Implementing ML Model for Predicting Building Longevity
Using Crack Detection
Guide : Dr Vignesh U(53145)
A REPORT submitted by
Devavrat Patel Harshit chudasama Sarthak Patel
21BCE5873 21BCE1951 21BCE5451
INTRODUCTION
• Artificial intelligence (AI) and deep learning have revolutionized the field of
computer vision, enabling applications ranging from facial recognition to medical
diagnostics However, the adoption of these methods often comes with a critical
challenge: lack of interpretability It becomes essential to understand why a model
makes certain predictions, especially in safety-critical domains such as crack
detection for structural integrity analysis
Problem Statement
• Deep learning models, especially convolutional neural networks (CNNs), have
shown remarkable success in various domains such as image classification, object
detection, and segmentation However, these models are often considered "black
boxes" due to their lack of interpretability, making it difficult to understand the
rationale behind their predictions This lack of transparency raises concerns,
especially in critical domains like healthcare, finance, and autonomous systems,
where explainability is crucial for trust, accountability, and compliance
Research Objectives
Enhance Model Interpretability
• Use explainable AI (XAI) methods like Grad-CAM or SHAP to make EdgeCrackNet-based crack
detection more transparent.
Improve Trust and Reliability
• Ensure predictions can be justified with visual explanations to reduce false positives/negatives.
Optimize AI for Safety-Critical Applications
• Balance high accuracy with interpretability for real-world structural assessments.
Develop an Explainable AI Framework
• Integrate interpretability techniques into crack detection models for better decision-making.
Proposed Algorithm - EdgeCrackNet
1 EdgeCrackNet is ideal for image-based tasks such as crack detection due to its
ability to extract spatial features from images
2 Basic approach involves: Input -> Preprocessing -> Convolutional Layers ->
Pooling Layers -> Fully Connected Layers -> Output
3 Activation function: ReLU (Rectified Linear Unit)
4 Loss function: Cross-Entropy for classification tasks
Proposed System Introduction
The proposed system aims to detect structural cracks in
buildings and predict their longevity using machine learning
techniques. Cracks are a primary indicator of structural
degradation, and their early detection is vital for safety and
maintenance planning.
• The system processes input images through preprocessing steps
like grayscale conversion and noise removal. It then uses
machine learning models—KNN, SVM, DT, and Edge Crack
Detection—to classify cracks and analyze their severity. The
outcome assists in estimating the building's health and long-term
durability.
Proposed System
Diagram
List of Modules
1 Image Acquisition Module
2 Preprocessing Module
3 Crack Detection Module
4 Feature Extraction Module
5 ML Classification
6 Longevity Prediction
7 Accuracy Evaluation
Image Acquisition
Module Preprocessing Module
• The Image Acquisition • This module prepares the
Module is responsible for acquired images for analysis
gathering high-quality by applying essential
images of building surfaces, transformations such as
either through existing grayscale conversion,
datasets or real-time resizing, noise removal, and
capture using cameras or normalization. These steps
drones. The quality and enhance image clarity,
resolution of the images are standardize input
crucial, as they directly dimensions, and ensure that
impact the accuracy of crack the crack detection model
detection and further receives clean, consistent
analysis. data for accurate results
Crack Detection Module
• This module uses the Edge Crack Detection model to
identify and localize cracks within the preprocessed
images. It analyzes pixel-level patterns and edges to
accurately segment cracked regions. The output is a
binary mask highlighting the presence and extent of
cracks, serving as the foundation for classification and
longevity prediction.
