CONDITION OF AN AIRCRAFT

Field of the invention

The invention relates generally to the field of aircrafts, whether manned or unmanned, more particularly to a device for detecting abnormal flight conditions of an aircraft.

Description of prior art

Since the last century, an increasing number of manned and even unmanned aircrafts have been developed. Those aircrafts are used for many different purposes such as leisure flying, passenger transportation, goods transportation, surveillance, military operations, etc

In that context, considerable progress has been made towards developing complex on-board systems aimed at increasing flight safety. An exemplary safety recovery system for an aircraft is described in US 2018/01 1 1695A1 . An on-board networked system for detecting anomalous flights is disclosed in US 2018/0288080A1 .

Despite the progress made over the years in terms of on-board avionics and safety systems, there is still a need for efficient, low cost and yet safe systems for detecting abnormal aircraft flight conditions.

Summary of the invention

It is an object of the invention to provide a device and a method for detecting an abnormal flight condition of an aircraft which addresses problems of the known devices.

The invention is defined by the independent claims. The dependent claims define advantageous embodiments.

According to the invention, there is provided a device for detecting an abnormal flight condition of an aircraft, said device comprising:

- acceleration sensors to deliver aircraft acceleration data;

- gyroscopic sensors to deliver aircraft angular data; and - a computing unit configured to perform the following operations: a) during a period of time: i. obtain samples of the aircraft acceleration data (4) and obtain samples of the aircraft angular data (6); ii. compute a flight condition categorization result for said period of time in function of the samples of the aircraft acceleration data and of the samples of the aircraft angular data; and b) decide whether said abnormal flight condition of the aircraft occurs in function of the flight condition categorization result, wherein said device comprises a housing into which are arranged at least the acceleration sensors, the gyroscopic sensors and the computing unit.

In the context of the present invention, the expression “abnormal flight condition of an aircraft” is meant to a designate a flight condition which detrimentally affects the ability of the pilot to maintain proper control of the aircraft and ensure normal flight operation and condition. Abnormal flight conditions may for example be hazardous flight conditions which may potentially lead a crash of the aircraft, such as for example loss of control (LOG), spin or stall of the aircraft.

Due its relatively autonomous nature, the device of the invention can operate without requiring on-board systems pre-existing in the aircraft and connections thereto, in particular to inertial sensors which aircrafts are conventionally equipped with, or even to any remote systems or sensors which are not necessarily located onboard the aircraft. This is directly linked to the fact that the acceleration and gyroscopic sensors as well as the computing unit are all located within the housing of the device so that the device is self-contained.

An additional benefit of this autonomous nature is that the device is much less prone to measurement errors originating from faulty aircraft sensors and equipment, which may for example occur due to such sensors being located off- body (i.e. outside the fuselage of the aircraft) and subjected to extreme conditions such as freezing temperatures for example. These characteristics logically translate into more reliable measurements, which in turn lead to an improved categorization and identification of the abnormal flight conditions of the aircraft.

In some embodiments, the device of the invention is a portable device. This is advantageous as it enables the device not only to operate without connections to external systems or components (onboard the aircraft or remote), but it also allows the device to be transported, installed and uninstalled easily without requiring complex installation steps and without interfering with the existing onboard avionics systems. Accordingly, there is generally no additional certification or safety control required by conventional aviation certification and regulatory agencies (such as e.g. the Federal Aviation Administration in the USA) for the device of the invention to operate in conventional aircrafts.

In some embodiments, the housing further contains an electrical power source to power the device, such as a battery for example. This is beneficial as it further enhances the autonomous (standalone or self-contained) nature of the device, as the latter may operate without requiring any power source provided by the aircraft itself. Further, the device may continue its operation even in case of electric power failure that the aircraft may experience, in particular in extreme flight conditions.

In some embodiments, the housing for use herein is a single housing. This embodiment is beneficial as it further enhances the autonomous nature of the device, its compacity and ease of installation.

