The evolution of industrial AI: A short history of AVEVA’s AI innovation
Posted: December 18, 2025
Industrial AI isn’t one thing. It’s machine learning, deep learning, reinforcement learning; it’s neural networks and predictive analytics, generative and agentic AI, and more. Altogether, it’s a broad set of technologies and techniques designed with deep domain expertise specifically for industrial applications.
The field may be evolving at a breakneck pace, but it’s hardly new. For more than two decades, AVEVA has been embedding AI into industrial software to help organizations deliver safer, more efficient and reliable outcomes. Industrial AI looks wildly different today than it did twenty years ago, and twenty years from now, that evolution will only get more dramatic.
In this blog, we’ll trace the evolution of industrial AI through AVEVA’s own industrial AI journey. From predictive analytics to the beginnings of autonomous operations and generative design, we’ll explore some of the key milestones that brought us here, and glimpse some of the breakthroughs just ahead.
How has industrial AI evolved over time?
This blog walks through six key stages in AVEVA’s industrial AI journey:
Predictive analytics and maintenance
2006
Predictive analytics applies machine learning to real-time and historical data to detect patterns and forecast issues, helping operators prevent failures and optimize performance.
Predictive analytics may be one of the older tools in the industrial AI toolkit, and it receives less attention in the news than cutting-edge large language models (LLMs), but it remains one of the surest generators of actual ROI.
Modern predictive analytics combines real-time data and historical data with machine learning (ML) models to identify patterns and forecast future outcomes. It enables industrial operators to anticipate equipment failures, optimize performance, shift to predictive maintenance, and make better operational decisions days, weeks, or even months before an operator might detect an issue.
Today, you’ll find predictive analytics just about everywhere, from the power industry to life sciences, from food and beverage to aerospace and defense. Any industry where anticipating failures, demand, or behavior leads to new value, predictive analytics is already an AI mainstay.
SCG Chemicals: The real-world benefits of predictive analytics
SCG Chemicals operates one of the largest petrochemical supply chains in all of Asia. Any asset failures can quickly lead to a ripple effect of costly disruptions up and down the supply chain. That’s why SCG partnered with AVEVA to create a digital twin of plant operations, which uses predictive analytics to predict asset health, prevent failures, and drive plant-wide reliability to 99%. In just six months, SCG Chemicals’ new platform achieved 9x ROI.
Predictive asset optimization
2022
By combining first-principles simulation models with AI-powered predictive analytics, predictive asset optimization helps operators make the most economic, sustainable operational choices.
Predictive asset optimization takes predictive analytics a step further by combining it with first-principles simulation to provide a 360-degree view of operational risk.
Using first-principles simulation, predictive asset optimization empowers industrial teams to determine how an asset should be performing, then compare that theoretical performance against the asset’s real-world performance. This allows them to isolate underlying asset degradation and cancel out any deviations caused by ambient conditions or changing levels of operation.
Once that baseline is established, predictive asset optimization uses AI-driven predictive analytics to determine whether an asset can reliably operate until the next planned outage.
It can also evaluate whether an immediate shutdown is more cost-effective or whether running at slightly reduced efficiency until the next scheduled maintenance window is the better option. An unplanned stop can be costly, but operating for months at reduced efficiency may be even more expensive. Determining the most economic and sustainable path of action requires the combined power of AI and physics-based simulation.
Gray-box simulation
2023
By combining physics-based and machine learning models, gray-box simulation achieves faster, more accurate, and more flexible simulations.
Traditional process simulation models are built on first principles—known physical equations that define processes within a given piece of equipment. These physics-based models are incredibly accurate when you have all the required input data. The trouble is that industrial organizations rarely have all that data. Equipment data is often incomplete; aging and fouling assets can lead to unexpected effects; complex physics can be too complicated to model.
Gray-box simulation addresses these limitations by combining traditional physics-based models (white box) with machine learning models (black box), anchoring the model in known physics while filling in the gaps with data-driven AI. The result is the best of both worlds: the accuracy and interpretability of physics-based models, and the flexibility, speed, and predictive power of AI.
This hybrid approach allows industrial organizations to achieve faster, near-real-time simulations that behave like physical processes even when data is incomplete or physics alone falls short.
Isu Chemicals: The real-world benefits of gray-box simulation
South Korea-based Isu Chemicals needed to model the performance of a new reactor while also transforming sample assay data into feed and product component structure. To predict reactor yield and catalyst performance and decay, the company deployed a hybrid simulation model built on AVEVA™ Process Simulation and a ML model. Now, across different recipes and operational environments, Isu Chemicals can predict yield with a striking 99.7% accuracy.
Industrial large language models
2024
With industrial LLMs, users can securely find, interpret, and share operational insights simply by talking to their tools.
The AI tools we’ve looked at so far primarily enhance the ways industry designs, operates, and optimizes. Generative AI, however, can actually create. It can respond to queries. It can surface insights from complex datasets.
LLMs like GPT and Gemini are some of the more high-profile instances of generative AI, but now, with tools like AVEVA’s Industrial AI Assistant, generative AI has entered the industrial domain in a safe and secure manner.
