🔬 condensed matter

Hydrogen production from blended waste biomass: pyrolysis, thermodynamic-kinetic analysis and AI-based modelling

This study investigates the thermochemical conversion of spent coffee grounds and date seeds for sustainable hydrogen production through pyrolysis, utilizing comprehensive kinetic and thermodynamic analyses alongside an AI-based LSTM model to optimize process efficiency and predict thermal degradation behavior with exceptional accuracy.

Original authors: Sana Kordoghli, Abdelhakim Settar, Oumayma Belaati, Mohammad Alkhatib, Khaled Chetehouna, Zakaria Mansouri

Published 2026-01-15

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sana Kordoghli, Abdelhakim Settar, Oumayma Belaati, Mohammad Alkhatib, Khaled Chetehouna, Zakaria Mansouri

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have two types of kitchen trash that usually end up in a landfill: old coffee grounds from your morning brew and the hard pits from dates. Instead of throwing them away, this paper asks: What if we could turn this waste into clean fuel (hydrogen) using heat, and use a super-smart computer brain to figure out the perfect recipe?

Here is the story of that experiment, broken down simply:

1. The Ingredients: Coffee and Dates

The researchers took Spent Coffee Grounds (SCG) and Date Seeds (DS). Think of these as two different types of "wood" made of plant fibers (cellulose, hemicellulose, and lignin).

Coffee Grounds: Like a dense, oily sponge.
Date Seeds: Like a hard, fibrous nut.

They didn't just test them alone; they mixed them together in different recipes (blends), like a chef testing different ratios of flour and sugar to see which makes the best cake. They tried three mixes: mostly dates, a 50/50 split, and mostly coffee.

2. The Cooking Process: The "Pyrolysis" Oven

To turn this waste into gas, they put the samples into a special oven called a pyrolysis reactor.

The Method: They heated the samples up to 650°C (very hot!) without any oxygen. It's like roasting a marshmallow in a fire, but instead of letting it burn with a flame, they trap the smoke and gases inside.
The Goal: When these plant fibers get super hot, they break apart. The researchers wanted to see how much Hydrogen gas (a clean fuel) they could catch in a bag, along with other gases.

3. The Detective Work: Kinetics and Thermodynamics

Before they could cook, they needed to understand how the ingredients behave under heat. They used three main tools:

The Scale (TGA): They weighed the samples as they heated up to see exactly when they started losing weight (turning into gas).
The Speedometer (Kinetics): They calculated the "energy cost" to break the bonds. Imagine trying to open a jar. Some jars (like the date seeds) are easier to open (lower energy), while others (like the coffee grounds) are stuck tight and need a lot of force (higher energy).
- The Surprise: The 75% Date / 25% Coffee mix was the easiest to "open" (it needed the least energy).
- The Trade-off: The 75% Coffee / 25% Date mix was the hardest to open (needed the most energy), but it produced the most hydrogen gas. It's like a high-effort workout that gives you the biggest reward.

4. The Computer Brain: AI and LSTM

This is where the paper gets futuristic. Instead of just running the oven experiment over and over again (which takes a long time and costs money), they built an AI model (specifically called an LSTM, which is like a robot that remembers patterns over time).

The Training: They fed the AI data from their oven experiments.
The Magic Trick: They gave the AI two versions of the "recipe book":
1. Basic Recipe: Just temperature and mixing ratios.
2. Advanced Recipe: Temperature, ratios, plus the specific chemical makeup of the plant fibers (how much cellulose, hemicellulose, and lignin was inside).
The Result: The AI became a crystal ball. It could predict exactly how the waste would behave in the oven with 99.9% accuracy.
- The "Advanced Recipe" AI was even better. It was so good at guessing that it could predict what would happen at temperatures and speeds it had never seen before. It's like a chef who, after tasting a few soups, can perfectly guess the taste of a new soup they've never made, just by knowing the ingredients.

5. The Final Verdict

The paper concludes with two main takeaways:

For the most Hydrogen: Use the mix with more coffee grounds (Blend 3). It gives the most gas, even though it takes more energy to get there.
For the easiest process: Use the mix with more date seeds (Blend 1). It requires the least amount of energy to start the reaction.
For the future: Using AI is a game-changer. It allows scientists to skip the messy, time-consuming trial-and-error phase and jump straight to the best settings for making clean fuel.

In a nutshell: This study proves that your morning coffee trash and date pits can be turned into clean hydrogen fuel. By mixing them just right and using a smart computer to predict the results, we can turn waste into energy more efficiently than ever before.

