Multi-sensor system for measuring acceleration

Abstract

Multi-sensor system (1) and method for measuring acceleration, including: one or more primary sensors (4) for measuring acceleration according to one or more axes of a spatial reference system (X, Y, Z); one or more secondary sensors (5), each configured to measure one or more physical quantities or parameters associated with the environmental conditions wherein said primary sensors (4) operate; one or more removable actuation means (8), configured to cause a variation of the acceleration measured by said primary sensors (4) and / or, for each of said one or more secondary sensors (5); a central unit (6) configured to process said data detected by said primary sensors (4) and said one or more secondary sensors (5) by means of Machine Learning techniques, to determine a correlation algorithm between: the data detected at said first predetermined instants of time by said one or more primary sensors (4) and secondary sensors (5).

Description

[0001] MULTI-SENSOR SYSTEM FOR MEASURING ACCELERATION

[0002] DESCRIPTION

[0003] Technical field of the invention

[0004] The present invention relates to a multi-sensor system for measuring an acceleration, for example the inertial or gravity acceleration, which allows to reduce or to remove the disturbances deriving from the particular environmental conditions of the detection site, which occur at time of the measurement itself.

[0005] To this purpose, the system is configured to carry out acceleration measures by accelerometers and to detect the values of physical quantities associated to the environmental conditions in which such accelerometers operate, and to use Machine Learning techniques to analyse all data collected with the purpose of constructing, in an automated way, an analytical correlation model between the values of the physical quantities associated to the environmental conditions and their influence on the contextually detected acceleration value. Once devised, for subsequent measurements, the model allows to obtain a posteriori the ‘correct’ value of the measured acceleration, according to the values of the physical quantities associated to the environmental conditions present at time of the acceleration measurement.

[0006] The herein proposed multi-sensor system can be applied in the most diverse scopes of the art, for example it is suitable for space, submarine, air use or as seismometer (for example lunar seismometer). Background

[0007] Up to now, several devices, systems and methods are available to the state of art in order to perform accelerometer measurements, considering the plurality of their possible uses.

[0008] For example, the measurements performed by accelerometer are essential to the purpose of performing gravimetric surveys, consisting in measuring and mapping the variations in the Earth’s gravitational acceleration. Such mappings provide precious information about the variations in mass density of the sub-soil, and then on its composition and morphology. Another example of using the accelerometers relates to the measurement of the inertial accelerations thereto a space satellite is submitted. In this case, an accelerometer on board a space satellite is useful since it provides crucial information for the control, stability and safety of the satellite, as well as for the study of the space environment in which the satellite is placed.

[0009] The different solutions currently available to the state of art can be based upon the use of various types of accelerometers. The most commonly used accelerometer is the “relative" one, but there are also absolute accelerometers, such as the absolute gravimeters. The latter, typically, cannot be transported and they are used to provide a reference value in fixed stations. As far as the relative accelerometers are concerned, they typically consist in a test mass, subjected to the force that one wants to measure, and an elastic element, exerting a restoring force to keep the mass in equilibrium. In such system, the variations in the restoring force correspond to a variation in the position of the test mass. By measuring the variations of this position, it is possible to determine the force variation which is associated to the acceleration.

[0010] It is to be noted that for the inertial accelerations and gravity, the measured force is proportional to the acceleration (for example, the force of gravity is the product between the acceleration of local gravity and the test mass of the gravimeter). When the accelerometer is configured and used to measure the gravity acceleration, it is called gravimeter.

[0011] The most traditional gravity field prospecting method consists in transporting a gravimeter in pre-established measuring stations and performing the measurement on site. Such method provides very accurate measurements, but it is very expensive in terms of time and often it does not allow the measurement in remote places, difficult to be reached. Moreover, if one wishes a high spatial resolution mapping, it is necessary to increase the number of stations and, consequently, the time used to perform the prospecting.

[0012] To obviate this problem, one can instal the gravimeter on a movable platform (for example, aircraft, helicopter, ship, submarine) that flies over or navigates the study area, or to hook it to the latter, even if externally. In this way, large surfaces can be covered in less time and with greater spatial resolution.

