ACHEMA Start-up Award Finalist: Phaidra

How far can you optimize an existing chemical plant? Most low-hanging fruits have already been picked, and incremental optimization is becoming ever more difficult. Phaidra has set out to bring back the dynamics into the chemical industry by using Artificial Intelligence.

“Artificial Intelligence” is a buzzword that has been around for some time in the process industry. With its giant amounts of data, it seems ideally fit to employ methods that get better the more information is available. Still, closing the gap between the very “physical” chemical and pharma industry with its pumps, reactors and pipes and the “volatile” information industry is not always easy. “The physical world today is filled with static infrastructure. Factories, power plants, and other industrial facilities are frozen in time. They operate the same way they've been operating for decades because their control systems are hard-coded. And by definition, hard-coded systems cannot change — leading to performance degradation and a lack of resiliency,” says Jim Gao, CEO of Phaidra.

Jim Gao and his co-founders Katherine Hoffman and Vedavyas Panneershelvam are taking this challenge on: “Phaidra uses AI to turn static physical infrastructure into intelligent systems. Phaidra creates AI-powered control systems that automatically learn, adapt, and get better over time. Just as your Netflix recommendations automatically improve with watchtime, and self-driving cars automatically improve with driving experience, so too will industrial facilities of the future. Our goal is to become the future of industrial automation.”

Enormous potential of reinforcement learning

The founding team relies on its extensive experience at DeepMind where co-founder Vedavyas was part of the AlphaGo team that beat the world Go champion Lee-Sedol  and where the team reduced the energy required to cool Google’s data centers by 40 %. “This helped us realize the enormous impact that reinforcement learning can have in controlling complex industrial systems.”

At the same time, they are well aware that key to any successful AI applications in the “real-world” is domain expertise. Rather than trying to fit this expertise into their product, they created a solution that builds on the deep understanding of chemical factoring present in the companies: “We therefore created Phaidra as a tool to help domain experts operate their facilities safely and efficiently. Our users - the domain experts - define what they want the AI control system to do  - e.g. maximize yield, reduce energy consumption - , and the operating constraints - e.g. temperatures, pressures. We take care of everything else.”

Industrial sector is overlooked by Silicon Valley

While AI applications are already broadly employed in advertising or by streaming services, innovation hubs like the Silicon Valley are mostly overlooking the “old economy” – despite the fundamental role chemical manufacturing and other industries play for the modern economy. “We wanted to bring the benefits of advanced AI technology to the industrial sector,” says Jim Gao.

However, he and his colleagues are also experiencing the cultural gap between the approaches that rely on autonomous systems and the hardware-oriented chemical industry. As Jim Gao says, “Changing people’s habits has been harder than expected. The industry is used to hard-coded controls systems that cannot change or adapt. The idea of an autonomous AI-based control system that automatically learns and gets better over time is foreign to most people. We need to win people’s trust and show them why AI can not only lead to better plant performance (e.g. yield optimization), but also increase plant safety and stability.”