Achieving Efficient And Effective Ai-Based Predictive Maintenance Capability Recommendations For Military Leaders

In recent years, transformational technologies like artificial intelligence(AI)- based predictive maintenance have shown the potential to significantly reduce unplanned maintenance, increase asset availability, improve inventory optimization, and reduce total cost of operation in military aviation. The U.S. Department of Defense (DoD) has recognized these technologies to be a game changer. The US Air Force (USAF) embraced this approach through its “Condition Based Maintenance+” (CBM+) concept and planned on tripling data-driven maintenance efforts to create a Predictive Maintenance Analytics Loop(PMAL) which helps identify system failures before they occur, representing a monumental shift away from the “fly to fail” approach.Looking at this transformation continuum in aviation maintenance, we proposehow implementationscan benefit froma holistic approach involving three key elements: AI-driven cross-correlation of varied datasets,AI-ready knowledge capture andquality assurance,and risk-profile based prioritization.

Proactive and predictive maintenance capabilities complement reactive and preventative maintenance programs.Predictive maintenance is now enhanced through AI and machine learning (ML), where self-learning models continuallytune themselves to changing conditions over time.For example, analytics of onboard sensor datacan detect overspeed events from evasive maneuvers and enable condition monitoring of engine, auxiliary power unit and hydraulic systems, allowing for early, less expensive repairs instead ofoverhauls or complete failures.AI/ML takes this one step further by detecting complex underlying relationships and identifying failure patterns in components and throughout a fleet—something that would otherwise be difficult to detect. It’s an evolution and a paradigm shift from measuring to calculating, from a descriptive to a predictive approach.

Justin Smith, Amentum’s Executive Director for USAF aircraft maintenance programsand former direct-report to the AFMC Commander, offers a glimpse into a maintenance operation fueled by data-driven insights. “Imagine an algorithm that continually cross-examines F-16 sortie/mission type data from GTIMS and break/maintenance data from IMDS. In short order, this algorithm may find a patternthat mostF-16 ISA breaks occur between 100 and 120 hours of SAT sorties since last Phase. Armed with this information, schedulers can develop a local ISA replacement plan, creating a substantial decrease in ISA breaks.”

Finally, strategically combining different maintenance programs reduces risk while accounting for differences in asset types, maintenance needs and regulatory requirements.

As the DoD recognizes the need for predictability, increased speed and enhancedreadiness, solutions must shift to gain a ‘decisionadvantage’.It is important to definecritical characteristics that would underpin a holistic approach to next-gen AI/ML solutions.

Using AI-driven Cross- Correlation to Analyze Varied Datasets

Traditional predictive maintenance technologies rely on equipment conditionsto predict maintenance requirements, which require configured instrumentation such as sensors. AI/ ML techniquespowered by scalable computing allow for identification ofassociations in non-sensor datasets such as those housed in maintenance and supply management information systems. AI/ML techniquescan aidpredictions by identifying correlations between a largenumber ofconditions. In an aviation maintenance domain, these could range from aircraft components and characteristics, flight hours, sortie/mission types, failure data including “mean time to failure” indicators, quality assurance data, pilot/ maintainerprofiles, environmental conditions and more.While this allows us toextract previously unknown insights, it also allows the sustainer to more effectively manageolder, simpler platforms where onboard data is not available.

The Defense Innovation Unit reported that recent USAF implementations of AI-based approachesdemonstrated the potential for a 3 to 6 percent improvement in mission capability; up to a 35 percentreduction of base-level occurrences of grounded aircraft awaiting parts; and upto a 40 percent reduction in unscheduled maintenance events.

Capturing Meaningful AI-Ready Dataat the Right Place, Right Time

AI systems can be significantly enhanced by capturing more meaningful data at the point of maintenance. In an aviation maintenance environment, inefficiencies and risks arise when maintenance knowledge resides within a small group of experts and is not fully captured or transferred to others.Thisnot only reducesthe level of shared understanding but also ultimately impacts operational readiness.It is essential that maintainers consistently, accurately, and comprehensively document discrepancies, flight configuration (e.g., flaps up, spoilers retracted), and anomalies in maintenance management systems. For example, rather than “the hydraulic pump is leaking” or “engine vibes high,” descriptions that will yield more useful insight would be “primary hydraulic pump is leaking from the pressure side 90-degree elbow” and “engine vibes were at the 1.0 IPS limit during transition from start to ground idle.” Training and quality assurance efforts are necessary to build this into the culture. In fact, AI-based approaches can enable identification of these gaps in data quality, which when mitigated cansignificantly enhance training as well as compliance.

Further, as we encourage better capture of knowledge, it is equally important toextract that knowledge from free-form text fields in in maintenance management systems. Natural language processing approaches (a subset of AI) allow extraction and must be integrated intothe overall solution. This would, in turn,aid verification of diagnostics and enable cross-domain knowledge transfer. These solutions also surface the gaps in captured knowledge, thus acting as a feedback loop to improve data quality, ultimately bolstering theeffectiveness of the predictive AI/ML models.

Scaled, Risk-based, and Agile Approach

Whileinitial solutions using AI-based approaches have shown significant potential, the challenge of scaling these efforts is exacerbated by large volumes of data produced by varied systems and the timeliness with which the information needs to be collected, disambiguated, analyzed and presented. Developing sustainable AI-driven predictive analytics solutions requires time,expertise, and if not carefully developed, significant up-front investment.Taking an agile development approach that is incremental and iterative, prioritizing efforts by keeping in mind the mission relevance and risk profile and adding security controls upfront will be key to the overall success of these programs.

Reducing maintenance time and costs long-term, improving speed and relevance of analysis, and ultimately achieving higher operational readiness requires a holistic approach that looks at not just maintenance approaches and AI algorithms, but also includes data lifecycle management, critical infrastructure view, and upfront operational/risk considerations.

The best solutions are those that complement existing expertise, helping humans understand the tangible, while machines sift through the intangible, transforming bits and bytes into information and knowledge that help forge the edge for tomorrow’s warfighters.

Prachi Sukhatankar is Vice President of Technology Solutions, andSandra Burnis-Holly is Director of Business Development for Amentum, a premier global technical and engineering services partner supporting critical programs of national significance across defense, aviation, security, intelligence, energy, and environment.