Water Content Detection of Red Sandstone Based on Shock Acoustic Sensing and Convolutional Neural Network

Zhicheng Qiu, Yang Liu, Yi Zhang, Xueqi Zhao, Dongdong Chen, Shengwu TuSensors 2025 · DOI

TL;DR

This paper introduces a novel, non-destructive method for detecting the moisture content of red sandstone by analyzing the acoustic signals generated from knocking the rock. By combining Mel Frequency Cepstral Coefficients (MFCCs) for feature extraction with a Convolutional Neural Network-Support Vector Machine (CNN-SVM) model for classification, the researchers achieved an impressive 94.4% accuracy, offering a practical solution for geotechnical engineering applications.

What Problem Does This Paper Address?

Red sandstone is a common building material, but its mechanical properties and stability are significantly affected by its moisture content. High moisture levels can lead to issues like tunnel collapse and reduced bearing capacity, making accurate moisture detection crucial for safety and project quality. Existing methods, such as the dryer drying method or fiber optic sensors, have limitations like being time-consuming, energy-intensive, damaging to the sample, or requiring complex equipment. This paper aims to fill this gap by proposing a simple, economical, efficient, and non-destructive method for detecting red sandstone's moisture content.

Methods

The core of this research lies in a multi-stage approach:

  1. Shock Acoustic Sensing: A simple knocking hammer is used to strike the red sandstone sample. A microphone captures the resulting acoustic signal. This is a non-destructive technique, meaning it doesn't damage the rock.
  2. Feature Extraction (MFCCs vs. CWT): The captured sound signals are processed to extract meaningful features. The paper compares two methods:
  3. Mel Frequency Cepstral Coefficients (MFCCs): This is a standard technique in speech recognition, known for its ability to mimic human auditory perception and capture subtle variations in sound. It involves several steps including pre-emphasis, framing, FFT, power spectrum calculation, filter bank application, and discrete cosine transform.
  4. Continuous Wavelet Transform (CWT): This method analyzes signals in both time and frequency domains, providing detailed information.
  5. Classification (CNN-SVM): The extracted features (MFCCs or CWT) are fed into a machine learning model for classification. The researchers employ a CNN-SVM architecture.
  6. CNN (Convolutional Neural Network): Acts as a powerful feature extractor, automatically learning relevant patterns from the MFCC or CWT data through convolutional and pooling layers.
  7. SVM (Support Vector Machine): Takes the features extracted by the CNN and performs the final classification, aiming to find an optimal hyperplane to separate different moisture content categories. The Adaptive Moment Estimation (Adam) optimizer is used for training the CNN.

Datasets

The study utilized custom-prepared red sandstone samples. These samples were:

  • Cylindrical, with dimensions of 50 x 100 mm.
  • Dried to a constant weight in an oven at 105°C for 8 hours.
  • Then, soaked in water for varying durations (0, 10, 20, 40, 60, 80, 100, 120, 160, 200, and 240 minutes) to achieve different moisture content levels.
  • The experiment involved 100 strikes per sample group under five distinct moisture content conditions (dry, soaked for 10 min, 20 min, 40 min, and 240 min). The sampling frequency for the acoustic signals was set at 51.2 kHz.

Key Insights and Ideas

  • Sound as a Moisture Indicator: The fundamental idea that the acoustic properties of a material change predictably with its moisture content is the bedrock of this research. The paper demonstrates this principle effectively for red sandstone.
  • MFCCs Outperform CWT for Acoustic Moisture Detection: The study highlights that MFCCs, despite being commonly used in speech, are superior to CWT for capturing the subtle acoustic variations indicative of moisture content in red sandstone. This is attributed to MFCCs' ability to better simulate human auditory perception.
  • CNN-SVM Synergy: The combination of CNN for robust feature extraction and SVM for precise classification proves to be a powerful approach, especially for datasets that might not be massive. The CNN learns the complex patterns, and the SVM leverages these learned features for accurate categorization.
  • Non-Destructive and Accessible: The method's reliance on simple tools like a hammer and microphone, coupled with its non-destructive nature, makes it highly practical for field applications where preserving samples or avoiding damage is important.

Key Results

  • High Classification Accuracy: The proposed method achieved a validation accuracy of 94.4% when using MFCCs as features.
  • Superiority of MFCCs: MFCCs significantly outperformed CWT, which achieved an accuracy of 85.6%.
  • Confusion Matrix Insights: For MFCCs, the confusion matrix showed 100% accuracy for samples soaked for 40 and 240 minutes, and 96% for samples soaked for 10 minutes, indicating excellent discrimination at higher moisture levels.
  • Robustness to Impact Force Variation: The study noted that variations in knocking force did not significantly compromise classification accuracy, adding to the method's practical reliability.
  • Performance Across Moisture Levels: Evaluation metrics (precision, recall, F1-score) remained high across different moisture content states, including dry, short-term, mid-term, and fully saturated conditions.

Limitations and Future Work

  • Sample Variability: The authors acknowledge that the shape, size, and geological characteristics of the sandstone samples can influence the knocking sound, potentially affecting detection reliability.
  • Environmental Noise: The study was conducted in a quiet environment. Real-world applications may involve ambient noise, which could interfere with signal acquisition and processing, impacting accuracy.
  • Generalization: While effective in laboratory settings, the model's generalization capability across different environments and rock types needs further validation.

Future Work Suggestions:

  • Expand Dataset: Increase the variety and quantity of samples to cover a broader range of red sandstone types and moisture levels.
  • Noise Reduction: Develop algorithms to mitigate the impact of environmental noise on knocking sound signals.
  • Model Optimization: Further refine the CNN-SVM architecture to enhance feature extraction efficiency and improve generalization.
  • Broader Applicability: Extend the approach to other rock types (e.g., shale, concrete) and materials.
  • Automated Systems: Explore the development of automated intelligent knocking systems for more efficient and accurate detection.

Summary updated

Abstract

In response to the challenge of changes in the physical and mechanical properties of red sandstone when it comes into contact with water during construction projects, this paper proposes a moisture content detection method for red sandstone based on the knocking method. Taking red sandstone as the research object, this study explores a moisture content detection approach by combining the knocking method with Convolutional Neural Network and Support Vector Machine algorithms (CNN-SVM). Specifically, this research involves knocking the surface of red sandstone specimens with a knocking hammer and precisely capturing the acoustic signals generated during the knocking process using a microphone. Subsequently, an effective detection of the moisture content in red sandstone is achieved through a method based on feature extraction from knocking sound signals and a Convolutional Neural Network classification model. This method is easy to operate. By utilizing modern signal processing techniques combined with the CNN-SVM model, it enables accurate identification and non-destructive testing of the moisture content in red sandstone even with small sample datasets. Mel Frequency Cepstral Coefficients (MFCCs) and Continuous Wavelet Transform (CWT) were separately used as features for detecting red sandstone specimens with different moisture contents. The detection results show that the classification accuracy of red sandstone moisture content using MFCCs as the feature reaches as high as 94.4%, significantly outperforming the classification method using CWT as the feature. This study validates the effectiveness and reliability of the proposed method, providing a novel and efficient approach for rapid and non-destructive detection of the moisture content in red sandstone.
BibTeX
@inproceedings{qiu2025water,
  author = {Zhicheng Qiu and Yang Liu and Yi Zhang and Xueqi Zhao and Dongdong Chen and Shengwu Tu},
  title = {{Water Content Detection of Red Sandstone Based on Shock Acoustic Sensing and Convolutional Neural Network}},
  booktitle = {Sensors},
  year = {2025},
  doi = {10.3390/s25237164},
}
Figure