In some embodiments, the aircraft acceleration data are selected from the group consisting of acceleration according to an X-axis, acceleration according to a Y-axis, acceleration according to a Z-axis, and any combinations thereof, wherein the X, Y and Z axis are forming an orthonormal referential, and the aircraft angular data are selected from the group consisting of rotation speed around an X’-axis, rotation speed around a Y’-axis, rotation speed around a Z’-axis, and any combinations thereof, wherein the X’, Y’ and Z’ axis are forming an orthonormal referential.

It is to be noted that the X, Y and Z axis may be the same or different from the X’, Y’ and Z’ axis, respectively.

In some embodiments, the operation in a)ii for of computing a flight condition categorization comprises the following operations :

- during the period of time, select a first pattern of the samples of the aircraft acceleration data (4) and a second pattern of the samples of the aircraft angular data (6) obtained during operation a)i;

- feed a pre-trained machine learning model of the aircraft with the said first and second patterns to obtain the flight condition categorization result.

In some embodiments, the pre-trained machine learning model of the aircraft is part of the device, for example part of the computing unit of the device.

In some embodiments, the pre-trained machine learning model is a supervised learning model or a deep learning model. Supervised learning models for use herein include support vector machine models, ensemble models or discriminant analysis models.

Deep learning models for use herein include models based on a convolution neural network (CNN) or on a Long-Short-Term-Memory (LSTM) neural network with categorization output.

Details on how an exemplary pre-trained model may be trained and used will be given hereafter.

Such a device is suitable to detect abnormal flight conditions of an aircraft, such as for example hazardous flight conditions due to a fast and abrupt change of the aircraft attitude and/or altitude. Such fast and abrupt changes, which are among the most complex and challenging to anticipate and detect, typically occur during fast transitions from Visual Meteorological Conditions (VMR or VFR) to Instrument Meteorological Conditions (IMG) or when the aircraft is subject to extreme external forces such as heavy wind, air pockets and/or turbulences. Due to its ability to detect such abnormal flight conditions of an aircraft, the device of the invention is particularly suitable to prevent or at least strongly mitigate the risks for the pilot to experience incapacitating mental state, as it is typically the case when a pilot is confronted to these abnormal flight conditions.

Without wishing to be bound by theory, it is believed that this suitability is due in particular to the combination of the sensors and the computing unit configured as described herein. It has indeed been found that this particular combination of technical features has an excellent ability to categorize and identify these abnormal flight conditions in a fast and accurate manner, yet without requiring complex computing models or high computational power.

In some embodiments, the operation a)ii is preceded by a calculation of at least one feature from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data obtained during operation a)i, and operation a)ii further comprises the following operations:

- during the period of time, select a third pattern of said at least one feature;

- feed the pre-trained machine learning model of the aircraft also with the said third pattern to obtain the flight condition categorization result.

In the context of the present invention, the terms “feature from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data” is meant to designate further technical parameters or functions calculated by the computing unit and based respectively on the obtained samples of the aircraft acceleration data and/or on the obtained samples of the aircraft angular data.

In some embodiments, the at least one feature calculated from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data is selected from the group consisting of : energy of acceleration according to the X-axis, energy of acceleration according to the Y-axis, energy of acceleration according to the Z-axis, energy of rotation speed according to the X’-axis, energy of rotation speed according to the Y’-axis, energy of rotation speed according to the Z’-axis, integration of acceleration according to the X- axis, integration of acceleration according to the Y-axis, integration of acceleration according to the Z-axis, integration of rotation speed according to the X’-axis, integration of rotation speed according to the Y’-axis, integration of rotation speed according to the Z’-axis, and any combinations thereof.

These features have been found to constitute meaningful features to calculate from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data for the purpose of detecting an abnormal flight condition of an aircraft.

