AI in MRO and predictive maintenance: regulatory considerations (FAA/EASA official material)

Official FAA and EASA-hosted pages discuss maintenance data, health monitoring and, in one research report, machine-learning remaining-useful-life. They do not publish an MRO-specific AI regulation. This article keeps those documents in their own words.

Status as of 14 October 2026: the FAA Roadmap HTML slug now returns the Version I PDF (HTTP 200). NASA ProgPy still describes an open-sourced python package supporting research and development of prognostics and health management and predictive maintenance tools. AC 43-218 and ForMuLA PDFs were not re-extracted as text this pass; those claims stay from the last editorial search (no AI/ML tokens in AC 43-218; ForMuLA still shall not constitute any binding advice).

What the FAA AI roadmap says about maintenance

The Roadmap for Artificial Intelligence Safety Assurance, Version I includes a Maintenance example: Many maintenance concepts already employ a digital twin, whereby a model of the component is represented in a virtual context that receives real inputs. When the performance of the real system varies from the digital twin, a call for preventive maintenance can be made. An AI application can conduct a thorough comparison, considering multiple factors beyond predefined triggers, particularly when the system is allowed to learn over time.

Under Retirement Phase, the same roadmap says AI can aid in predictive maintenance, optimizing the decommissioning and recycling of aircraft parts, and managing assets efficiently. Those sentences sit in a living policy roadmap. They are not an advisory circular for maintenance organizations. See the FAA roadmap explainer.

AC 43-218 is IAHM guidance; a search found no AI words

AC 43-218 (7 July 2022; editorial update 19 December 2025) addresses Operational Authorization of Integrated Aircraft Health Management Systems. It says Aircraft Health Monitoring (AHM) for maintenance uses onboard sensors, data transmission, and data analysis to provide information regarding aircraft system performance and structural condition. The audience listed in the circular includes maintenance, repair, and overhaul (MRO) organizations and Part 145, Repair Stations. It is an acceptable means, but not the only means, to comply with the cited 14 CFR sections.

We searched the editorial PDF for “artificial intelligence”, “machine learning”, “AI ” and “predictive” and found none of those tokens. We cite AC 43-218 as IAHM and MRO health-management guidance, not as AI guidance.

ForMuLA is EASA-hosted research, not a regulation

EASA hosts the public extract of Formal Methods use for Learning Assurance (ForMuLA) (April 2023, Version 1.0), an Innovation Partnership Contract report by EASA and Collins Aerospace. The disclaimer says information in the report shall not constitute any binding advice, nor they shall create or be understood as creating any expectations with respect to any future certification or approval whatsoever.

The use case is a deep learning-based estimator for remaining useful life of mechanical bearings in airborne equipment. The output of the estimator is used for on-ground maintenance applications. RUL estimation discussed in the ForMuLA IPC is performed for a mechanical bearing component installed in the drivetrain assembly of a rotorcraft. Intended users include the MRO (Maintenance, Repair and Overhaul) team, the CAMO (Continuing Airworthiness Management Organization) team, and the pilots. The report says the use case maps onto Level 1A and Level 1B in the EASA classification because estimated RUL does not automatically drive the maintenance and inspection tasks, but only supports the human user in taking a related decision.

On-ground Condition-Based Maintenance (CBM) is described as moving from preventive maintenance to predictive maintenance. That is the research ConOps, not an EASA implementing rule.

A NASA research package, not an FAA or EASA tool

NASA’s ProgPy is an open-sourced python package supporting research and development of prognostics and health management and predictive maintenance tools. The Prognostics Center of Excellence addresses prognostic technology gaps within the application areas of aeronautics and space exploration and lists Validation and verification of prognostics among its challenges, using machine learning among other methods. Those pages do not name ProgPy as an FAA or EASA acceptable means of compliance. The listing is on the ProgPy tool page.

We do not add an “MRO” industry slug. No official taxonomy we cite uses that as a sector tag, and AC 43-218 is already aviation health-management guidance.

Questions

Is there an FAA advisory circular for AI in MRO?

AC 43-218 is the IAHM circular we cite. A search of that PDF found no “artificial intelligence” or “machine learning” wording. The FAA AI roadmap mentions digital-twin preventive maintenance and retirement-phase predictive maintenance as examples, not as an MRO AC.

Does ForMuLA approve autonomous maintenance?

The public extract says estimated RUL does not automatically drive the maintenance and inspection tasks, but only supports the human user, and that the report shall not constitute any binding advice or create expectations of future certification or approval.

Did we invent an MRO industry filter?

No. Tools that name aeronautics, aviation or space on their official pages are tagged with those existing industries. ProgPy is tagged from the PCoE aeronautics and space sentence.