Concrete Defect Detection Using BiLSTM and Instantaneous Frequency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current non-destructive testing methods for concrete structures face challenges in accurately detecting internal defects due to high computational costs, time requirements, and reliance on human expertise, particularly when dealing with long sequences and external noise interference.

Innovation Solution

An artificial intelligence-based system using a bidirectional LSTM model processes data with long sequences, incorporating a signal detection unit and data processing unit to generate frequency signals and instantaneous frequency data, which are used to determine defects through deep learning, minimizing computation time and improving classification accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of stationary object

If traditional non-destructive testing methods (ultrasonic, magnetic, potential methods) are used to detect internal defects, then measurement coverage is improved, but measurement precision deteriorates due to external noise interference and difficulty in interpreting results

Engineering Contradiction:
Improvemeasurement coverageVSAvoiddefect detection accuracy
Core Design Contradiction:
Area of stationary objectVSMeasurement precision

Solution Approach 1:

The patent segments the measurement process into multiple frequency components through Fourier transformation of the reflection signal. By analyzing specific frequency ranges corresponding to different defect depths and types, the system isolates defect-related signals from background noise, thereby improving measurement precision while maintaining broad coverage

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary processing layer (signal processing unit performing Fourier transformation and frequency analysis) between the raw reflection signal and the defect determination. This intermediary extracts meaningful frequency characteristics that serve as a bridge, converting noisy time-domain signals into interpretable frequency-domain features for accurate defect detection

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If core sampling method is used to accurately detect defect locations, then measurement precision is improved, but loss of time increases due to coring operations and repair work required

Engineering Contradiction:
Improvedefect location accuracyVSAvoidinspection and repair time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the mechanical core sampling process with a non-destructive signal-based detection system. By using impact excitation and reflection signal analysis, the system obtains precise defect location information without physical coring, eliminating the time-consuming mechanical extraction and subsequent repair operations while maintaining high measurement precision

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary defect detection and localization before any physical intervention is needed. By accurately identifying defect positions through frequency analysis of reflection signals, the system provides advance information that eliminates the need for exploratory coring operations, thereby saving significant time in the overall inspection and repair process

Inventive Principle:
Principle #10Preliminary action

3Reliability

If deep learning models process long sequence data for defect detection, then reliability is improved, but loss of time increases due to computational requirements

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts only the essential frequency characteristics from the reflection signal through Fourier transformation, isolating the most relevant features for defect detection. By taking out and focusing on specific frequency components rather than processing entire long-time sequences, the system reduces computational burden while maintaining high classification accuracy through targeted feature analysis

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the time-domain reflection signal into the frequency domain through Fourier transformation, changing the parameter representation from time sequences to frequency spectra. This parameter change converts long temporal sequences into compact frequency distributions, reducing the dimensionality and computational complexity while preserving the essential information needed for reliable defect classification

Inventive Principle:
Principle #35Parameter changes

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

The system enables rapid, accurate detection of internal defects in concrete members by leveraging bidirectional LSTM models and instantaneous frequency data, reducing training time and enhancing classification accuracy, allowing for real-time, wide-range non-destructive testing.

Implementation Method 1

detects a reflection signal by excitation applied to a concrete member

Methodology Applied
Scientific EffectReflection: Reflection

Implementation Method 2

generates a frequency signal from the impact echo signal, and extracts instantaneous frequency data from the frequency signal

Methodology Applied
Scientific EffectSignal processing and Fourier transformation:

Data Source

PatentEP4647752A1System for detecting internal defects in concrete members on basis of artificial intelligence
Publication Date: 2025.11.12 SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION
  • EP4647752A1 patent drawingFigure 1
  • EP4647752A1 patent drawingFigure 2
  • EP4647752A1 patent drawingFigure 3

AI summary

The present invention relates to a system for detecting internal defects in concrete members on the basis of artificial intelligence, wherein, when defects are determined by using deep learning in non-destructive inspection of concrete structures, a defect occurrence determination model is constructed by using a bidirectional LSTM model, and thus, the system is not only suitable for processing data having long sequences but can also improve the accuracy of classification through deep learning and minimize computation time.