Short description of the drawings

These and further aspects of the invention will be explained in greater detail by way of examples and with reference to the accompanying drawings in which:

Fig.1 shows a diagram depicting a schematic configuration of a device according to an exemplary embodiment of the present invention;

Fig.2 shows a flow diagram depicting an exemplary embodiment of the operations performed by an exemplary computing unit for use in the device according to the invention;

Fig.3 shows a flow diagram depicting another exemplary embodiment of the operations performed by an exemplary computing unit for use in the device according to the invention;

Fig .4 shows a flow diagram depicting another exemplary embodiment of the operations performed by an exemplary computing unit for use in the device according to the invention;

Fig.5 shows a graph of exemplary aircraft flight data over time for an exemplary flight;

Fig.6 shows a graph of exemplary aircraft flight data over time for another exemplary flight;

Fig.7 shows a graph of exemplary aircraft flight data over time for another exemplary flight;

Fig.8 shows a diagram depicting a schematic configuration of a device according to an exemplary embodiment of the present invention; Fig.9 shows a flow diagram depicting an exemplary embodiment of the operations performed by an exemplary computing unit for use in the device according to the invention;

Fig .10 shows a diagram depicting a schematic configuration of a device according to an exemplary embodiment of the present invention.

The drawings of the figures are neither drawn to scale nor proportioned. Generally, similar or identical components are denoted by the same reference numerals in the figures.

Detailed description of embodiments of the invention

Fig.1 shows a diagram depicting a schematic configuration of a device according to an exemplary embodiment of the invention. The device (1 ) comprises a housing (2) containing acceleration sensors (3) to deliver aircraft acceleration data (4), gyroscopic sensors (5) to deliver aircraft angular data (6), and a computing unit (7). Optionally, the housing may also comprise an electrical energy source (20) for powering the device, such as a battery for example.

In some embodiments, the device is a portable device. The term portable means that the device can be carried by a person such as a pilot of the aircraft for example.

In some embodiments, the housing is a single housing.

Exemplary acceleration sensors for use herein include, but are not limited to, single three-axis (X, Y and Z) accelerometers or sets of three accelerometers fixed along the three axis X, Y and Z, wherein the three axis X, Y and Z are preferably forming an orthonormal referential.

Exemplary gyroscopic sensors for use herein include, but are not limited to, single three-axis (X’, Y’ and Z’) gyroscopes or sets of three gyroscopes aligned with the three axis X’, Y’ and Z’, wherein the three axis X’, Y’ and Z’ are preferably forming an orthonormal referential.

The X,Y and Z axis may be the same of different from the X’, Y’ and Z’ axis, respectively. Such acceleration and gyroscopic sensors are well known in the art and may for example constitute an Inertial Measurement Unit (IMU) implement as a Micro Electro-Mechanical System (MEMS), such as for example implemented in conventional commercial smartphones.

As will be apparent to those skilled in the art, the computing unit (7) is configured to acquire data provided by the acceleration sensors and by the gyroscopic sensors, and to perform various operations and functions with these data. Exemplary operations include processing and analysing such data, requesting input from a user, providing instructions to the user, generating messages and alerts, triggering other devices such as a parachute deployment subsystem for example, etc.... The computing unit for use herein may execute suitable computer programs, software, algorithms, machine learning models, and may be provided with artificial intelligence capabilities. A suitable computing unit is for example a microcomputer or a microcontroller.

The computing unit for use herein is configured to perform a series of operations as detailed for example in Fig.2 which shows a flow diagram depicting an exemplary embodiment of the operations performed by an exemplary computing unit for use in the device according to the invention.

In this embodiment, the computing unit is configured to perform the following operations: a) during a period of time: i. obtain samples of the aircraft acceleration data (4) and obtain samples of the aircraft angular data (6); ii. compute a flight condition categorization result for said period of time in function of the samples of at least the aircraft acceleration data and the samples of the aircraft angular data; and b) decide whether said abnormal flight condition of the aircraft occurs in function of the flight condition categorization result.

To perform operation a)ii, the computing unit may for example categorize a flight condition as normal if the aircraft acceleration data and the aircraft angular data sampled during said period of time are all within preset ranges, and categorize a flight condition as abnormal if at least some of the aircraft acceleration data and/or some of the aircraft angular data sampled during said period of time are outside these ranges. Such flight condition categorization result may for example be “normal flight condition”, “loss of control of the aircraft (LOG)”, “spin of the aircraft”, “stall of the aircraft”, etc.. .

