AIC115: AI Methods for Chemical and Process Engineers

Duration: Approximately 7 Hours Online (Self-Paced Video Course).

Audience: Process Engineers, Chemical Engineers, Control Engineers, Reliability Engineers, and Engineering Students.

Prerequisites: Familiarity with process variables, sensors, and basic process control. No prior machine learning or programming experience is required.

Course Material: Software Products used in Course - Pitops, Google Colab Notebooks, Training Slides, Supplementary PDF Reference Materials.

Many chemical and process engineers hear about artificial intelligence and machine learning but are unsure how these methods actually apply to their process, their instruments, and their day-to-day decisions. This course closes that gap by teaching AI concepts through recognizable process engineering problems - fault detection, soft sensing, batch monitoring, and equipment degradation.

The course starts with a foundation in how to recognize patterns in process data and how to frame an AI project correctly before any model is built. It then moves through classical machine learning methods for fault detection and soft sensing, time-series methods for batch monitoring and predictive maintenance, and deep learning methods for anomaly detection. The course closes with a practical discussion of how AI models are safely deployed, monitored, and maintained in a plant environment, followed by a live demonstration of an AI model working inside Pitops.

No coding is required to complete the course - every model is demonstrated through narrated video and pre-built Colab notebooks. Learners who want to go deeper into the mathematics behind each method will find detailed supplementary PDF material for every algorithm covered.

After completing this course, attendees will be able to recognize recurring pattern types in process data, identify the best-suited AI method for a given process problem, and explain the reasoning behind that choice to both engineering colleagues and data science teams.

Attendees will be able to interpret machine learning model outputs in process engineering terms - feature importance, confusion matrices, false alarms, missed detections, and anomaly scores - and will understand the difference between a model that interpolates safely within known conditions and one that is asked to extrapolate beyond them.

Attendees will also be able to correctly frame an industrial AI project - defining the problem, the data, the labels, and the success criteria - and will understand what is required to move a model safely from a demonstration into plant use, including human oversight, fallback behavior, and ongoing monitoring for model drift.

The following topics are covered in this course:

• Pattern Recognition in Process Data.
• Drift, Anomalies, Cyclic Patterns, Correlation Shifts, and Fault Signatures.
• Features, Classification, and Regression in Machine Learning.
• Supervised versus Unsupervised Learning.
• Overfitting and Underfitting.
• Framing an Industrial AI Project.
• Exploratory Data Analysis and Data Preprocessing.
• Supervised Fault Classification Using Random Forest, XGBoost, and Support Vector Machines.
• Unsupervised Anomaly Detection Using Isolation Forest.
• Model Comparison and Validation on Unseen Data.
• Soft Sensors for Predicting Product Quality.
• Interpolation versus Extrapolation.
• Time-Series Data and Introduction to Neural Networks.
• Batch Process Monitoring and Early Fault Detection.
• Predictive Maintenance and Equipment Health Indicators.
• Autoencoders for Unsupervised Anomaly Detection.
• Deploying and Monitoring Industrial AI Models.
• Case Study: Automatic Detection of Good Segments for System Identification (Pitops).