• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Sunday, November 22, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Convolutional Neural Network Accurately Predicts Particle Sizes in Recycled Stabilized Soil

by
November 21, 2026
in Technology
Reading Time: 6 mins read
0
Convolutional Neural Network Accurately Predicts Particle Sizes in Recycled Stabilized Soil

Convolutional Neural Network Accurately Predicts Particle Sizes in Recycled Stabilized Soil

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Geotechnical engineering researchers have developed a convolutional neural network framework capable of rapidly predicting the particle size distribution of recycled stabilized soils from low-resolution digital photographs, achieving determination coefficients exceeding ninety percent across ninety-five percent of test evaluations. Published in the International Journal of Geo-Engineering, the investigation demonstrates that computer vision can extract grain-size metrics from construction materials composed of clay, paper sludge ash, and blast furnace slag cement without requiring time-consuming mechanical sieving. The deep learning system operates on downsampled 112-by-112-pixel images, converting surface visual patterns directly into mass percentages retained across standardized sieve sizes. Across an evaluation dataset of one hundred eighty soil samples, the artificial intelligence model produced an average mean absolute error of 0.0158 and an average root mean square error of 0.0264. By interpreting optical features from moist, binder-treated particles, the non-destructive method establishes an efficient quality-control pathway for industrial facilities converting urban construction sludge into usable civil engineering fill.

The management of surplus soil generated during urban excavation presents an escalating logistical and environmental challenge in metropolitan regions where landfill capacity remains constrained. Massive quantities of high-moisture mud and soft construction sludge are routinely produced, prompting efforts to stabilize these weak geomaterials with chemical binders so they can be repurposed as structural fill in road subgrades and embankments. Conventional stabilization traditionally incorporates cement-based agents, but the resulting materials frequently display high alkalinity around pH 12 that can disrupt surrounding ecosystems. In response, modern processing facilities have introduced supplementary additives such as paper sludge ash and industrial slag, occasionally paired with carbon dioxide curing treatments to neutralize alkalinity down to levels near 8.6. As applications for these recycled stabilized soils diversify into sensitive civil infrastructure, establishing rapid quality assurance protocols has become indispensable for verifying material integrity prior to site placement.

Despite the need for continuous quality monitoring during soil recycling operations, existing laboratory characterization procedures struggle to balance representativeness with operational speed. Hydraulic conductivity and permeability represent foundational performance indicators for engineered earth structures, particularly in zoned embankments that demand differentiated drainage layers and impermeable cores. Because soil permeability correlates strongly with particle size distribution, grain-size gradation serves as a critical proxy for predicting water movement through compacted fills. Advanced laboratory alternatives, such as laser diffraction and dynamic image analysis, introduce systematic vulnerabilities when applied to recycled geomaterials. Dynamic image analysis requires extensive thermal drying and mechanical dispersion, which can artificially fracture delicate binder bonds, break aggregated clay clusters, and alter in-situ gradation. Furthermore, recycled soil particles frequently exhibit coarse dimensions and cohesiveness exceeding the measurement limits of standard optical dispersion systems.

To construct a rigorous experimental foundation, the research team synthesized controlled recycled soil mixtures in the laboratory using Ao clay, a Japanese geotechnical material classified as clay with a low liquid limit. Standard laboratory testing established that Ao clay exhibited a liquid limit of 40.7 percent, a plastic limit of 23.7 percent, and a particle density of 2.716 grams per cubic centimeter. The soil was blended with distilled water across nine target water contents spanning from 0.1 to 0.9 times its liquid limit. The researchers incorporated two industrial additives: a paper sludge ash-based stabilizer, added at a dry mass ratio of three percent relative to dry Ao clay, and blast furnace slag cement type B, blended at ratios of five or ten percent. The paper sludge ash stabilizer exhibited a particle density of 2.603 grams per cubic centimeter, while the slag cement possessed a density of 3.04 grams per cubic centimeter with slag content between thirty and sixty percent.