Generally speaking in the context of the present invention, each flight condition categorization result may for example be attributed a pre-defined numerical value by the computing unit, such as 1 (one) for normal flight, 2 (two) for LOO, 3 (three) for spin of the aircraft, and 4 (four) for stall of the aircraft for example.

Deciding whether an abnormal flight condition of the aircraft occurs in function of the flight condition categorization result is then straightforward.

In some embodiments, an abnormal flight condition of the aircraft may for example be decided in case the flight condition categorization result is loss of control (LOO) of the aircraft, or spin of the aircraft, or stall of the aircraft, or combinations thereof.

In some embodiments, the sampled aircraft acceleration data (4) are selected from the group consisting of acceleration according to an X-axis (Accel X), acceleration according to a Y-axis (Accel Y), acceleration according to a Z-axis (Accel Z), and any combinations thereof, wherein the X, Y and Z axis are forming an orthonormal referential, and the sampled aircraft angular data (6) are selected from the group consisting of rotation speed around an X’-axis (Rotation speed X’), rotation speed around a Y’-axis (Rotation speed Y’), rotation speed around a Z’-axis (Rotation speed Z’), and any combinations thereof, wherein the X’, Y’ and Z’ axis are forming an orthonormal referential.

In some embodiments, as illustrated in Fig.3, the operation in a)ii for of computing a flight condition categorization comprises the following operations : - during the period of time, select a first pattern of the samples of the aircraft acceleration data (4) and a second pattern of the samples of the aircraft angular data (6) obtained during operation a)i; - feed the pre-trained machine learning model of the aircraft with the said first and second patterns to obtain the flight condition categorization result.

As illustrated on Fig.4, in some embodiments, the operation a)ii is preceded by a calculation of at least one feature from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data obtained during operation a)i, and operation a)ii further comprises the following operations:

- during the period of time, select a third pattern of said at least one feature;

- feed the pre-trained machine learning model of the aircraft also with the said third pattern to obtain the flight condition categorization result.

In some embodiments, the at least one feature calculated from the samples of the aircraft acceleration data are for example selected from the group consisting of energy of acceleration according to the X-axis (AccelEnergy X), energy of acceleration according to the Y-axis (AccelEnergy Y), energy of acceleration according to the Z-axis (AccelEnergy Z), integration of acceleration according to the X-axis (Accel Integral X), integration of acceleration according to the Y-axis (Accel Integral Y)„ integration of acceleration according to the Z- axis(Accellntegral Z), wherein :

AccelEnergy X = (Accel X) x (Accel X);

AccelEnergy Y = (Accel Y) x (Accel Y) ;

AccelEnergy Z = (Accel Z) x (Accel Z);

Accellntegral X = trapezoidal integration of (Accel X); Accellntegral Y = trapezoidal integration of (Accel Y); Accellntegral Z = trapezoidal integration of (Accel Z).

In some embodiments, the at least one feature calculated from the samples of the aircraft angular data are for example selected from the group consisting of energy of rotation speed according to the X’-axis (GyroEnergy X’), energy of rotation speed according to the Y’-axis (GyroEnergy Y’), energy of rotation speed according to the Z’-axis (GyroEnergy Z’), integration of rotation speed according to the X’-axis (GyroIntegral X’), integration of rotation speed according to the Y’-axis (GyroIntegral Y’), integration of rotation speed according to the Z’-axis (GyroIntegral Z’), and any combinations thereof, wherein :

GyroEnergy X’ = (Rotation speed X’) x (Rotation speed X’); GyroEnergy Y’ = (Rotation speed Y’) x (Rotation speed Y’); GyroEnergy Z’ = (Rotation speed Z’) x (Rotation speed Z’); GyroIntegral X’ = trapezoidal integration of (Rotation speed X’); GyroIntegral Y’ = trapezoidal integration of (Rotation speed Y’); GyroIntegral Z’ = trapezoidal integration of (Rotation speed Z’).

