The industrial sector is betting big on AI. Globally, 37% of industrial companies are investing heavily in the technology — a figure shared during an IBM-hosted panel discussion and cited by the Singapore Manufacturing Federation. Of those companies, 60% are described as AI leaders.
During the panel, several enterprises shared practical use cases for AI, alongside challenges to implementation and opportunities to scale.
For the Singapore Manufacturing Federation, which comprises over 5,000 members, automation is not a new concept, but the pace of advancements in recent years has accelerated.
“The successful formula in Singapore, to some extent, is the combination of multinational companies and local enterprises working together, because many of our local enterprises are supply chain partners or component partners to the larger MNC ecosystem,” observed Dennis Mark, CEO, Singapore Manufacturing Federation.
Cautious outlook
For industrial automation firm Yokogawa, a generally cautious approach to AI prevails among its customers.
“There are incremental innovations like AI assist, for example, in predictive maintenance. You gather data from the equipment, analyse it, and identify potential weaknesses. Then you adjust the schedule or carry out on-demand maintenance. Those kinds of applications are quite common,” said Savant C S, General Manager, Yokogawa.
Then there’s safety and risk mitigation, where AI is typically used for anomaly detection.
But when it comes to physical threats like fire, Savant noted that AI may not yet be ready to take on such challenges.
“When it comes to the real control and operation of the industrial plant, customers are still very conscious. These are industries which have thrived on the certainty and explainability of physics, thermodynamics, and other similar principles, so it’s very hard for them to really get into a cluster relationship with AI,” he explained.
Most of the processes, Savant said, are tightly controlled, with human accountability and strict adherence to standard operating procedures remaining key concerns — all of which add to the dilemma of adopting AI in such environments.
Despite these concerns, one of the company’s notable achievements was its reinforcement-learning-based AI algorithm called Factorial Kernel Dynamic Policy Programming (FKDPP), co-developed with the Nara Institute of Science and Technology (NAIST). The FKDPP was deployed at an ENEOS Materials chemical plant, where it demonstrated strong performance in controlling a distillation column for about a year.
“The operation of this particular plant’s distillation column is so delicate that even the most expert operator in the world could make mistakes, because the boiling points of these liquids are very close, and even slight variations can cause problems. Therefore, we implemented this project, and it’s been running there for two years now, fully automated, with no involvement from any human,” Savant said.
Data challenges
Going back 10 years, multi-modal public transport operator SBS Transit wanted to extract data from its subsystems so it can analyse them and derive insights.
“Our autonomous train is a very complex system. The train itself is one subsystem. We also have a signalling system, then the track side controls, the power systems, and the integrated supervisory control systems. You can just imagine that the management and maintenance of all the assets behind these systems is a huge task,” said Lim Kok Leong, Vice President, Head of Technology (Mobility), SBS Transit.
However, since its rail system is considered critical information infrastructure (CII), any data extraction efforts must undergo a strict cybersecurity risk assessment. This did not deter SBS Transit from navigating the roadblocks and moving ahead with AI deployment.
“For public transport, the first priority is really minimising downtime. We adopt computer vision AI for inspecting our trains. Basically, that augments the laborious efforts of our engineers and technicians to inspect all the trains,” he said.
Lim also highlighted AIVA, the company’s digital concierge, which assists passengers with their travel queries at Ang Mo Kio Bus Interchange and Punggol Coast MRT station.
Then there’s the Rail Rover on the Downtown line, an internally developed multi-function track trolley that swiftly inspects the tracks for defects to enhance rail reliability and safety.
Finally, SBS Transit, in partnership with IBM, has developed the MaxiMobility solution, aimed at integrating intelligent asset lifecycle management into its rail operations.
“By integrating advanced monitoring and predictive maintenance solutions, SBS Transit’s engineers and technicians will have access to near real-time data and predictive insights, enabling them to identify and address faults or system failures before they occur,” SBS Transit said in a statement.
Industry development
According to Mark, manufacturing comprises about 20% to 22% of Singapore’s gross domestic product (GDP), and the government intends to increase it by 15% by 2030. Investment is therefore focused not just on capabilities, systems, machinery, and infrastructure, but also on people.
“One of our key focuses now is to forecast what skillsets will be needed in the next three to five years. With all this advancement in technology and AI, we must start to train our people towards them. That is one of the key enablements,” he said.
To maximise AI opportunities, the key for enterprises is democratisation, remarked Lim.
“I have my engineer assessing some of the conditions of our assets, monitoring diagnostic information using AI, because it gives operations better insight into asset health,” he said.
Then again, because AI is heavily reliant on data, issues about data privacy and security go hand in hand with innovation.
“If your data isn’t protected and ends up in the open, it can attract unwanted attention, especially toward your proprietary information. That’s why we’re also promoting cybersecurity as a necessary defence. The more data you collect, the more powerful your AI becomes; but at the same time, someone could use AI to gain too much insight into your internal operations,” Mark warned.
He added that whatever qualifies as critical data should be protected — possibly kept on-premises — and safeguarded within the firewall. At the same time, there may be data worth sourcing externally for industry comparisons, which makes a hybrid cloud approach the most realistic option.
According to Mark, organisations need to be clear about which data can reside outside and which should stay on-premises, and plan accordingly. Otherwise, in his words, “by embarking on the data highway, you don’t suddenly have too many wrong vehicles jumping onboard your highway.”
