Constructing statistical models and performing data clustering applied to technical sensor data: non-technical

The application underlying the discussed decision concerns a method and system for monitoring the health of helicopters by determining the severity of flight missions and scheduling maintenance accordingly. The key features at issue were the construction of mission types from flight data descriptors using data clustering, the estimation of severity models per mission type, and the programming of maintenance operations based thereon. The Board considered these features as an abstract modeling method that did not produce a credible technical effect and was therefore non-technical.

Here are the practical takeaways from the decision: T 1360/22 (Santé d’hélicoptères/SAFRAN) of 6 October 2025, of the Technical Board of Appeal 3.5.01.

Key takeaways

Abstract data modeling steps such as constructing statistical models and performing data clustering, even when applied to technical sensor data from helicopters, do not automatically produce a technical effect. When the claim lacks sufficient detail to demonstrate a credible technical improvement over the prior art, such features cannot support inventive step.

The invention

The Board of Appeal summarized the invention as follows:

Helicopters are designed and used for different types of missions (emergency medical service, utility, tourism, VIP, maritime, police, etc.), each having a different influence on the condition of the helicopter and its components. Traditionally, mission types are determined during the design phase based on theoretical profiles established by the manufacturer through client consultation, which may not reflect actual usage. The invention allows the determination of mission types based on flights actually performed by multiple helicopters, thereby deriving mission types from real usage rather than assumed usage. The method involves acquiring flight data (physical data from sensors) and maintenance data (component failures, replaced components) for a fleet of helicopters. Physical flight data are reduced to vectors of predetermined dimension (descriptors), which are then partitioned into subsets forming mission types. Severity models are estimated for each mission type based on both flight and maintenance data, defining component aging estimations per mission type. For maintenance purposes, the actual mission type performed by a specific helicopter is determined from its flight data, and maintenance operations are programmed based on the associated severity model.

Subsidiary Request, Claim 1

A method for maintaining a helicopter, characterized in that it comprises:

a severity determination step, comprising:

a step of acquiring and storing flight data from helicopter flight missions, said flight data comprising, for each flight of a helicopter, the physical data recorded by at least one sensor of the helicopter,

a step of acquiring and storing maintenance data from the plurality of helicopters, said maintenance data comprising at least information relating to component failures of each helicopter and components replaced in each helicopter as a consequence of the flight missions,

a mission type construction step (10), comprising:

a sub-step (14) of constructing relative descriptors, in which the physical flight data are each reduced to a vector of predetermined dimension forming a descriptor, all descriptors having the same dimension,

a sub-step (18) of partitioning the descriptors, adapted to partition the descriptors into subsets forming the mission types,

a sub-step (20) of assigning a mission type to each flight by associating the descriptor of said flight with the subset in which this descriptor is found, and creating a mission type model associating physical flight data with each mission type,

a severity interpretation step (30) for the mission types, comprising:

a sub-step (32) of estimating severity models from the flight data and the maintenance data, each severity model defining an estimation of component aging of the helicopters as a function of the mission types,

a sub-step (34) of associating a severity model with each mission type determined in the mission type construction step (10),

a step of determining the mission type performed by the helicopter from the flight data of the helicopter,

a step of programming maintenance operations based on the severity model associated with said determined mission type, and

a maintenance step according to said programming of maintenance operations.

Is it patentable?

The Examining Division’s position

The examining division rejected the application on the ground that the subject-matter of the claims of both the main request and the subsidiary request lacked inventive step (Article 56 EPC). The division considered the distinguishing features over the prior art, in particular D1 (US 2010/0235108), as constituting the implementation of a plan falling within the exercise of intellectual activities on known technical infrastructure. D1 disclosed a helicopter maintenance method based on estimating component aging from actual usage data, including flight regimes and usage cycles recorded by sensors. For the main request, the division also relied on D2 (GB 2536766), D3 (US 2010049379), and D4 (US 2012179326). In essence, the division found that all features distinguishing the claims from D1, namely the construction of mission types via data descriptors and clustering, and the estimation of per-mission-type severity models, were non-technical and could not contribute to inventive step.