The inventors have indeed found from experience with flight data acquired from real or simulated aircraft flights, that the aircraft acceleration data (4) and/or the aircraft angular data (6) and/or features calculated from these data reveal specific patterns (i.e. sequences of data and/or sequences of calculated features) which are indicative of normal flight conditions, or of abnormal flight conditions such as loss of control (LOG), spin or stall of the aircraft for example.

Such a pattern is for example shown on Fig. 5 which is a graph of exemplary aircraft flight data for an exemplary normal flight. On this figure are shown aircraft angular data (rotation speed around X’, Y’, Z’) and features calculated from these angular data (energy of rotation speed around X’, Y,’ Z’) as respectively sampled and calculated over the considered period of time and which are considered to correspond to normal flight conditions.

As one can see on Fig .5, all values are well below 1 (one), including for the right part of the patterns which correspond to some degree of turbulence and is considered as a normal flight condition.

Another exemplary pattern is shown on Fig. 6 which is a graph of exemplary aircraft flight data for an exemplary abnormal flight during which the aircraft is spinning. On this figure are shown the values of the same data types as for Fig.

5 (rotation speed around X’, Y’, Z’ , and energy of rotation speed around X’, Y,’ Z’) as respectively sampled and calculated over the considered period of time for such abnormal flight condition. As one can see on Fig .6, many data values are well above 1 (one), which corresponds to a spinning of the aircraft, and is considered as an abnormal flight condition.

Yet another exemplary pattern is shown on Fig. 7 which is a graph of exemplary aircraft flight data for an exemplary abnormal flight during which the aircraft is undergoing Loss Of Control (LOG). On this figure are shown the values of the same data types as for Fig. 5 and Fig. 6 (rotation speed around X’, Y’, Z’ , and energy of rotation speed around X’, Y,’ Z’) as respectively sampled and calculated over the considered period of time for such abnormal flight condition. As one can see on Fig.7, many values are well above 1 (one), which corresponds to a LOO of the aircraft, and is considered as an abnormal flight condition.

As one can also see, the patterns of data of Fig.7 are different from the patterns of data of Fig.6, which permits to distinguish a condition of spinning of the aircraft from a condition of LOO of the aircraft.

To the inventor’s knowledge, LOO is one of the most complex accident categories, involving numerous contributing factors that act individually or, more often, in combination. Considering all the contributing factors, which are very different in each aircraft and highly dependent on its configuration and the external conditions, a generalized approach to the problem of determining that an aircraft is out of control is almost impossible to be solved using classic stochastic theory and algorithms. To date, LOG remains a significant threat and is still the leading cause of fatalities worldwide.

Hence, a machine learning model of the aircraft may be pre-trained with such pre-acquired patterns of flight data to deliver corresponding flight condition categorizations. The pre-training of the machine learning model of the aircraft may be performed according to techniques commonly known in the art, whether supervised or not supervised. The pre-training of the machine learning model of the aircraft is typically performed off-line, for example with the use of a (separate) computer. According to an exemplary training procedure, a large number of flight data are collected from simulated flights (using computer flight simulators) and/or from real flights, whether manned or unmanned. Patterns of these collected flight data, which comprise at least patterns of aircraft acceleration data and patterns of aircraft angular data, possibly complemented with patterns of at least one feature calculated from the samples of the aircraft acceleration data and/or from the samples of the aircraft angular data, are used as input to a machine learning algorithm for training the machine learning model of the aircraft and for categorizing aircraft flight conditions.

The patterns of Fig.5 are for example used for pre-training the machine learning model and for corresponding to a normal flight condition.

The patterns of Fig.6 are for example used for pre-training the machine learning model and for corresponding to an abnormal flight condition, more specifically to a condition where the aircraft is spinning.

The patterns of Fig.7 are for example used for pre-training the machine learning model and for corresponding to an abnormal flight condition, more specifically to a condition where the aircraft undergoes a LOG.

Many other patterns of various types of flight data may be used for pre-training the machine learning model. Obviously, the types of flight data which were used for pre-training the machine learning model of the aircraft are the same as the types of flight data sampled and fed by the computing unit of the device into the pre-trained machine learning model of the aircraft in order to obtain the flight condition categorization result during a real flight of the aircraft onboard of which the device is installed and used.