[0013] Unfortunately, such advantage is at the expense of the measurement accuracy, which would require measurement systems far more complex than those used on the ground. In fact, the movable platform is a non-inertial reference system in which the masses are subjected to inertial (fictitious, or apparent) forces not easily distinguishable from the gravity variations which one wants to measure. Moreover, these fictitious forces are typically much bigger than the gravitational signal which one wants to measure; therefore, they can even determine a malfunction of the gravimeter (for example, saturation of the output signal).

[0014] To obviate such problem, the gravimeter is installed inside big and heavy active stabilizing platforms (for example with overall dimensions equal to about 1 m3, weight greater than 30Kg and requiring much electric power supply), not ideal to be used on a movable platform in which, generally, space, power supply and capability of transporting weight are poor resources. The suspensions of these stabilizing platforms compensate or mitigate accurately the external disturbances determined by the environmental conditions in which the gravimeter operates, such as oscillations, tilts, temperature variations. Despite that, the detected data have then to be filtered to obtain the desired information, and this requires to perform complex procedures. Moreover, often the filtering does not allow to remove completely the effect of the disturbances on the obtained acceleration measurement.

[0015] Summary of the invention

[0016] The technical problem placed and solved by the present invention then is to allow to perform accelerometer measurements in an improved way, by obviating the drawbacks mentioned above with reference to the known art.

[0017] Such problem is solved by a multi-sensor system for measuring acceleration, configured to detect in the same instant of time, apart from the piece of data on acceleration, even data related to the environmental conditions of the detection site - that is the conditions thereto the system itself is subjected, with particular reference indeed to the sensors to detect the acceleration included therein - by a plurality of sensors configured to this purpose. The detected acceleration measurement is subsequently corrected by applying an analytical model which punctually takes into consideration the influence of the above-mentioned environmental conditions on the measured acceleration value, as defined in the independent claim 1.

[0018] The system is configured to use Machine Learning (ML) techniques to analyse data related to a plurality of environmental conditions and corresponding contextual acceleration values, measured by the same system in the same detection site, with the purpose of constructing, in automated way, an analytical model of correlation between the values of the physical quantities measured by the multi-sensor system and the acceleration value which one wants to measure. Once devised, the model allows to calculate a posteriori the ‘correct’ acceleration value, as function of the contemporary measurements of the environmental conditions and of the acceleration performed by the multi-sensor system.

[0019] The ML algorithms which can be used in the context of this invention are the ones supervised for the regression. There are many algorithms of different type which can be used successfully, thereamong: Linear Regression, Support Vector Machine for the regression, Random Forest, Neural Networks, and so on.

[0020] By way of example, the linear regression method is described hereinafter. The linear regression is a fundamental machine learning algorithm, widely used to analyse and model the relationship between a dependent variable (in the subject case, the acceleration which one wants to measure) and one or more independent variables (the measurements obtained by the sensors of the multisensor system). The main purpose of the linear regression is to detect the best linear adjustment of the data, by determining the coefficients which minimize the difference between the above-mentioned values of the model and the real values observed in the training dataset.

[0021] The classical form of the linear regression is expressed by the equation: y=bo+bi*xi+b2*X2+...bn*xn, where y represents the dependent variable, bo is the intercept, bi, b2 bn are the coefficients of the independent variables xi, X2, ... xn. The main purposes of the linear regression are: estimating accurately these coefficients and evaluating the quality of the model adjustment to the data.

[0022] In other words, the analytical model of correlation between the values of the physical quantities associated to the environmental conditions of the detection site and their influence on the detected acceleration value is constructed by using ML techniques which process the training data (training data or "labelled" data), which comprise a plurality of sets of data of the environmental conditions and of the acceleration contextually measured by the same multi-sensor system, at different instants of time therefore the environmental conditions and the acceleration result to be different each time.