AVEVA’s Industrial AI Assistant is a generative AI chat tool, an industrial large language model designed and trained specifically for industrial data. Through one natural language interface, you can ask questions about specific assets, sites, plants, and more. You can create visual displays, or quickly find and summarize information from across applications and data types.
Although the Industrial AI Assistant was released on CONNECT in 2024, its evolution is still ongoing. More on that in a moment.
AI-infused autonomous operations
2025
AI-powered autonomous operations extend automation beyond steady-state control to manage complex transient states like startup, shutdown, and process disruptions.
Industry has been automating steady-state operations for some time now—tweaking set points automatically to maintain a process. It’s a powerful technology, and although some implementations may include an AI/ML layer on the existing control structure, it is not an inherently AI-driven technique.
The power of AI is taking autonomous operations further now to include transient states, like startup or shutdown, major disruptions, changing feed levels, and other dynamic states that are beyond the capabilities of traditional approaches to autonomous operations.
AI autonomous operations are built on the integration of dynamic simulation tools and deep reinforcement learning, then trained on massive synthetic datasets. The resulting automation system is smarter, more flexible, and more powerful than traditional systems. It can reduce start-up and shut-down times, maximize throughput, improve control of batch processes, and stabilize after upsets faster than human operators.
These new, autonomous AI capabilities are still in development, but they’ve emerged from the lab and are right now being tested in real-world proof-of-concept projects with key AVEVA customers.
What is synthetic data?
Unlike real-time or historical operations data, synthetic data isn’t collected from sensors, documents, or other sources. It’s artificially generated data that mimics the statistical patterns, structure, and behavior of real-world data. When combined with real-time sensor data, synthetic data forms the foundation for physical AI like robots or autonomous operations control systems. In an industrial context, it’s essential for a few reasons:
- Privacy: Synthetic data allows us to train AI models without exposing real customer data.
- Scale: We can generate vast volumes of synthetic data rapidly and inexpensively, which we can then use to improve ML models.
- Coverage: Synthetic data is essential in automating transient-state operations simply because transient operations data is scare. Startups, shutdowns, disruptions, and other edge cases are relatively rare, short-lived, and often too dangerous and costly to reproduce for the purpose of data collection. Synthetic data allows us to generate the massive datasets required to train the reinforcement learning models that underpin autonomous operations.
Generative design
2025
Generative design AI accelerates engineering workflows by rapidly generating optimized layouts.
Generative design AI can rapidly generate feasible, clash-free design options using environment-aware AI, accelerating design time by reducing the amount of human effort required to create 3D models.
Truly generative design tools go beyond basic geometry algorithms, actually analyzing design constraints and generating layouts. As the human designer, you just have to set your parameters, then the generative AI tools will create options for you to compare and select from.
Today, AVEVA’s generative design AI, or GenDAI, a new feature in the latest update of AVEVA™ Unified Engineering, focuses on auto-route piping. You can automatically route one pipeline or multiple pipelines simultaneously for a more scalable, efficient approach to design. In the updates ahead, its capabilities will expand from piping to include HVAC systems, structural elements, and more.
The auto-route piping capabilities of AVEVA Unified Engineering can automatically route one pipeline or multiple pipelines simultaneously.
Agentic AI
In development
Agentic AI blurs the line between intelligent tool and trusted collaborator, making advanced capabilities accessible to everyone, not just specialists.
What’s the next big breakthrough for industrial AI?
While generative AI helps industrial teams understand, share, and even create information, agentic AI actually uses industrial information to act on your behalf.
In the not-so distant future, AI assistants like AVEVA’s Industrial Assistant won’t just answer your questions. They’ll carry out your requests. Through a human-like chat interface, you’ll be able to create and task AI agents to perform specific tasks.
For example, say a power plant operator is concerned about a drop in performance in one of the condensers. He can simply ask his Industrial AI Assistant to create a monitoring agent for the specific condenser unit, train it to monitor active power and turbine exhaust pressure based on relevant parameters, and get to the bottom of the issue.
Soon, using AVEVA’s Industrial AI Assistant, operators will be able to create and task AI agents on the fly. Watch the full presentation to learn more.
Together, generative and agentic AI transform what industrial software can do, and, just as dramatically, they also transform the user experience. For as long as we’ve had software tools, learning them was a necessary first step, whether that meant vendor-led SCADA training or the “Introduction to Excel” class down at your local library. Now, the smarter software gets, the more invisible it becomes. Increasingly, instead of learning how to make a tool do what you want, you’ll simply ask for what you need.
This shift does more than remove the friction from day-to-day work. It fundamentally lowers the expertise barrier. Engineers and operators will no longer need to be data scientists to analyze trends, validate assumptions, or interact with complex models.
Compounding progress is accelerating innovation
If the last twenty years have shown anything, it’s that AI innovation doesn’t slow down. It accelerates. Predictive analytics defined the mid-2000s, but here in the 2020s, the breakthroughs are arriving one after another, each leap enabling the next. Today, AI is enhancing how industry works. Tomorrow, it will transform it.
Reference:
Chappell, Jim. 2025. From Data to Decisions: How AVEVA AI Is Shaping the Future of Industry. Presentation at AVEVA World, San Francisco, April 8, 2025.
Related blog posts
Stay in the know: Keep up to date on the latest happenings around the industry.