Following blending, the soil mixtures were subjected to two distinct curing regimes designed to reflect differing industrial stabilization processes. Under the first condition, moist soil batches were sealed in airtight plastic bags and cured for seven continuous days to allow baseline binder hydration. Under the second protocol, specimens underwent sealed curing for six days followed by twenty-four hours of carbonation curing inside an environmental chamber at twenty-five degrees Celsius under a constant ten percent carbon dioxide atmosphere. The carbonation exposure triggered reactions between carbon dioxide and calcium hydroxide within the soil matrix, precipitating calcium carbonate and lowering soil pH from approximately 12.0 to 8.6. These chemical transformations generated irregular aggregate particles containing open and closed intraparticle voids that expanded as particle dimensions increased. To preserve these fragile moisture-rich void networks, subsequent reference gradation testing omitted standard oven-drying procedures, preventing artificial thermal desiccation from distorting grain geometries prior to mechanical sieve classification.

To assemble an extensive image repository for neural network training, the researchers devised an optical capture platform capable of recording surface textures in seconds. Soil samples representing thirty-six distinct formulation and curing combinations were poured from a height under five centimeters into a tray measuring twenty-four by eighteen by three centimeters without compaction. A digital camera positioned twenty-eight centimeters above the tray recorded RGB photographs of a ten-by-ten-centimeter surface area under paired lighting units that suppressed shadowing. Thirty independent pours were photographed for every formulation, yielding 1,080 multi-size fraction images capturing placement variations. The investigators complemented these full-mixture scenes with single-size fraction images across ten sieve intervals from 0.075 to 26.5 millimeters. Applying spatial data augmentation, the team segmented and recombined these single-fraction photographs into 2,000 two-by-two composite grids and 2,000 four-by-four composite grids, measuring packed densities across each sieve range to calculate precise theoretical mass gradations for four thousand synthetic training scenes.

Rather than deploying parameter-heavy commercial architectures such as VGG16 with 138 million parameters or ResNet-50 with 25.6 million parameters—which demand massive labeled datasets rarely available in geotechnical engineering—the team engineered a bespoke convolutional neural network containing 6.55 million parameters. The investigators avoided general-purpose models like MobileNet, whose pretrained weights derive from natural object photography bearing little resemblance to soil textures. The custom architecture incorporates convolutional blocks and spatial pooling layers that compress feature maps before flattening them into fully connected layers with rectified linear unit activations. The final layer applies a softmax activation across eleven sieve categories, guaranteeing that predicted retained mass fractions remain non-negative and sum to one before post-processing generates monotonic cumulative distribution curves.

Model training utilized the Adam optimization algorithm to adjust network weights against a mean absolute error loss function calculated across all eleven sieve fractions, continuing until validation loss converged below 0.02. When evaluated on the independent test dataset of one hundred eighty unseen multi-size fraction images, the trained convolutional architecture achieved high predictive fidelity, with 171 images—representing ninety-five percent of the test collection—achieving coefficients of determination of ninety percent or greater. Eight test images exhibited moderate accuracy between eighty and ninety percent, while only a single sample registered low accuracy below eighty percent. Across the complete test population, the network sustained an average mean absolute error of 0.0158 and an average root mean square error of 0.0264. Benchmarked against gray-level co-occurrence matrix texture extraction coupled to a multilayer perceptron, which attained high accuracy on only 91.67 percent of the test images, the convolutional model demonstrated superior capability in interpreting the complex reflectance and shadow variations inherent to recycled soils.

Examination of convolutional feature maps revealed how the system interprets grain attributes, demonstrating an ability to infer fine particles smaller than the 0.892-millimeter pixel resolution. Initial convolutional filters responded to sharp boundaries and high-contrast edges typical of coarse aggregates measuring between nineteen and 26.5 millimeters, whereas subsequent filters responded to diffuse surface brightness patterns and pixel density fluctuations associated with fine-grained clay matrices. Beyond predicting complete cumulative distribution curves, the neural network accurately quantified key classification fractions, specifically fines content under 0.075 millimeters, sand content between 0.075 and two millimeters, and gravel content exceeding two millimeters. The model attained modified determination coefficients of 91.39 percent for fines, 91.37 percent for sand, and 94.41 percent for gravel, with corresponding mean absolute errors of 0.0264, 0.0350, and 0.0523. Estimation accuracy for fine materials proved elevated in soil mixtures containing over twenty percent fines, indicating that clay binders generate coherent textural cues that facilitate algorithmic detection.

