In February 2025, Schneider Electric received a patent for technology that uses AI to help prevent process safety hazards. The EcoStruxure Triconex Safety solution automatically or semi-automatically analyzes potential hazards in industrial processes in an attempt to prevent them. The project is part of Schneider’s initiative to bolster functional safety through artificial intelligence.
To get a better understanding of the bigger picture, from employee buy-in to data models, Schneider Electric’s senior director of offer management Chris Stogner provided Design & Development Today some insight. Stogner oversees Schneider’s AI advancements, the interview has been edited for length and clarity.
How will AI identification be fully implemented and used?
Artificial Intelligence has been dominating the news cycles for the last several years and there seems to be no stopping the momentum.
AI will touch most parts of human life and there is much speculation of what it will eventually be able to do. In heavy industries, manufacturing and automation companies are making substantial investments in order to create new and smarter ways of working. This is driven in part by the continuous need to drive efficiency, but also to address needs, like staffing challenges and employee safety.
Owner/operators in chemical, [oil and gas], and other heavy industries face challenging demographics in that there is an imbalance in the age of employees. There aren’t enough under 55 employees to replace those who will be retiring over the next 10 years. This problem can’t be solved by adding more people, so it must be solved by adding innovations—and this is where AI will be put to use.
Have there been any issues with employee buy-in?
Change management is always a challenge, however there is a willingness to adopt AI and other tools stemming from a digital transformation in process automation platforms. This isn’t necessarily the case when it comes to process safety. People who are specialists in this field tend to be very conservative in adopting new innovations, and with good reason. In a field where consequences of failure can lead to multiple fatalities, it’s better to get it right than to get it fast.
However, the skilled workforce demographics are particularly challenging in process safety. There simply won’t be enough subject matter experts in this field in five to 10 years. In this case, new ways of working must be adopted in order to maintain safety. This is not a situation where the goal is to replace the human with a machine. People on these teams are going away through natural attrition. Rather, it is a situation in which AI will mitigate a lack of expertise and help maintain a level of safety which guarantees workers are safe in these critical and often hazardous industrial environments.
What data model is the data model built upon?
Schneider Electric’s patented AI model is built on reinforced learning. It is intended to automate a HAZOP (Hazards and Operability Study). This is a very time intensive process involving many subject matter experts in which they assess what could potentially go wrong in a plant environment and what must be done to prevent these undesirable outcomes. The tool works by having two agents, one trying to drive the plant to an unsafe state, while the other is trying to prevent it from going to the unsafe state.
As opposed to supervised learning or generative AI, reinforced learning has the benefit of not needing to be trained in advance. The agents try random changes to the system and either get a reward or not, depending on whether the goals were met. To contrast, imagine AI being applied to a game of chess. A supervised model would need to have some level of chess strategy programmed into it. In an unsupervised model, the agents try random moves and learn. Over numerous games, the agent would always know based on how the board was configured which move leads to the highest likelihood of winning the game.
Through this process of automating the HAZOP, after numerous iterations the plant hazards can be defined as well as the most effective mitigating layers of protection.
What’s the next step?
Given the consequences of a potential failure, trust will need to be developed in AI tools to apply them to process safety work processes. This trust will take time, which means that we must act soon to take advantage of the time we have left before significant numbers of process safety experts retire. In the next few years, an AI tool can be used to check the results from the traditional HAZOP human-led methods. There will always be a need for someone competent in functional safety to sign off on the results, but once the tool is proven out then in the future that person can then sanity check the tool, rather than vice versa.
There is a perception that since HAZOPs are conducted by subject matter experts that it is an objective process. In reality, it is highly subjective, and different teams can come up with widely different results when assessing identical designs. This has proven out in large enterprises who have many near identical process units. Humans bring bias and personality into a process, whereas the AI will not. By combining the efforts and benefits of human expertise and artificial intelligence, we will ensure that industrial environments can continue to be both profitable and safe in the decades to come.