This learning procedure results into the pre-trained machine learning model of the aircraft, which is then stored into the device (1 ) and thereafter used by the computing unit (7) of the device for the flight categorization during a real flight of the aircraft.

The pre-trained machine learning model of the aircraft is thus part of the device, for example part of the computing unit of the device and may for example be a pre-trained supervised learning model or a pre-trained deep learning model. Supervised learning models for use herein include support vector machine models, ensemble models or discriminant analysis models. Deep learning models for use herein include models based on a convolution neural network (CNN) or on a Long-Short-Term-Memory (LSTM) neural network with categorization output.

In some embodiments, as illustrated on Fig .8, the device or the housing (2) of the device further contains a GPS unit (9) to deliver aircraft positional data (10) and the computing unit (7) is further configured to perform the following operations:

- obtain samples of the aircraft positional data (10) from the GPS unit in operation a)i.;

- calculate at least one feature (12) from the samples of the aircraft positional data (10) in operation a)ii.; and

- feed the pre-trained machine learning model of the aircraft also with the samples of the aircraft positional data (10) and/or with the at least one calculated feature (12) from the samples of the aircraft positional data (10), to obtain the flight condition categorization result for said period of time.

The aircraft positional data are for example the GPS coordinates of the device. In some embodiments, the at least one feature (12) calculated from the samples of the aircraft positional data (10) is selected from the group consisting of aircraft ground speed, aircraft vertical speed, aircraft bearing, and any combinations thereof, and which can conventionally be calculated based on the samples of the aircraft positional data.

In some embodiments, an abnormal flight condition of the aircraft may for example be decided by the computing unit in case the flight condition categorization result is loss of control (LOG) of the aircraft, or spin of the aircraft, or stall of the aircraft, or any combination thereof.

In some embodiments, the computing unit is configured to :

- repeat operation a) a plurality of times, to obtain a corresponding plurality of flight condition categorization results; - calculate a score as a function of the plurality of flight condition categorization results;

- compare the score with a predefined threshold value; and wherein operation b) is replaced by the following operation: b) decide whether said abnormal flight condition of the aircraft occurs in function of a result of the comparison of the score with the predefined threshold value.

Operation a) may for example be repeated a number Ms of times, thereby providing Ms flight condition categorization results, wherein Ms is for example a value comprised between 2 and 100. In some examples, Ms is equal to 100. Each flight condition categorization result may for example be attributed a predefined numerical value by the computing unit, such as 1 (one) for normal flight, 2 (two) for LOG and 3 (three) for SPIN for example.

Those Ms flight condition categorization results may then be summed up or averaged by the computing unit and the result of this sum or average is the score. In case the Ms results are summed up and if Ms = 100, the value of the score is for example equal to 100 (1x100) for normal flight, 200 (2x100) for LOO and 300 (3x100) for Spin. The computing unit may thereafter compare the score with a predefined threshold value. In this example, the predefined threshold value is for example equal to 150 and the computing unit will for example decide that an abnormal flight condition of the aircraft occurs if the score is higher than 150.

An exemplary embodiment of a device calculating said score is shown in Fig.9.

In some embodiments of the device of the invention, operation a)ii. performed by the computing unit as described hereinabove is preceded by an operation of low-pass filtering the aircraft acceleration data and/or of low-pass filtering the aircraft angular data and/or of low-pass filtering the aircraft positional data sampled during operation a)i. This filtering operation may be performed using conventional low-pass filters applied by the computing unit. This embodiment is advantageous since it allows reducing noise which might be produced by the various sensors of the device and/or by external vibrations, as for example from the engine, propeller or other moving parts of the aircraft. In some embodiments, the computing unit comprises a memory into which it stores the aircraft acceleration data, the aircraft angular data and optionally the aircraft positional data, as sampled during the period of time, as well as the features as described hereinabove and as calculated based on these data. The stored values or patterns of values are then fed into the pre-trained machine learning model to obtain the flight condition categorization result. In case this sequence is repeated a plurality of times to obtain a plurality of flight condition categorization results, these results may also be stored into the memory by the computing unit to thereafter calculated the score as described hereinabove. The computing unit then decides whether or not an abnormal flight condition of the aircraft occurs in function of the flight condition categorization result for said period of time or in function of the calculated score for the plurality of periods of time.

