kitchener, Canada
Professor (Contract), Quality Engineering
Jan 2024 - Dec 2024

Term Summary Report

QUAL8361 - Current Topics in Quality Engineering

Delivered lectures on Industry 4.0 (I4.0), Internet of Things (IoT), Artificial Intelligence (AI), Machine Learning and Deep Learning (ML & DL), and Digital Process Simulation (DPS).
Designed and supervised individual and group projects on ML and DPS, ensuring alignment with current industry trends and demands while providing hands-on experience with relevant tools and methodologies.

QUAL8126 - Data Analytics

Taught core data analytics concepts including data acquisition, data quality check, data cleaning, exploratory data analysis (descriptive statistics, data visualization, correlation & trend analysis), outlier detection, data transformation, and feature engineering.
Developed and implemented quizzes, assignments, in-/off-class practical exercises, and capstone projects to reinforce learning and promote practical application of data analytics concepts.

Project on Data-Driven Mortgage Risk Assessment using Machine Learning

Developed a classification model to predict mortgage default likelihood using real-world datasets from the Freddie Mac repository.
Collected and visualized large-scale datasets in Tableau to explore geographic and temporal trends.
Preprocessed data in Azure ML using Python libraries; performed outlier detection using a scalable, unsupervised framework and applied stratification to ensure balanced state-level representation.
Trained a neural network model optimized for ROC and AUC while minimizing false negatives; used penalty functions and a custom batch sampler to handle class imbalance and improve convergence.
Applied SHAP for model interpretability, identifying key features influencing default predictions.

Interact with Web Model

Project on AI-powered virtual assistant for real-time interaction

Designed and built lifelike virtual avatars using the MetaHuman library to enable natural user engagement.
Developed and deployed conversational avatars with real-time interaction capabilities in both Unreal Engine and Unity, supporting multi-platform performance.
Created a web-based interface to allow seamless real-time interaction and assistance in research and industry applications.
Developed a mobile application integrating augmented reality to enhance user experience through immersive, location-aware visualization.

Deep Learning Integration

University of Waterloo Waterloo, Canada Graduate Research Assistant May 2021 - Present

Research on Deep Learning Assisted Damage Characterization for Fiber Reinforced Composites

Task 4: Deep learning-based CNN model for the qualitative and quantitative characterization of damage in non-crimp fabric (NCF) composites (completed)

Developed a CNN model to effectively identify different damage modes, locate damage initiation, and track damage progression, enabling detailed assessment of damage mechanisms in the material.

Preprocessed and annotated experimental images into six damage-related classes; split data into training and testing sets.

Employed a U-Net architecture with skip connections, dropout, and weighted Dice loss to train and evaluate the model under various hyperparameters, including batch size, class weights, and the number of epochs.

Integrated an automated crack-counting algorithm to enable quantitative damage assessment, supporting more precise evaluation of structural integrity.

Advanced predictive maintenance strategies for automotive and aerospace applications, enhancing structural health monitoring and improving failure prevention.

Reference paper: Erli Shi, John Montesano, Yu Zeng, Khizar Rouf, “Deep Learning Based CNN Model for Damage Characterization in Non-crimp Fabric Composites” (under review, submitted to Composite Science and Technology).

Task 3: Integrated digital imaging technique for correlation of damage evolution and stiffness degradation under cyclic loading (completed)

Developed a novel in-situ integrated digital imaging technique to simultaneously characterize damage and monitor stiffness degradation in the material subjected to tension-tension cyclic loading.

Enabled automated identification of damage mechanisms, localization of damage initiation, and quantification of damage propagation (e.g., crack length increase) in a single test specimen.

Addressed a key gap in fatigue testing methodologies by integrating digital image correlation (DIC) with in-situ monitoring for real-time damage assessment.

Generated critical insights into fatigue damage mechanisms in NCF composites used in wind turbine blades and automotive components.

Provided foundational data for developing progressive fatigue damage models, including metrics like the number of cycles required to initiate tow cracks and the evolution of tow crack density and length increase.​

Reference paper: Erli Shi, John Montesano, “Integrated digital imaging technique for correlation of damage evolution and stiffness degradation in non-crimp fabric composite materials under cyclic loading” Composite Structures (accepted, waiting for doi. July 23, 2025).