Feature Extraction
Module ML Classification Module
• The Feature Extraction • This module classifies the
extracted features using
Module is responsible for machine learning algorithms
such as KNN, SVM, and
analyzing detected cracks Decision Tree. Each model
by extracting key properties analyzes the characteristics
of cracks to determine their
that contribute to building type or severity level. The
classification results serve
longevity prediction. It as a key input for evaluating
processes edge-detected the building’s structural
condition and predicting its
images and computes remaining lifespan.
quantitative features
Accuracy Evaluation
Longevity Prediction Module Module
• This module estimates the • This module assesses the
remaining structural lifespan performance of each
of a building based on the machine learning model
classified crack data. It maps using metrics such as
the severity, spread, and accuracy, precision, recall,
density of cracks to F1-score, and confusion
predefined degradation matrix. It helps determine
patterns using predictive the most reliable model for
models. The output helps crack detection and
engineers assess risk levels longevity prediction,
and plan maintenance or ensuring the system's
reinforcement activities effectiveness in real-world
effectively. applications.
Conclusion & Future Work
This project successfully demonstrates how machine learning and deep learning
techniques—specifically Edge Crack Detection, KNN, SVM, and Decision Tree—can be used
to accurately detect cracks and predict the longevity of building structures. The project
effectively utilizes machine learning algorithms—KNN (95.81%), SVM (94.83%), Decision
Tree (96.62%), and Edge Crack Detection (99.8%)—for accurate crack detection in
building structures.
The system automates a traditionally manual and error-prone process, achieving a peak
accuracy of 99.8% using Edge Crack Detection. It offers a reliable, scalable solution for
structural health monitoring and preventive maintenance.
Future Work
• Expansion of the dataset with diverse building materials and lighting conditions for better
model generalization.
• Development of a mobile or web-based interface to make the tool more accessible to
engineers and site inspectors.
• Implement adaptive learning so the model improves as more real-time data is collected.
• Deploy as a cloud-based service or API for structural monitoring companies and smart city
infrastructure.
Guide Approval
Research Paper Submitted Status
References
• Zhang, Z., Liu, Q., & Wang, Y. (2018). Road crack detection using deep convolutional neural network. 2018
International Conference on Image and Graphics Processing (ICIGP).
• Sun, W., Shen, S., & Wang, X. (2020). Image-based concrete crack detection using
Convolutional Neural Networks. Automated Construction, 109, 102958.
• Mangal, M., & Kumar, R. (2021). ML-based crack detection for smart cities infrastructure.
International Journal of Computer Applications, 183(3), 12–16.
• Jangid, P., & Verma, R. (2022). Implementation of deep learning techniques for infrastructure damage detection.
Materials Today: Proceedings, 62, 2436–2442.
• Xu, Y., Su, H., & Wang, Q. (2019). Concrete crack detection based on deep learning and data augmentation. Symmetry,
11(3), 431.
• Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical
image segmentation. Medical Image Computing and Computer-Assisted Intervention, 234–241.
• Dorafshan, S., Thomas, R. J., & Maguire, M. (2018). Comparison of deep convolutional neural
networks and edge detectors for crack detection in concrete. Construction and Building
Materials, 186, 1031–1045.
• Kumar, A., & Singh, V. (2022). Crack Detection in Concrete Structures Using Deep Learning.
International Journal of Computer Applications, 184(17), 34–39.
• Verma, P., & Gupta, R. (2021). Automated Crack Detection in Concrete Walls Using Deep
Learning. International Journal of Engineering Research & Technology (IJERT), 10(3), 415–419.
• Li, Y., & Zhao, X. (2019). Crack recognition system for concrete bridge inspection using
machine learning. Automation in Construction, 103, 256–269.
• Zhang, L., Yang, F., Zhang, Y. D., & Zhu, Y. J. (2016). Road crack detection using deep
convolutional neural networks. Image Processing (ICIP), IEEE, 3708–3712.
• Adam, A., & Ammayappan, L. (2025). "Comparative analysis of deep learning models
for crack detection in concrete structures." Scientific Reports, 15, 85983
• Li, Y., et al. (2023). "Building surface crack detection using deep learning technology."
Buildings, 13(7), 1814.
• Chen, H., et al. (2023). "Deep learning-based automated detection of cracks in
historical buildings." Buildings, 13(12), 3113.
THANK YOU