In some embodiments, the device or the housing of the device according to the invention further comprises a warning unit (16) as shown in Fig. 10.

The warning unit may for example be selected from the group consisting of visual alarms, audible alarms, sensory alarms, or any combinations thereof. In this embodiment, the computing unit is operationally connected to the warning unit and will activate the warning unit when an abnormal flight condition of the aircraft has been detected.

In some embodiments, the computing unit is configured to generate a recommendation for the pilot of the aircraft when an abnormal flight condition of the aircraft has been detected, for example an audible and/or a visual recommendation, through the waring unit, to indicate that the pilot should activate a parachute launch mechanism (if the aircraft if equipped with such a mechanism).

In some embodiments, the computing unit is configured to generate a recommendation for the pilot of the aircraft when an abnormal flight condition of the aircraft has been detected, for example an audible and/or a visual recommendation, through the waring unit, to indicate that the pilot should activate a parachute launch mechanism (if the aircraft if equipped with such a mechanism), but only if and when an altitude of the aircraft is lower than a predefined altitude, such as a predefined altitude of 600 meters above ground level for example. To this end, the computing unit may for example use the data (vertical or altitude coordinate) from the GPS unit or from an additional barometric sensor to obtain the altitude of the aircraft above ground level.

In some embodiments, the device or the housing of the device according to the invention further comprises a parachute launch triggering unit (17) which is operationally connected to the computing unit, as shown in Fig. 10, and the computing unit is configured to control the parachute launch triggering unit (17) to automatically activate a parachute launch mechanism when an abnormal flight condition of the aircraft has been detected.

In some embodiments, the computing unit for is configured to control the parachute launch triggering unit (17) to automatically activate the parachute launch mechanism when an abnormal flight condition of the aircraft has been detected, but only if and when an altitude of the aircraft is lower than a predefined altitude, such as a predefined altitude of 600 meters above ground level for example (i.e. when both conditions are met).

The parachute launch mechanism itself may or may not be part of the device according to the invention. The parachute itself is a parachute adapted to be mounted on the aircraft and to slow down the fall of the aircraft when it is deployed.

In some embodiments, the device or the housing of the device according to the invention comprises a means accessible to the pilot and to enable him to impeach the device (1 ) to automatically activate the parachute launch mechanism. Such means may for example be a push button arranged on the housing of the device and which, when pushed, sends a signal to the computing unit so that it does not activate the parachute launch triggering unit (17), whatever the circumstances. Such means may alternatively be a push button, such as a conventional emergency button for example, arranged on the housing of the device or detached from the said housing and which, when pushed, interrupts an electrical link or a data link between the parachute launch triggering unit (17) and the parachute launch mechanism.

The present invention has been described in terms of specific embodiments, which are illustrative of the invention and not to be construed as limiting. More generally, it will be appreciated by persons skilled in the art that the present invention is not limited by what has been particularly shown and/or described hereinabove.

Reference numerals in the claims do not limit their protective scope. Use of the verbs "to comprise", "to include", "to be composed of", or any other variant, as well as their respective conjugations, does not exclude the presence of elements other than those stated. Use of the article "a", "an" or "the" preceding an element does not exclude the presence of a plurality of such elements.

The invention may also be described as follows: an autonomous device (1 ) for detecting an abnormal flight condition of an aircraft. The device comprises a housing (2) containing acceleration sensors (3) to deliver aircraft acceleration data, gyroscopic sensors (5) to deliver aircraft angular data, and a computing unit (7) configured to perform specific operations on these data, to obtain a flight condition categorization result in function a result of these operations, and to decide whether an abnormal flight condition of the aircraft occurs in function of the flight condition categorization result.