A legitimate use of AI has emerged in predictive maintenance, where the symptoms produced by a component are analysed over time to identify patterns of wear and degradation.

It is thought that using an AI trained on data from everyday smooth running, it will be possible to identify outliers and warning signs early, enabling parts to be replaced at precisely the right time.

The hope is that if a part is diagnosed early, a replacement can be sourced, a repair booked during an existing downtime interval, and that the worst-case scenario of a catastrophic failure – leading to huge costs and long downtime – can be avoided.

In October 2025, Opearl LNG Ship Management, based in Hong Kong, made an agreement with Wärtsilä to provide AI-enabled predictive maintenance on 14 LNG carriers, combining its Dynamic Maintenance Planning (DMP) and Expert Insight (EI) predictive maintenance platform. The system uses AI to hunt for anomalous readings which could indicate potential failures.

“We currently manage tight delivery schedules and require operations with minimal downtime and reduced maintenance interruptions,” said general manager Captain Nomura, OPearl LNG Ship Management. “This long-term agreement with Wärtsilä is intended to support these operational requirements and assist us in reliably meeting our delivery commitments to our customers.”

Shortly afterward, in January of this year, Wärtsilä signed another agreement, this time for 12 LNG Carriers with MOL Global Ship Management and incorporating both DMP and EI.

“Wärtsilä’s Lifecycle Agreement will optimise our vessel operations and maintenance, ensuring that we can maximise uptime and performance,” said Namit Mathur, director, MOL Global Ship Management. “In addition, this agreement will play a crucial role in supporting the sustainable operations of our fleet by helping us reduce emissions and operate more efficiently.”

Semi-submersible crane vessel Saipem 7000, in partnership with BIP, and ultra-deep-water drillship Saipem 12000, in collaboration with rig assurance company ADC, are now having essential systems monitored by AI predictive maintenance systems.

Rigged with networks of internet-of-things (IOT) sensors, the vessels are watched for early signs of wear or degradation in shipboard components.

As a drillship, maintenance downtime on Saipem 12000 is extremely costly. But in the low-tolerance context of deep-water drilling, the vessel incorporates various systems which could lead to severe negative consequences if there is a critical failure in an important component.

In 2010, some 11 workers were killed and 4.9m barrels of oil discharged into the Gulf of Mexico after the failure of a blowout-preventer (BOP) on ultra-deep-water semi-submersible drilling rig Deepwater Horizon. An internal BP audit months before the explosion revealed that 3,900 maintenance tasks were overdue, including on the BOP, and that deferred maintenance – of the sort Saipem 12000’s AI predictive maintenance system could help to avoid – was an endemic issue.

But the industry should take care that it does not over-rely on AI, and allow human inspection, maintenance and repair skills to atrophy. In recent studies of AI predictive maintenance by Lloyd’s Register, it was found that AIs have not only supposed degradation where none existed (a ‘false positive’); but have overlooked problems when they have occurred (‘false negative’).

As usual, the accuracy of models will improve over time when there is more training data present, but shipowners should not assume pinpoint accuracy from the outset. Particularly on such critical vessels as drillships, operators must resist the urge to neglect maintenance of a crucial wear part, on the say-so of an AI.