In a military naval environment, an invisible drift can be enough to compromise an entire mission.
On a multi-mission frigate (FREMM frigate), certain failures leave no second chance. A malfunctioning sonar, even partially degraded, can lead to a loss of underwater detection capability, unplanned operational downtime, or even an early return to port.
In a military context, maintenance cannot rely solely on fixed thresholds or theoretical schedules. Embarked systems must remain available, reliable and high-performing over long life cycles, in extreme and evolving environments.
This use case demonstrates how to detect early drifts in a critical embarked sonar system, without exploitable failure history, without disrupting operations, and while providing information that is genuinely actionable for the maintenance and support chain.
Operational context: a sonar at the heart of anti-submarine warfare missions
The European Multi-Mission Frigate, better known by its acronym FREMM, is equipped with a high-performance sonar designed to:
detect and track underwater threats, including hostile submarines
conduct detailed analysis of the acoustic environment
map the seabed and identify obstacles, terrain features or mines
Unlike radar, sonar relies on the emission and reception of acoustic waves, which are highly sensitive to environmental conditions. It is a complex electronic and software system composed of acoustic sensors, embedded electronics, mechanical actuators and advanced signal processing capabilities.
The operational continuity of this sonar is strategic. A gradual, undetected degradation can affect detection performance long before a clear-cut failure occurs.
Specific challenges in maintaining critical naval systems
The FREMM sonar operates under particularly demanding conditions:
prolonged immersion
mechanical shocks
high salinity
significant thermal variations
acoustic and electromagnetic interference
In this context, failures are rarely abrupt. They most often manifest as progressive drifts, difficult to detect using traditional approaches.
Maintenance and support teams also face several major limitations:
absence or scarcity of failure data
normal behaviour varying according to operational conditions
fixed thresholds ill-suited to dynamic environments
difficulty in precisely locating the root cause of a degradation
The result is often reactive maintenance, or conversely costly over-maintenance, with no guarantee of preventing critical failures.
The limits of traditional approaches
Conventional strategies rely on predefined thresholds, static rules and scheduled intervals.
On a multivariate system such as an embarked sonar, these methods quickly reach their limits:
inability to detect weak signals
multiplication of false positives
difficulty distinguishing normal variation from genuine drift
strong dependence on human expertise
In Defence, unavailability is not an option. It is necessary to move from a simple alert logic to a true decision-support logic.
The DiagFit solution: AI designed for the maintenance and support chain
Unsupervised learning tailored to Defence environments
DiagFit is based on an industrial AI approach designed for contexts where failures are rare and poorly documented.
The principle is straightforward:
the model learns solely from healthy operating data
it builds a reference model of the system’s nominal behaviour
any significant deviation from this behaviour is automatically detected
No failure data is required for training, which is a decisive advantage in military environments.
From alert to actionable information
The objective is not to generate yet another alert.
The objective is to produce contextualised information that can be used directly by the maintenance and support chain. For each detected drift, teams are provided with:
a precise indication of the system or sub-system concerned
a severity level
a time evolution view to assess the dynamics of the degradation
This structuring transforms detection into a genuine decision-making tool. It enables informed choices regarding:
continued operation
system reconfiguration
planned intervention
enhanced monitoring
Data becomes an objective basis for decision-making, rather than an additional alarm.
Multi-level health analysis of the sonar
DiagFit was used to analyse the system at two complementary levels:
global level: assessment of the overall health state of the sonar
local level: sensor-by-sensor analysis to precisely identify sources of drift
This approach enabled maintenance teams to:
detect early anomalies
understand which sub-systems were involved
prioritise corrective actions
support diagnosis with objective, data-driven evidence
Results achieved
Reliable early detection, even without failure history
On the FREMM sonar, DiagFit demonstrated its ability to:
detect 100 percent of the available labelled faults
identify additional unlabelled anomalies, revealing previously unnoticed degradations
operate without failure data for training
These results confirm the robustness of the approach in complex, multivariate and critical environments.
As the volume and diversity of available data increased, the model was progressively refined, reducing false positives and further improving detection accuracy.
Deployment compatible with Defence constraints
In the military sector, sovereignty and data security are paramount.
The solution is compatible with the following requirements:
on-premise deployment
non-intrusive integration into existing architectures
compatibility with closed systems
processing of potentially classified data
full traceability of analyses
No reliance on a public cloud is required. The system can be integrated into constrained environments in full compliance with Defence security requirements.
Operational benefits for Defence systems
Using DiagFit on an embarked sonar makes it possible to:
anticipate drifts before they impact the mission
improve the operational availability of critical systems
reduce unplanned downtime
strengthen the reliability of maintenance decisions
limit reliance on static thresholds and rigid rules
DiagFit therefore provides a predictive maintenance capability that is genuinely usable in demanding Defence contexts.
Going further
Are you operating critical systems subject to similar constraints, such as embarked systems, naval equipment, complex sensors or harsh environments?
👉 Let’s discuss your predictive maintenance and anomaly detection challenges in Defence environments.
