Optimizing Cement Formulation with AI | RapidMiner

Methods for Predicting Cement Strength

How does the AI model process different process parameters to predict cement strength?

Process: Data Collection & Preparation > Data Input into the Model > Model Training > Prediction & Validation > Application & Optimization

1. Data Collection and Preparation:

   • Collected 1,030 sets of experimental data from cement manufacturing processes. These datasets include process parameters such as cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, and fine aggregate.

2. Data Input into the Model:

   • These datasets are imported into Altair AI software RapidMiner for training. Each data point contains the kilograms of each material per cubic meter of concrete and the final compressive strength.

3. Model Training:

   • RapidMiner analyzes these data and trains an AI model for predicting compressive strength. The model inputs are the various process parameters, and the output is the predicted cement strength.

4. Prediction and Validation:

   • After training, the model is used for prediction. The predicted values are then compared with the experimental results (as shown in the figure below). The red line represents the AI model’s predicted strength, while the blue line represents the measured experimental strength. The chart shows that the prediction and experimental results match closely, with a prediction success rate of over 90%.

5. Application and Optimization:

   • The AI model can not only predict cement strength, but also be combined with optimization software to find the best mix by defining upper and lower bounds for each parameter. For example, you can limit the cement weight below a certain value while still achieving the ideal strength.

Prediction Results

• 90% prediction success rate: The high prediction success rate of the AI model demonstrates its effectiveness in practical applications.

• By using the AI prediction model together with HyperStudy, the optimal cement mix can be obtained.

RapidMiner accurate cement strength prediction

 

Technical Highlights: Why Can Accuracy Reach Over 90%?

1. Large Amount of Supporting Data:

   • A total of 1,030 cement test datasets were collected, covering a wide variety of process parameters and results, providing sufficient samples to train the AI model.

   • With a large volume of data, the model can better learn and identify relationships between parameters, improving prediction accuracy.

2. Correct Parameter Recording:

   • Besides having a large dataset, selecting the right parameters is also crucial. The parameters must significantly affect the results so that the AI model can accurately learn the influence of each parameter on cement strength.

3. Powerful AI Tool (RapidMiner):

   • Using a powerful AI platform like RapidMiner for model training. RapidMiner has advanced machine learning algorithms and analytics capabilities, enabling it to efficiently process large datasets and make accurate predictions.

↘ Learn more about RapidMiner

1. Model Optimization and Adjustment:

   • During model training, continuous optimization and tuning were performed to ensure the accuracy of the predictions. Through iterative refinement, a model with a prediction success rate of over 90% was eventually obtained.

2. Validation with Real Data:

   • The predicted results were compared with the experimental data, and the red prediction curve nearly overlapped the blue experimental curve. This real-data validation further proves the model’s accuracy and reliability.

3. Integrated Optimization Capability:

   • In this case, after building the AI prediction model, it was further combined with HyperStudy for cement mix ratio optimization. The optimal cement formula can be obtained without additional experiments.
     
     
     

RapidMiner integrated with HyperStudy for optimal cement mix design