AI-Optimized Drying of Bio-Based Concrete

Bio-based concretes combine a mineral binder with plant-based aggregates such as hemp shiv, flax shives, or wood particles. They are increasingly recognized as strategic materials for decarbonizing the building sector. Their natural hygrothermal regulation capacity, low embodied energy, and ability to store biogenic carbon directly address the objectives of the RE2020 and the National Low-Carbon Strategy.

Despite growing scientific and industrial interest, a major challenge remains: post-production drying is slow, energy-intensive, and poorly controlled.

Highly porous and hydrophilic plant-based aggregates interact in complex ways with the binder during hydration. Drying conditions (temperature, relative humidity, duration, and ventilation) directly influence binder hydration kinetics and the microstructure of the composite, and consequently govern the final hygrothermal properties of the hardened material [1,2].

This dependency has been partially documented in the literature [3–5], but the available data remain scattered, obtained using heterogeneous protocols, and rarely compiled within a unified framework.

To our knowledge, there is currently no systematic and predictive approach linking drying conditions and formulation to measured hygrothermal performance. The BBC-DROP project directly addresses this gap and aims to develop an AI-assisted optimization approach for drying protocols of bio-based concretes.

It combines three methodological components:

1. Predictive modeling using AI-based models;
2. Multi-objective optimization;
3. Experimental development and validation.

BBC-DROP is structured around four complementary and progressive research areas, all achievable within a three-year PhD project (see Fig. 1).

Research Area 1: Database Development and Experimental Enrichment

The first step consists of a systematic and critical literature review on the properties of bio-based concretes. The review focuses on the influence of formulation and drying conditions on mechanical and hygrothermal performance.

The data extracted from publications will cover:

• Density;
• Compressive strength;
• Porosity;
• Thermal conductivity;
• Specific heat capacity;
• Water vapor permeability;
• Sorption isotherms;
• Capillary absorption.

These data will be structured and harmonized into a centralized database, which will form the main foundation for training the AI models.

In a second stage, targeted experimental campaigns will be conducted at CEREMA under the supervision of the research team, which specializes in the behavior of bio-based materials. These campaigns will specifically target configurations that are underrepresented or absent from the existing literature.

Hemp and flax concrete specimens will be subjected to controlled drying protocols in climate chambers. The conditions will vary in:

• Temperature: 20°C to 60°C;
• Relative humidity: 30% to 80%;
• Duration: 7 to 90 days.

The database will therefore be enriched with high-quality original data covering a broad and representative range of conditions.

Research Area 2: AI-Based Predictive Modeling

Machine learning models will be trained on the consolidated database to predict hygrothermal properties.

The input variables will include formulation parameters such as:

• Properties of plant-based aggregates;
• Binder properties;
• Mixture properties;

as well as drying conditions:

• Temperature;
• Relative humidity;
• Ventilation duration.

The outputs will include the resulting mechanical and hygrothermal properties, such as:

• Compressive strength;
• Thermal conductivity;
• Water vapor permeability;
• Porosity.

The algorithms considered may include neural networks, gradient boosting, and coupled models.

Transfer Learning techniques may also be considered at this stage to address the potential lack of data.

Particular attention will be paid to model interpretability through SHAP values and sensitivity analyses. These tools will identify the most influential variables and provide physical insight into the observed trends.

Research Area 3: Multi-Objective Optimization

The validated AI models will serve as objective functions within a multi-objective optimization framework using the NSGA-II algorithm.

For a given set of mechanical and hygrothermal performance targets, the algorithm will identify optimal combinations of drying protocols and formulations.

The outcome will consist of Pareto fronts linking drying conditions, formulation parameters, and performance.

These maps can be directly used as decision-support tools within the work carried out by CEREMA teams on bio-based concretes.

Research Area 4: Experimental Validation of Optimized Configurations – CEREMA Strasbourg

The configurations identified as optimal following Research Area 3 will be manufactured, dried according to the optimized protocols, and characterized at CEREMA Strasbourg.

An experimental campaign will be conducted to characterize these concretes manufactured and dried under controlled conditions.

After drying, the manufactured concretes will be characterized in terms of:

• Mechanical properties;
• Hygrothermal properties;
• Microstructural properties.

The results obtained from the characterization campaign will then be compared with the model predictions and the performance targets initially defined.

This stage aims to assess the reliability and generalization capability of the developed optimization approach.

The closed loop between numerical optimization and experimental confirmation is a key and distinctive element of the BBC-DROP project.

Figure 1. Overview of the PhD research topic.

Work Programme

The BBC-DROP work programme will run for 36 months and will be organized into six progressive and interdependent work packages.

Recruitment

Recruitment process: Application review and interview.

The application must include:

• A detailed CV;
• A cover letter explaining the candidate’s motivation to pursue a PhD;
• Post-secondary academic results and corresponding transcripts;
• Any recommendation letters;
• Any other supporting document deemed relevant.

Applications will be processed in the order in which they are received. Therefore, this PhD offer will expire once a candidate has been selected.

Organization

Laboratories:

• CESI LINEACT Strasbourg
• CEREMA Strasbourg

Start date: October 2026

Duration: 36 months

Supervision

Djaoued Beladjine, Dr. HDR
Associate Research Professor, CESI LINEACT
PhD Supervisor

Etienne Gourlay, PhD
Research Scientist, CEREMA Strasbourg
Co-supervisor

Mazhar Hussain, PhD
Research Professor, CESI LINEACT Strasbourg
Co-supervisor

 

Skills required

Diplôme de Master en Génie Civil, Science des Matériaux, Physique du bâtiment, Informatique ou domaine connexe. Connaissances en formulation et caractérisation des matériaux de construction, idéalement biosourcés ou cimentaires. Notions en physique du bâtiment et en transferts hygrothermiques. Intérêt pour l’expérimentation, la modélisation numérique, l’analyse de données et l’intelligence artificielle. Compétences en programmation, idéalement en Python ; une expérience en Machine Learning et/ou en optimisation multi-objectif constitue un atout. Capacité à réaliser une synthèse bibliographique, à gérer des données scientifiques, ainsi que d’excellentes capacités de rédaction et de communication en français et en anglais.