Skip to main content
Graphic of computer code
November 17, 2026

How AI and ML can accelerate chemical engineers’ work

Written By:

Categories:

Before algorithms and digital streams entered the laboratory, chemists Fritz Haber and Carl Bosch engineered a new process to create ammonia, a key ingredient for fertilizer. The critical discovery made organized, high-yield farming possible, which supported large communities and laid the foundation for modern civilization.

High-strength Kevlar, mass production of penicillin, reduction of harmful air pollutants with catalytic converters, energy storage in lithium-ion batteries, water purification: All these inventions built chemical engineering’s reputation for tackling global issues, adopting advanced tools and scaling them to change the world. 

Today, we face an unprecedented wave of complex digital information. We now generate far more data and information than humans can manually process. Human health and psychology experts classify this challenge as information overload, which can lead to burnout and lost performance, and impacts the quality of individuals’ decision making. 

From designing tiny molecules and operating massive manufacturing facilities, chemical engineers in particular manage a high volume of information, including laboratory measurements, sensor data, process diagrams, simulations and more. Making sense of vast, multiscale complexity demands more than traditional intuition, it makes modern computational tools like AI and ML an absolute necessity. 

Artificial intelligence can be a powerful companion in helping teams quickly analyze enormous data sets. Then researchers can prioritize what humans do best: explore new ideas, develop creative solutions, actively innovate and focus our time on the decisions that matter most. 

So what is AI?

Artificial intelligence (AI) is any machine-based system capable of performing tasks that traditionally require human intelligence. Using provided inputs, it generates outputs, which includes everything from predictions, content, recommendations, and decisions. For example, a researcher might utilize AI to organize raw data in spreadsheets and plot key trends or flag unusual results, while a plant manager could use it to monitor equipment sensors and flag subtle changes over time.

Yet, to tackle major global issues—like discovering zero-carbon fuels, designing recyclable polymers, or optimizing energy-intensive reactions—data analysis isn’t enough. We need tools that can look into the unknown, predict untested behavior, and navigate billions of possibilities in a short period of time.

To dive even deeper, we turn to machine learning

Machine learning (ML) is a specialized branch of AI in which algorithms learn patterns directly from data rather than following rigid, pre-written code. ML algorithms, or models, generally learn through four main approaches:

  • Supervised learning: Models train on historical, organized data that maps input with known results. For example, when trained on past production data with temperatures, pressures, flow rates and final quality, a model can estimate the quality of a new batch before laboratory testing is complete.
  • Unsupervised learning: Models ingest unidentified data to discover hidden structures, recurring patterns or natural groupings without prior guidance. For instance, a model can analyze years of equipment sensor readings and organize them into distinct clusters, helping engineers identify unusual operation combinations that signal equipment degradation.
  • Reinforcement learning: Models learn through trial and error within simulated environments, refining their decisions and outputs by receiving feedback on performance. For example, a model learns optimal operating techniques by adjusting temperature or mixing speeds in a simulated reactor. It earns rewards for maximizing product yield or receives penalties for wasting energy or creating unsafe conditions.
  • Self-supervised learning: Models learn by creating their own practice problems from raw data, masking or manipulating portions of the data and training themselves to predict the missing pieces. A problem or two might not seem like enough, but frequent repetition across vast data sets effectively teachers a model important underlying patterns. It is particularly useful for modern AI systems that have enormous amounts of undefined data. For example, a model might hide segments of a molecular structure to learn about chemical properties and formation, allowing it to understand chemical rules before fine-tuning property predictions on new molecules.

In practice, these insights become vital safeguarding tools for modern industry. Set up to analyze data from factory sensors and equipment, models can quickly identify emerging mechanical faults or quality issues before major problems occur, like hazardous shutdowns or food contamination. Chemical engineers can also avoid putting personnel at risk and damaging expensive equipment by using models to test new processes, controls and other strategies in a digital environment–otherwise known as a simulation.

So by assuming exhaustive data analysis, routine tasks and calculations, AI and ML transform how chemical engineers work—saving them valuable time investigating every single possibility. Continuous learning also unlocks additional insights that go far beyond simple calculations. Models can quickly distill information and analyze different research avenues so chemical engineers focus on the most promising results. With that time savings, chemical engineers can pursue creative solutions, engage in problem-solving, and undertake uniquely human tasks to tackle major societal challenges.

Learning models also power a rapidly expanding range of AI tools, like generative AI (think high-level ChatGPT). The most novel of this technology allow us to create entirely new content and perform advanced simulations. With the right information and data, AI assistants can design unique materials or chemical structures, provide real-time plant simulations, and more. This further vets possible research paths so the focus is on unconventional, high-performing solutions, and in turn we can limit energy-intensive simulations to these solutions.

With our domain expertise guiding generative AI to ensure realistic and sustainable solutions, it continues to improve. Then we can validate responses with real-world testing, update AI with the results, and so on. All of this dramatically and exponentially accelerates scientific discovery, shaving years, or even decades, off testing and experimentation.

Applying good engineering judgement

From executing code, verifying information, solving complex problems, to even molecular discovery, computational tools have made significant improvements in the past few years. And while they make less factual errors, they’re still not infallible.

For example, a model might create an error-free computer code that solves the wrong equation, suggest a modification without recognizing a physical or operational constraint, or generate a molecule that is difficult or unsafe to synthesize. These are reminders that even sophisticated computational tools need to be tested and validated, particularly when consequences of a mistake are significant.

As AI becomes more capable, perhaps the most interesting questions is no longer simply what it can do, but how we should use it. Where can it help us explore ideas that were previously out of reach? When should we trust its recommendations, and how do we know when to question them?

Chemical engineering has always advanced by combining new tools with scientific understanding and practical judgment. AI offers another powerful opportunity to continue this legacy.

Joel Paulson is the Gerald and Louise Battist Associate Professor in the Department of Chemical and Biological Engineering. He is a leading expert in artificial intelligence and machine learning in chemical engineering.