Technical Summary: Hydrogen Production from Blended Waste Biomass

Problem Statement
The global transition away from fossil fuels necessitates the development of sustainable "green hydrogen" production methods to mitigate climate change and reduce carbon emissions. While biomass offers a promising renewable feedstock, the inherent complexity of biomass composition and the variability of thermal degradation processes pose significant challenges for optimization. Specifically, underutilized agricultural residues like Spent Coffee Grounds (SCG) and Date Seeds (DS) remain largely unexplored for hydrogen production. Furthermore, traditional experimental methods for characterizing pyrolysis kinetics and predicting thermal behavior are often labor-intensive and time-consuming. There is a critical need to optimize the pyrolysis process for these specific feedstocks and to leverage advanced computational tools, such as Artificial Intelligence (AI), to enhance modeling accuracy and process efficiency.

Methodology
This study employed a comprehensive approach combining experimental characterization, kinetic and thermodynamic analysis, and AI-based predictive modeling:

Sample Preparation and Characterization: SCG and DS were dried, ground, and analyzed individually and as three distinct blends (75% DS–25% SCG, 50% DS–50% SCG, and 25% DS–75% SCG). Analyses included proximate and ultimate composition, fiber content (cellulose, hemicellulose, lignin), and thermogravimetric analysis (TGA) under inert nitrogen atmosphere.
Pyrolysis Experiments: Experiments were conducted in a fixed-bed reactor at 650°C with a heating rate of 10°C/min. Gas products were collected and analyzed using Micro-Gas Chromatography (Micro-GC) to quantify hydrogen and other volatile components.
Kinetic and Thermodynamic Analysis: Isoconversional methods (KAS, FWO, and Friedman) were applied to TGA data to determine activation energies ( $E_a$ ) and pre-exponential factors. Thermodynamic parameters (enthalpy $\Delta H$ , Gibbs free energy $\Delta G$ , and entropy $\Delta S$ ) were calculated to assess energy requirements.
AI-Based Modeling: Long Short-Term Memory (LSTM) neural networks were developed using Python (TensorFlow/Keras) to predict TGA mass loss curves. Two models were trained:
- Model 1: Utilized basic features (temperature, heating rate, blend ratios).
- Model 2: Extended Model 1 by incorporating lignocellulosic composition data (Cellulose, Hemicellulose, Lignin percentages) which were dynamically adjusted based on temperature ranges.
  Hyperparameter tuning was performed using Keras Tuner and RandomSearch. The models were validated against unseen data, including a new heating rate of 25°C/min.

Key Results

Physicochemical Properties: Both SCG and DS exhibited high volatile matter and suitable heating values for bioenergy. DS showed higher hemicellulose content (48.2%), while SCG had higher cellulose content (32%).
Kinetic Analysis: The KAS model was identified as the most accurate for pure samples.
- Blend 1 (75% DS–25% SCG): Exhibited the lowest average activation energy ( $E_a \approx 161.75$ kJ/mol), indicating the highest energy efficiency and easiest decomposition.
- Blend 3 (25% DS–75% SCG): Showed the highest activation energy ( $E_a \approx 313.24$ kJ/mol), suggesting a more complex and energy-intensive degradation process.
Thermodynamics: Thermodynamic parameters confirmed that Blend 1 requires the least external energy input, while Blend 3 demands the highest, consistent with kinetic findings.
Hydrogen Yield: Micro-GC analysis revealed that increasing the SCG proportion in blends increased hydrogen concentration. Blend 3 (75% SCG) produced the highest hydrogen yield (3.4378 mole %), followed by Blend 2 and Blend 1. Pure SCG yielded 4.6259 mole % H₂, significantly higher than pure DS (0.9024 mole %).
AI Modeling Performance:
- Model 1 achieved $R^2$ values between 0.9658 and 0.9958.
- Model 2, which included lignocellulosic composition, demonstrated exceptional accuracy with $R^2$ values ranging from 0.9996 to 0.9998 across various heating rates, including the unseen 25°C/min rate. Model 2 significantly reduced Mean Squared Error (MSE) and Mean Absolute Error (MAE) compared to Model 1.

Significance and Contributions
The paper claims to fill a notable gap in the literature by applying advanced AI/ML techniques specifically to the pyrolysis of SCG, DS, and their blends. The primary contributions are:

Optimization of Blends: The study identifies a trade-off between energy efficiency and hydrogen yield. Blend 1 is optimal for energy-efficient pyrolysis, while Blend 3 is superior for maximizing hydrogen production, despite higher energy demands.
AI Integration: The research demonstrates that integrating lignocellulosic composition data into LSTM models significantly enhances the prediction of thermal degradation behavior, achieving near-perfect accuracy ( $R^2 > 0.999$ ). This offers a cost- and time-efficient alternative to extensive experimental testing.
Waste Valorization: The work validates the potential of converting underutilized food wastes into sustainable energy carriers, supporting the transition toward a circular hydrogen economy.

The authors conclude that while the LSTM models show high predictive capability, future work should focus on expanding dataset diversity to further improve model generalization across different biomass types and blend ratios.

1. The Ingredients: Coffee and Dates

2. The Cooking Process: The "Pyrolysis" Oven

3. The Detective Work: Kinetics and Thermodynamics

4. The Computer Brain: AI and LSTM

5. The Final Verdict

More like this