Industrial Machine Learning with PLC–MATLAB OPC UA Communication
This project demonstrates how industrial process data can be used to train a machine learning model and how such models can be connected to PLC systems using OPC UA.
The main goal is to predict product quality in an industrial mining process and show how ML can support automation and process optimization.
- Communicate with PLC systems using OPC UA (MATLAB side)
- Train an industrial ML model in Python
- Predict Silica percentage (Silica %) in a flotation process
- Use real industrial data from Kaggle
- Demonstrate ML usage for Industry 4.0
Source: Kaggle
Dataset: Quality Prediction in a Mining Process
🔗 https://www.kaggle.com/datasets/edumagalhaes/quality-prediction-in-a-mining-process/data
- Real data from an iron ore flotation plant
- Time-series industrial sensor data
- The target variable is Silica % in iron concentrate
- Large dataset (millions of rows)
Selected process parameters:
- % Iron Feed
- % Silica Feed
- Starch Flow
- Amina Flow
- Ore Pulp pH
- Ore Pulp Density
- Flotation Columns Air Flow
- % Silica Concentrate
Steps used in the notebook:
- Load CSV data
- Select relevant industrial features
- Remove unnecessary columns
- Random sampling (100,000 rows) to speed up training
- Train/Test split
- Basic data visualization and inspection
- Problem type: Regression
- Goal: Predict Silica % from process parameters
- Language: Python
- Libraries:
- pandas
- numpy
- scikit-learn
- matplotlib
- seaborn
- Linear Regression
- Random Forest Regressor
Random Forest showed better performance for nonlinear industrial data.
Evaluation metrics used:
- Mean Absolute Error (MAE)
- Mean Squared Error (MSE)
- R² Score
The trained model is suitable for process monitoring and decision support, not for direct closed-loop control.
This project is designed to work together with:
- PLC systems
- OPC UA communication
- MATLAB as OPC UA client
This heatmap shows missing values in the dataset.
All selected features have no missing data, so no imputation is required.
This matrix shows correlations between process variables and product quality.
Main observations:
- Strong negative correlation between % Iron Feed and % Silica Feed
- Strong negative correlation between % Iron Concentrate and % Silica Concentrate
- Strong positive correlation between Flotation Column 01 Air Flow and Column 02 Air Flow
This chart shows how each feature correlates with the target variable.
Key points:
- Amina Flow has the strongest positive correlation
- Air Flow parameters have negative correlation
- Most variables have weak to moderate influence
The histogram and boxplot show the distribution of the target variable.
Observations:
- Mean value is around 2.3
- Data is slightly right-skewed
- Some outliers are present
Box plots for all features show the presence of outliers.
Outliers are expected in industrial process data and were not removed to keep realism.
Comparison of regression models using:
- RMSE
- MAE
- R² Score
Results:
- Random Forest shows the best performance
- Linear and Ridge regression perform worse on nonlinear data
This plot shows which features are most important for prediction.
Key insight:
- % Iron Concentrate is the most influential feature
- Process parameters like air flow and pH also contribute
This scatter plot compares real and predicted values.
- Points close to the diagonal line mean good predictions
- Random Forest shows strong accuracy
Residual plots show:
- Errors are centered around zero
- Residuals are approximately normally distributed
- No strong bias is visible
This chart compares initial process settings with optimized values suggested by the model.
Purpose:
- Support operator decision-making
- Not intended for direct automatic control
Show the video of realization work
- Almat Abdimalik
- Rassul Zeinulla