[0023] According to a preferred aspect, the training data have also to include data corresponding to a ‘correct’ acceleration value which is already known, or measured with an independent measurer in absence of disturbances, with reference to the same detection site of the above-mentioned training data. Once ended the training phase, for the subsequent measurements with the multisensor system, once detected the acceleration piece of data and the corresponding data inherent to the environmental conditions, a ML algorithm applied automatically by the system (previously devised by processing the training data) will provide as output the ‘correct’ acceleration value.

[0024] With reference to the training data which the system uses to apply the ML techniques, it is noted that the acquisition of such data has always been a problem in many ML applications. In fact, the fact of acquiring them would involve field measurements, in contexts in which the acceleration (for example that of gravity) is known with sufficient accuracy. Such operation could not be feasible (for example, in space applications) or at least expensive (for example, in air or submarine applications).

[0025] The present invention overcomes this problem by providing the possibility of generating the training data in laboratory, without having to operate on the field. In order to do this, the system, the independent claim 1 relates to, further comprises means adapted to determine variations in the acceleration which one wants to measure and / or selective variations in the detection environmental conditions, for example variations in temperature, variations in pressure, variations in asset (translations and rotations of the accelerometer, or a support thereto it is associated, according to one or more space directions). In other words, the system is configured to repeat a plurality of operating conditions in which one wants to perform the accelerometric measurements.

[0026] It will be then appreciated that the system solves even the training functionality for generating the training data, since it is configured to simulate different predetermined values of the above-mentioned environmental physical quantities so as to be able to measure, for each value, the corresponding acceleration, and be able to determine the relationship existing between the acceleration value and the one of the environmental conditions at which the measurement is performed. According to preferred variants, the training acceleration measurements are compared with a measurement of acceleration performed with an independent accelerometer, thereto such variations in the environmental conditions are not applied. Alternatively to the independent accelerometer, indirect measurement systems can be used, therefrom it is possible to calculate the acceleration which one wants to measure. For example, by measuring the position along the vertical and deriving the latter twice with respect to time, the vertical acceleration is obtained. Therefore, more generally, an acceleration independent measuring device can be indicated, which is meant to operate only during the phases for acquiring the training data whereas, during the operating phases, it is physically removed from the system or deactivated since, however, it would not be capable of providing useful measurements.

[0027] In order to detect the data of acceleration and of environmental conditions, the multi-sensor system of the invention comprises one or more primary sensors sensitive to acceleration (accelerometers) and one or more secondary (or supplementary) sensors sensitive to the environmental conditions under which such primary sensors operate. Moreover, the system comprises means for implementing predetermined variations in such environmental conditions and / or in the acceleration itself (for example according to one or more space directions), and a programmed control central unit for managing the means and the already mentioned sensors, as well as for processing the data detected by ML techniques, with the purpose of determining the model / algorithm binding the environmental conditions to the measured value of acceleration.

[0028] The configuration of the proposed system allows the use thereof in association with a passive stabilizing platform, having much smaller sizes and weight with respect to the active ones. This makes the system particularly suitable to couple with movable platforms, thereto it can be connected or hooked externally, or thereon it can be directly installed on board. In particular, the system of the invention can be installed on small sized vehicles (for example UAV, Unmanned Aerial Vehicle), as well as also on space satellites, ships, submarines. Consequently, the gravimetric surveys could be performed with fewer costs and with more automated modes than the current ones, by using less personnel.

[0029] For example, an UAV equipped with the multi-sensor system of the invention and other accessory sensors (magnetometers, cameras, and so on) could perform geophysical and environmental surveys in an almost wholly automated manner, by collecting precious data related to the territory.

[0030] The invention further relates to a method for measuring acceleration performed by the above-mentioned multi-sensor system, as defined in the independent claim 11.

[0031] Other advantages, features and use modes of the present invention will result evident from the following detailed description of some embodiments, shown by way of example and not for limitative purposes.