The Appellant’s arguments

The appellant argued that the distinguishing features were indeed technical. In particular, the appellant submitted that these features enabled the creation of digital data representative of physical data (i.e., flight data from sensors), the obtaining of coherent subsets to form mission types, and the estimation of helicopter component aging, all of which should be considered technical effects. The appellant further contended that D1 could not serve as closest prior art because it only disclosed a single static model based on average fleet usage, where the estimated wear evolved identically over time for all helicopters. According to the appellant, the invention’s mission-type-specific severity models allowed a more personalized and usage-adapted aging estimation. The appellant also argued that the invention’s models were dynamically updated based on measured data, unlike D1’s allegedly static approach with flight time as the only parameter.

The Board’s analysis

Prior art assessment (D1)

The Board found that D1 discloses a helicopter maintenance method where component wear is determined from the actual usage data of a specific helicopter (usage monitor counts and usage profiles), multiplied by a factor (UMRF) derived from fleet-wide average usage to increase estimation accuracy.
The Board disagreed with the appellant’s characterization of D1 as static. The D1 model is also dynamic, as fleet usage data are transmitted daily and used to determine the UMRF factor. Moreover, D1 accounts for specific flight regimes actually encountered during each flight, not just flight time.
The Board also noted that the claim wording itself did not reflect the dynamic model updating alleged by the appellant during oral proceedings.

Distinguishing features

The Board identified the following features as not disclosed by D1: acquisition of maintenance data from the reference fleet; construction of mission types via descriptors and data clustering; estimation of severity models per mission type from flight and maintenance data; determination of the actual mission type and programming of maintenance accordingly.

Technical character of the distinguishing features

The Board acknowledged that determining the condition (in particular, the aging) of a component and programming maintenance operations accordingly constitutes a technical problem. However, this does not automatically render every feature or constraint used for that purpose technical.
Applying the established COMVIK approach (T 641/00), the Board held that only features that credibly produce a technical effect across the entire scope of the claim can support inventive step. This is particularly difficult to demonstrate when features are expressed in a very general and abstract manner, without a direct functional link to a specific technical context.
The Board held that, as such, the mere construction of statistical models and data clustering (partitioning) according to proximity or similarity criteria do not imply any technical effect, citing T 1635/19.
Crucially, the Board could not identify any credible improvement, in terms of precision of aging estimation or maintenance efficiency, resulting from developing different severity models per mission type compared to D1’s approach based on individual flight regimes. The claim contained no details regarding the physical data recorded by the sensors or the partitioning criteria used to define the mission types.
The Board therefore concluded that the features concerning mission type construction from descriptors, severity interpretation, mission type determination from flight data, and maintenance programming based on the associated severity model represent an abstract modeling method that does not reflect any technical consideration about the underlying technical systems and does not provide a credible technical effect.
Using historical maintenance data from the helicopter fleet for developing aging models was considered an obvious step for the skilled person.
D1 already discloses the programming of component removal, and thus maintenance operations, based on estimated remaining component life.

Main request

The main request was formulated more broadly than the subsidiary request. The Board therefore concluded that the same finding of lack of inventive step applied a fortiori.

Conclusion

The Board dismissed the appeal, finding that all requests lacked inventive step (Article 56 EPC). The distinguishing features over D1, namely the construction of mission types via descriptor partitioning and the estimation of per-mission-type severity models, were considered an abstract modeling method devoid of any credible technical effect. Although the invention processes technical data from helicopter sensors and addresses the undeniably technical problem of maintenance scheduling, the abstract and generic formulation of the data processing steps in the claims failed to demonstrate any tangible technical improvement over the prior art. This case illustrates that operating on technical data and serving a technical purpose is not sufficient if the claimed data processing steps themselves lack a credible, demonstrated technical effect within the scope of the claim.

More information

You can read the full decision here: T 1360/22 (Santé d’hélicoptères/SAFRAN) of 6 October 2025, of the Technical Board of Appeal 3.5.01.

Stay in the loop

Never miss a beat by subscribing to the email newsletter. Please see our Privacy Policy.

Privacy policy Yes, I consent to the collection, processing and use of my above-mentioned personal data for the purposes of processing my message and for the purposes of contacting me via email. The legal basis of the processing shall be formed by my consent pursuant to Art. 6 (1) lit. a GDPR. The data will be deleted three months after expiry of the purpose, provided that longer retention periods are not required by law. I can revoke this consent with future effect at any time. I have taken note of the privacy statement and consent to it. With regard to the processing of my data, I am entitled to inalienable rights, information on which can be found in the privacy statement.

* = Required field