To probe the generalizability of the trained system under uncontrolled industrial conditions, the researchers conducted supplementary validation trials on commercial recycled soil acquired from an operating recycling plant without prior disclosure of its raw feedstocks, binder additives, or moisture history. Evaluated against laboratory mechanical sieve benchmarks, the convolutional model predicted a fines content of 11.3 percent versus a measured 7.1 percent, a sand content of 52.8 percent versus 43.2 percent, and a gravel fraction of 35.9 percent versus 49.7 percent. While commercial prediction discrepancies slightly exceeded those observed on controlled laboratory mixtures, the authors noted that such divergence is expected given uncalibrated external feedstocks and foreign inclusions. Crucially, the outcome demonstrates that the lightweight neural network architecture can function across individual recycling facilities, provided plant operators assemble site-specific calibration databases reflecting local soil origins. The research team highlighted that embedding this rapid optical methodology into plant control rooms will permit real-time classification and continuous permeability estimation, advancing sustainable material circulation across modern civil infrastructure projects.

Subject of Research: Convolutional neural network prediction of particle size distribution in recycled stabilized soils

Article Title: Image-based prediction of particle size distribution in recycled stabilized soils using convolutional neural networks

Article References: Anh, T. T. P., Hayano, K., & Thang, B. A. (2026). Image-based prediction of particle size distribution in recycled stabilized soils using convolutional neural networks. International Journal of Geo-Engineering, 17(1), Article 7. https://doi.org/10.1186/s40703-026-00263-x

Image Credits: AI Generated

DOI: 10.1186/s40703-026-00263-x

Keywords: recycled stabilized soil, convolutional neural network, particle size distribution, construction sludge, Ao clay, paper sludge ash, blast furnace slag cement, carbonation curing, sieve analysis, fines content, sand content, gravel content, soil classification, image-based particle estimation, intraparticle porosity, gray-level co-occurrence matrix, geotechnical quality control

News Source: Cassandra Pierce. (November 21, 2026). Convolutional Neural Network Accurately Predicts Particle Sizes in Recycled Stabilized Soil. Scienmag.

Tags: Ao clayblast furnace slag cementcarbonation curingconstruction sludgeconvolutional neural networkfines contentgeotechnical quality controlgravel contentgray-level co-occurrence matriximage-based particle estimationintraparticle porositypaper sludge ashparticle size distributionrecycled stabilized soilSand contentsieve analysissoil classification
Share12Tweet7Share2ShareShareShare1

Related Posts

Novel Probe Cell Accelerates Localized Testing of Fuel Cell Catalyst Layers

Novel Probe Cell Accelerates Localized Testing of Fuel Cell Catalyst Layers

November 22, 2026
Wrinkled leaf-like electrocatalyst accelerates oxygen evolution reaction with high stability

Wrinkled leaf-like electrocatalyst accelerates oxygen evolution reaction with high stability

November 22, 2026

Machine Learning Estimates Transparent Wood Transmittance More Accurately Than Haze

November 22, 2026

Controlled Forging Deformation Optimizes Aluminum Alloy Strength and Corrosion Resistance

November 22, 2026

POPULAR NEWS

  • Analysis Links Ganoderma Lucidum Target NR3C2 and Gut Microbiota to Insomnia

    Analysis Links Ganoderma Lucidum Target NR3C2 and Gut Microbiota to Insomnia

    29 shares
    Share 12 Tweet 7
  • THAM Catalyzes Efficient Green Gewald Synthesis of Diverse 2-Aminothiophene Heterocycles

    29 shares
    Share 12 Tweet 7
  • Waste Glass and Sisal Fibers Increase Concrete Strengths in Neural Network Study

    29 shares
    Share 12 Tweet 7
  • Cementless Megaprosthesis Achieves Favorable Long-Term Stability After Bone Tumor Resection

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Pediatric Urticaria Linked to Higher Risk of Subsequent Allergic Airway Diseases

Synergistic Machine Learning Framework Identifies Biologically Relevant Cancer Genes

Normative Baseline Data Established for Japanese Tablet-Based Cognitive Assessment App

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.