[0032] Brief description of figures

[0033] The figures of the enclosed drawings will be referred to, wherein:

[0034] ■ Figure 1 shows a front perspective view of a preferred embodiment of the system according to the present invention;

[0035] ■ Figure 2 shows a front perspective view of a preferred embodiment of an operative chamber of the system according to the present invention;

[0036] ■ Figure 3 shows a block diagram exemplifying a preferred embodiment of the system according to the present invention;

[0037] ■ Figure 4 shows a block diagram related to a preferred embodiment of the method for acquiring the training data according to the present invention; and

[0038] ■ Figure 5 shows a block diagram related to an additional preferred embodiment of the method for acquiring the training data according to the present invention. It will be understood that the above-mentioned Figures are introduced by pure way of example and not for limitative purposes.

[0039] Detailed description of preferred embodiments

[0040] Several embodiments and variants of the invention will be described hereinafter, and this with reference to the above-introduced Figures.

[0041] Analogous components are designated in the different Figures with the same numeral reference.

[0042] The different embodiments and variants described hereinafter are likely to be used in combination, where compatible.

[0043] The present invention relates to a solution for measuring acceleration by means of an accelerometer included in a multi-sensor system, configured to reduce or remove the disturbances deriving from the environmental conditions thereto the accelerometer is subjected. Such conditions can refer to one or more among the following ones: pressure, temperature, humidity, as well as accelerations, variations in height or orientation of the accelerometer itself, with respect to known or predetermined reference conditions.

[0044] It is noted that, within the present invention, the fact of measuring the gravity acceleration or the inertial acceleration (for example of a satellite) is wholly equivalent, as well as defined by Einstein’s equivalence principle.

[0045] The proposed solution is based upon the use of the Machine Learning to analyse data related to a plurality of different environmental conditions and of corresponding acceleration values, with reference to the same detection site and the same detection system of such environmental and acceleration data, with the purpose of constructing, in automated manner, an analytical model of correlation between the values in the physical quantities and those of the accelerations. The purpose is to determine, for the considered detection system, the influence of the environmental parameters on the detected acceleration value, and then to detect an algorithm which expresses such relationship. For the construction of the model, a set of training data is required, to generate which data the system, the invention relates to, is further configured. In other words, as it will be clearer hereinafter, the proposed system acts both as training system and as measuring system.

[0046] The system, in fact, is configured to simulate predetermined variations in one or more values of the above-mentioned physical quantities associated to the environmental parameters and / or acceleration (according to one or more space directions) so as to be able to measure, for each value, the corresponding acceleration (for example, gravity acceleration), and to reconstruct the relationship between the variations in such physical quantities and the corresponding measurements or variation in the measurand acceleration. Optionally, to this purpose it is possible to know the acceleration value to be taken as reference for the considered detection site, which can be an acceleration value measured herein independently from external disturbances, or better under known or predetermined environmental conditions (standard conditions).

[0047] With reference to Figure 1 , a preferred embodiment of a multi-sensor system according to the present invention is designated with 1 . As already described, the system 1 is configured for measuring acceleration, wherein the effect of the environmental conditions (or external environmental disturbances) on the measurement is reduced or removed thanks to the application of a Machine Learning algorithm.

[0048] The system 1 first of all includes one or more primary sensors 4 configured for measuring acceleration according to one or more space directions, and one or more secondary sensors 5, each one configured to measure one or more physical quantities or parameters associated with the environmental conditions under which the primary sensors 4 operate. Such primary sensors 4 can be housed inside an operative chamber 10, a preferably closed chamber, that is isolated with respect to the external environment, so that the environmental conditions under which the primary sensors 4 operate are those of the space inside such chamber 10.

[0049] With reference to Figure 2, three primary sensors are represented such as the accelerometers designated with the numeral references 41 , 42, 43, suitably arranged so as to measure the three orthogonal components of the acceleration vector, according to three orthogonal space directions of a reference system X, Y, Z. The accelerometer which, in use, results to be arranged along the Earth’s vertical can act as a gravimeter.

[0050] The parameters or the physical quantities corresponding to the environmental conditions thereto the primary sensors 4 are subjected, are object of measurement by the secondary sensors 5 and they can be one or more among the following ones, shown by way of not exhaustive example: temperature, pressure, humidity, rotations around one or more axes of a spatial reference system (for example a Cartesian reference system with three dimensions X, Y, Z), translations according to such axes, position with respect to such reference system.