Task 2: Characterization of damage in NCF composites at low temperature using an in-situ digital imaging technique (completed)

Developed an automated in-situ digital imaging technique to characterize damage in the material subjected to quasi-static tensile loading at room temperature (22°C) and low temperature (-50°C).

Developed a custom algorithm to automatically detect the initiation and growth of main observed damage modes, including 90° tow cracks, matrix cracks, and 0° tow cracks.

Assessed temperature-dependent damage evolution using quantitative metrics, including crack density, crack density rate, and crack length increasement.

Validated damage modes and failure mechanisms through optical microscopy; provided a schematic illustration of the damage evolution process.

Contributed to the development of next-generation thermoplastic composite materials, supporting the design of lightweight, recyclable, durable, and impact-resistant structures for automotive and aerospace applications.

Task 1: A comparative study on mechanical properties and failure mechanisms of a NCF/reactive thermoplastic composites at room and low temperatures (completed)

Comparatively characterized the mechanical performance and failure behavior in NCF glass fiber/reactive thermoplastic acrylic composite (GF/acrylic) against a conventional epoxy-based counterpart (GF/epoxy) at 22°C and -50°C.

Conducted longitudinal, transverse, and 10° off-axis tensile tests to evaluate tensile and shear properties; employed DIC and post-test optical microscopy to evaluate damage evolution and failure mechanisms.

Provided essential insights for optimizing acrylic-based composites in cold-environment applications.

Reference paper: Erli Shi, John Montesano, “Effect of low temperature on the mechanical properties and failure characteristics of an infused non-crimp fabric glass fiber-reinforced reactive thermoplastic” (under review, submitted to Composite Science and Technology)

Conestoga College Kitchener, Canada Professor (Contract), Quality Engineering Jan 2024 - Dec 2024

QUAL8361 - Current Topics in Quality Engineering

Delivered lectures on Industry 4.0 (I4.0), Internet of Things (IoT), Artificial Intelligence (AI), Machine Learning and Deep Learning (ML & DL), and Digital Process Simulation (DPS).

Designed and supervised individual and group projects on ML and DPS, ensuring alignment with current industry trends and demands while providing hands-on experience with relevant tools and methodologies.

QUAL8126 - Data Analytics

Taught core data analytics concepts including data acquisition, data quality check, data cleaning, exploratory data analysis (descriptive statistics, data visualization, correlation & trend analysis), outlier detection, data transformation, and feature engineering.

Developed and implemented quizzes, assignments, in-/off-class practical exercises, and capstone projects to reinforce learning and promote practical application of data analytics concepts.

Project on Data-Driven Mortgage Risk Assessment using Machine Learning

Developed a classification model to predict mortgage default likelihood using real-world datasets from the Freddie Mac repository.

Collected and visualized large-scale datasets in Tableau to explore geographic and temporal trends.

Preprocessed data in Azure ML using Python libraries; performed outlier detection using a scalable, unsupervised framework and applied stratification to ensure balanced state-level representation.

Trained a neural network model optimized for ROC and AUC while minimizing false negatives; used penalty functions and a custom batch sampler to handle class imbalance and improve convergence.

Applied SHAP for model interpretability, identifying key features influencing default predictions.

Project on AI-powered virtual assistant for real-time interaction

Designed and built lifelike virtual avatars using the MetaHuman library to enable natural user engagement.

Developed and deployed conversational avatars with real-time interaction capabilities in both Unreal Engine and Unity, supporting multi-platform performance.

Created a web-based interface to allow seamless real-time interaction and assistance in research and industry applications.

Developed a mobile application integrating augmented reality to enhance user experience through immersive, location-aware visualization.

Contact

+1 613 816 5891 erli.shi@uwaterloo.ca

​

University of Waterloo
Waterloo, Canada
Graduate Research Assistant
May 2021 - Present

Provided foundational data for developing progressive fatigue damage models, including metrics like the number of cycles required to initiate tow cracks and the evolution of tow crack density and length increase.

Conestoga College
Kitchener, Canada
Professor (Contract), Quality Engineering
Jan 2024 - Dec 2024

+1 613 816 5891
erli.shi@uwaterloo.ca