AI in industrial control systems

The story of AI in industrial control is a story of engineers learning to handle complexity. Early systems captured the judgement of experienced operators in rule-based logic, helping plants monitor conditions and respond more consistently. Fuzzy logic made it easier to deal with messy, real-world conditions where things are not simply on or off, high or low. Neural networks added the ability to learn patterns from data, opening the door to better prediction and more adaptive control. None of this removed the need for good engineering. If anything, it has made that need more obvious.

What is changing the game today is the amount of information industrial systems produce and the new type of processing ability. Sensors, cameras, connected equipment and cloud platforms are generating a constant stream of operational data. That gives AI far more to work with than it had in the past. It can help spot faults earlier, predict maintenance needs, improve performance and create digital models that let engineers test ideas before touching the real plant. In simple terms, control systems are becoming more aware, more connected and more capable of supporting better decisions.

AI tools are starting to assist with writing programmable logic controller code, generating operator screens and speeding up parts of automation design that used to take a lot of human effort. On paper that sounds brilliant, and in many cases it will be. But there’s a catch. Fast is not the same as safe. Something can look polished, compile correctly and still contain a serious flaw. In industrial control systems, a missed interlock, a poor alarm response or a hidden cybersecurity weakness is not just a software bug. It can quickly become a plant, safety or reliability issue.

Even before AI, integrators had tools that could automate some of these tasks with deterministic outcomes. Functional safety is especially important here, because safety-related control functions must perform predictably and be designed, verified and maintained to defined standards. AI may support engineering work, but it cannot replace the rigorous lifecycle, independence, experience and evidence required to show that a safety function will act correctly when needed.

Engineering rigour, therefore, matters as much now as ever. As AI becomes easier to use, professional responsibility becomes more – not less – important. Competent people need to be involved in decisions about where AI should be trusted, how its outputs should be checked and who remains accountable when something goes wrong. The real issue is not whether AI will replace engineers. It is whether it could outrun the checks and balances that make engineering safe, reliable and worthy of trust.

What stands out most is that the biggest challenge ahead is not technical capability, but governance. Industrial control systems operate in the real world. They move energy, water, chemicals and other services people rely on every day. If AI is introduced without careful validation, human oversight and clear rules, the problem is not just poor software. It can become an operational, cybersecurity, or – in the worst cases – safety problem. The sensible path is not to reject AI, but to use it as a tool inside a disciplined engineering framework.

AI in industrial control systems is not new, so it should be judged with perspective. The combination of AI with connected industrial data is already changing how systems are monitored, maintained and improved. The next wave of AI-assisted engineering will almost certainly speed things up, but it will also raise important questions about competence, verification and security. The most useful future is one where AI expands what engineers can do while competent people, robust standards and clear accountability remain firmly in place.

Chris Bonner CMEngNZ CPEng IntPE(NZ) is Principal Control Systems Engineer at electricity generator and energy retailer Mercury.

This article was first published in the September 2026 issue of EG magazine.