Courses

Where AI and machine learning fit in chemical engineering, then Python from first principles in Google Colab.

1. Introduction to AI, machine learning and data science

   Supervised, unsupervised and reinforcement learning, the data-to-model workflow, and use cases across the process industries.

2. Getting started with Google Colab

   Notebooks, uploading datasets and working in the browser.

3. Python crash course for chemical engineers

   Data types, lists, loops, functions and libraries, each shown on an engineering calculation.

4. Cooling water pump flowrate analysis

   Cooling water pumps. Functions and loops in plain Python, with unit conversion.

5. Heat loss in a pipe with NumPy

   Insulated pipework. Vectorized engineering calculations with NumPy arrays.

6. ASTM D86 distillation curves with pandas

   ASTM D86 distillation data. Loading, filtering, sorting and plotting a distillation curve.

7. Batch reactor kinetics with SciPy

   Batch reactor. Solving and fitting reaction kinetics numerically.

Turning raw plant data into data a model can learn from: gaps, outliers, scales and imbalance, then a complete pipeline.

1. Data preprocessing and feature engineering in chemical engineering

   Why plant data needs preparing, and the steps from raw historian data to a model-ready table.

2. Handling missing values

   Distillation tray temperatures. Dropping, filling, interpolation and KNN imputation, compared on accuracy.

3. Handling missing values: advanced methods

   Distillation tray temperatures. Edge cases in imputation and choosing a method for time-ordered data.

4. Detecting and treating outliers

   Distillation tray temperatures. Boxplots, z-score and IQR methods, treatment, and system-wide correlated outliers.

5. Normalization and scaling

   Pump operating data. Min-max normalization and z-score standardization, and when to use each.

6. Handling imbalanced data

   Compressor failure data. Class distribution, undersampling and oversampling for rare failure events.

7. A complete preprocessing pipeline

   Polymer property prediction. Domain checks, train-test split, outlier capping, imputation, feature engineering and scaling, end to end.

Building the inputs that let a model see the process the way an engineer does.

1. Feature engineering for filtration processes

   Filtration data. Creating physically meaningful features, and comparing models on raw and engineered inputs.

2. Feature extraction and selection

   Heat exchanger condition monitoring. Correlation analysis, PCA and LDA, then filter, wrapper and embedded selection methods.

Regression, classification and unsupervised methods, each validated the way plant models need to be.

1. Introduction to machine learning and model types

   Where machine learning sits in the automation hierarchy, training and testing, overfitting and the bias-variance trade-off.

2. Linear regression

   Distillation column product purity. Fitting, coefficients and residual analysis on feed, reflux, duty and temperature data.

3. Regularization and PLS

   NIR spectroscopy for glucose. Ridge, lasso and elastic net on high-dimensional spectra, then partial least squares.

4. Cross-validation and hyperparameter tuning

   K-fold, leave-one-out, stratified and time series validation; grid, random and nested search.

5. Cross-validation for reactor models

   Catalytic reactor conversion. Single and repeated splits, k-fold, expanding and sliding windows, and walk-forward validation.

6. Hyperparameter tuning

   Catalytic batch reactor. Ridge sensitivity, learning curves, GridSearchCV, elastic net and tree-based models.

7. Ensemble methods for regression

   Wastewater treatment effluent COD. Decision trees, random forests and boosting, from unstable single trees to production-ready models.

8. Binary classification

   Ethylene oxide reactor fault detection. From regression to classification, logistic regression and decision thresholds.

9. Classification metrics

   Pipeline leak detection. Confusion matrices, precision and recall, ROC and precision-recall curves for rare events.

10. Multi-class classification

    Catalyst state. Softmax regression, per-class metrics, macro and micro averaging, one-vs-rest and one-vs-one.

11. Unsupervised learning

    Clustering, dimensionality reduction and anomaly detection when the data has no labels.

12. Clustering

    Crude oil blends. K-means, choosing the number of clusters, hierarchical clustering and profiling the operating regimes.

13. Anomaly detection

    Batch quality screening. Z-score screening, PCA-based process monitoring and isolation forest, compared on the results.