[0051] For example, the secondary sensors 5 can include one or more of the following: one or more thermometers 51 , preferably arranged so as to measure the distribution of temperatures at a plurality of points of the multi-sensor system 1 and of the movable platform thereon it will be installed during the operating phase; a triaxial gyroscope, whose components 57, 58, 59 are suitably arranged so as to measure the angular speeds of the primary sensors 4 around one or more axes of the above-mentioned spatial reference system, or additionally / alternatively a sensor 53 for determining the asset (for example SLAM, LIDAR systems, star sensors, solar sensors); a satellite receiver for the positioning and navigation (for example a GPS receiver), alternatively to the sensor 53 for determining the position (for example SLAM, LIDAR systems, altimeters, depth gauges); a barometer 52 configured to measure the pressure variations.

[0052] For one or more, or for each one, of the primary 4 and / or secondary 5 sensors, the system 1 comprises one or more mechanical or dielectric means, defined hereinafter even as actuation means 8, suitable to implement a variation in the physical quantity or parameter detected by the corresponding sensor, to simulate the environmental disturbances which can occur while measuring acceleration. In particular, the actuation means 8 can be further configured to vary the acceleration which can be measured by the primary sensors 4.

[0053] According to preferred variants of the invention, each one of the primary sensors 4, secondary sensors 5 and actuation means 8 can be activated independently with respect to the other ones.

[0054] The actuation means 8 is configured in order to be able, upon need, to be removed from the system 1. In fact, their operation is useful exclusively during a preliminary phase for using the system 1 , wherein it is necessary to generate a sufficient number of training data for training the ML algorithm, such phase being necessarily preparatory to the real measurements to be then effected on the field, wherein the acceleration value measured by the primary sensors 4 will be automatically corrected by applying the previously generated ML algorithm.

[0055] According to preferred variants, the system 1 can include a training platform 11 , including one or more of such actuation means 8, or which still more preferably comprises all actuation means 8. According to such variants, the training platform 11 is configured to be able to be removed from the system 1 , once finished the phase for generating the training data.

[0056] The actuation means 8 can include one or more of the following, shown by way of not exhaustive example: linear actuators 82 configured to vary the height and tilting angle of the primary sensors 4, sliders 83 to move horizontally the main sensors 4, for example motorized slides, rotary devices 84 (for example motorized rotating platforms) configured to put in rotation the main sensors 4 around one or more axes of a spatial reference system (for example, a Cartesian reference system X, Y, Z), heating means 81 , which can be preferably oriented so as to be able to direct the heat selectively (for example infrared lamps), and cooling means. For example, in order to vary the acceleration along the vertical, the linear actuators 82 can be used.

[0057] In order to improve the effectiveness of the training phase, redundant actuation means 8 can be provided, which can be actuated independently therebetween, to simulate predetermined variations in one or more values of the same environmental physical quantities and / or of the same acceleration according to one or more space directions. For example, in order to induce a temperature variation, both lamps and heating mats can be provided in the system.

[0058] According to additional variants, and as already partially anticipated, the system 1 can include an operative chamber 10 configured to house all primary sensors 4, and preferably one or more of the secondary sensors 5. According to preferred variants, one or more of the secondary sensors 5 are arranged externally with respect to the operative chamber 10. More in particular, even one or more of the actuation means 8 are arranged outside the operative chamber 10.

[0059] According to preferred embodiments, the training platform 11 comprises one or more of the actuation means 8 adapted to simulate a controlled variation in the acceleration which one wants to measure with the primary sensors 4, as well as a controlled variation in the environmental conditions under which such primary sensors 4 operate. Such variations obviously are to be considered with respect to known reference conditions or parameters, for example measured by the sensors 5 before starting the phase for generating the training data.

[0060] According to preferred embodiments of the invention, the training platform 11 is configured to be mechanically connected to, or better to house, the operative chamber 10. Such coupling between the training platform 11 and the chamber 10 is reversible: the training platform 11 and the operative chamber 10 are configured to be separated in case of removal of the platform 11 from the system 1.

[0061] Within the present invention, it is specified that the term ‘platform’ does not mean exclusively a plate-like support, or more generally with flat development; the platform can be implemented by a general mechanical supporting element configured to house, or to connect with, at least one of the actuation means 8 and / or the operative chamber 10.

[0062] According to preferred variants of the invention, once determined the ML algorithm, the operative chamber 10 can be released from the training platform 11 and directly constrained to the frame of the movable platform (for example the aircraft, the submarine or the space satellite) thereby one wants to perform subsequent ‘correct’ accelerometer measurements.

[0063] According to preferred variants of the invention, the operative chamber 10 comprises hooking means configured to implement a (temporary) mechanical connection to a vehicle for land, submarine, air or space transport, or more generally to a movable platform.

[0064] In particular, the chamber 10 can be temporarily constrained to the frame of the movable platform through a towing system, so as to be towed or dragged by the moving platform while the accelerometer measurements are performed. A connection of this type can be implemented by a hook provided at the operative chamber 10 (or the movable platform), and a respective hooking seat carried on the movable platform (or on the operative chamber 10). In this case, the operative chamber can be further configured to protect from the external environment (for example from rain or water for submarine applications) the primary and secondary sensors 4, 5 - which will be all carried inside thereof - being suitably sealed. Obviously, the chamber 10 will be implemented so as to be able to resist, without being damaged, under even extreme environmental conditions characterizing the detection site (for example, high submarine pressures).

[0065] Still, the training platform 11 can include an independent measurer of the acceleration which one wants to measure. Such independent measurer could be an independent accelerometer on which the disturbances generated by the actuation means 8 act, or an indirect measurement system, therefrom the acceleration is to be obtained, exploiting for example, positioning sensors.

[0066] The system 1 further comprises a control central unit 6, configured to store the data detected by the primary sensors 4 and by the secondary sensors 5 in predetermined instants of time, in correspondence of which the data detected by the primary 4 and secondary 5 sensors may have undergone a variation with respect to a known reference condition due to the activation of one or more actuation means 8. Therefore, the central unit 6 is configured to store data related to specific environmental conditions in association to the value of the acceleration measured in presence of such conditions. The central unit 6 can be physically remote with respect to the operative chamber 10 and / or the training platform 11 , if present.

[0067] The central unit 6 can be further configured to acquire the acceleration value detected at the installation site of the multi-sensor system 1 by an independent measurer, that is a sensor placed under known or standardized environmental conditions, for example included in the training platform 11 , according to what already described. Such value is meant to be the ‘correct’ acceleration value to be used as reference value, since not affected by external disturbances.

[0068] The central unit 6 is configured, or better programmed, to process all sets of data data detected by the primary sensors 4, secondary sensors 5 and the possible reference acceleration value - already known or detected by the independent measurer - by means of Machine Learning techniques, with the purpose of determining the mathematic model which binds therebetween the values detected by the primary sensors 4 and the values detected by the secondary sensors 5, and in case the reference acceleration value. Such sets of data are determined at different instants of time, thereto different modes for activating the actuation means 8 correspond.

[0069] Once determined the Machine Learning algorithm which allows to obtain the ‘correct’ acceleration value starting from the data detected by the primary sensors 4 and by the secondary sensors 5 (and in case the reference acceleration value) and validated such algorithm, for each subsequent measurement performed by the system 1 , the correct acceleration value is obtained by applying the algorithm to the value obtained by the primary sensors 4, considering as input such value and the data detected by the secondary sensors 5. Therefore, once determined the algorithm, the multi-sensor system 1 allows to provide as output the ‘correct’ acceleration value with respect to the particular measurement environmental conditions.

[0070] Figure 3, which relates to a preferred embodiment of the multi-sensor system comprising the training platform, on this regard shows a block diagram which highlights that the training platform is used only for the training phase, preparatory to subsequent measurements.

[0071] By referring again to Figure 1 , the central unit 6 can further include, even wireless, communication means, for transmitting / receiving data and control signals. Such communication can take place with the primary and secondary sensors 4, 5, the actuation means 8, the possible independent measurer, and / or with external electronic devices (pc, smartphone) to allow to control and monitor the system 1 even remotely.

[0072] The method for measuring the acceleration in a target site according to the present invention, implementable by the already described multi-sensor system 1 , then provides a preliminary training phase, during which the data for training are generated and at the end thereof the algorithm for the correlation between the measurements of the primary and secondary sensors at the same site is determined, in function of the different environmental conditions caused by activating the actuation means 8. Once established the algorithm, the multisensor system could be used for subsequent measurements, in different sites, therefore the acceleration value detected by the primary sensors 4 will be automatically corrected to take into considerations the corresponding environmental conditions detected herein by the sensors 5, as well as possible external disturbances at the expense of the acceleration itself, detected by the same sensors 4.

[0073] According to an additional preferred embodiment, the steps provided by the method are more specifically listed hereinafter:

[0074] - (optional) acquiring a reference gravity acceleration value or, at a first instant of time, measuring a reference gravity acceleration value at a determined site, under a reference condition wherein one or more physical quantities corresponding to the environmental conditions of such site, as already described, are known;

[0075] - at second instants of time, activating the actuation means 8 in order to cause in the site a variation in the physical quantities corresponding to the environmental conditions and / or in acceleration, and contextually measuring the respective acceleration punctual values and the parameters related to the environmental conditions in the detection site (that is, the environmental conditions in which the primary sensors 4 operate) for each one of the second instants of time by the primary 4 e secondary 5 sensors, and optionally detecting corresponding reference acceleration values by independent measurers;

[0076] - storing such sets of punctual values, wherein each set is meant referred to a second instant of time;

[0077] - processing such sets of punctual values by applying Machine Learning techniques, to determine a correlation algorithm between the acceleration measurements and those of the contextually measured environmental parameters;

[0078] - once this algorithm has been determined, at a third instant of time, acquiring a punctual measurement of acceleration in the target site by the primary sensors 4, with the actuation means 8 deactivated and / or removed from the system 1 , and contextually acquiring by secondary sensors 5 the values corresponding to said one or more physical quantities associated to the environmental conditions at the detection site (that is, at the primary sensors 4);

[0079] - applying said algorithm so as to calculate the ‘correct’ acceleration punctual value at the detection site, which is function of the values corresponding to the measurements performed by the primary 4 and secondary 5 sensors nearby; and

[0080] - providing as output the ‘correct’ acceleration value.

[0081] Figure 4 shows the way in which the contributions of the primary 4 and secondary 5 sensors and of the possible independent measurer can be acquired and processed in detail. As far as the independent measurer is concerned, the value of the measurand has to be known with great accuracy since it will represent the desired output of the training data. Therefore, the disturbances induced by activating the actuation means 8 have not to act on the independent measurer. For example, as primary sensor and independent measurer two gravimeters can be used, placed next to each other, so that the force of gravity measured thereby is the same. However, one should work in absence of disturbances, the other one will be placed for example on the training platform 11 , and however under the action of the disturbances induced by activating the actuation means 8.

[0082] It is repeated that the independent measurer (represented in Figure 4 with dotted line) is necessary only during the acquisition of the training data and not during the subsequent operative phases for performing the measurements, when the correlation algorithm has already been calculated. At last, it is to be highlighted that the ML algorithm has as inputs all measurements, both of the primary sensors and of the secondary sensors; on the contrary, the algorithm output is the desired signal, which for the training data is provided by the independent measurer.

[0083] Figure 5 shows schematically an additional preferred embodiment of a method for acquiring the training data, wherein the contribution of the independent measurer is not present and wherein a variation in the measurand acceleration by the actuation means is not induced. According to such variant, the training data, for example, can be acquired when the measurand (the acceleration which one wants to measure) is constant or, however, its variations are negligible. In such circumstance, the output of the accelerometer (that is of the primary sensors) is only due to the external disturbances, that is to the variations in the environmental parameters due to the activation of respective actuation means. In other words, it represents exactly what in operating phase is considered measurement error. As shown in the scheme, in this mode the algorithm ‘learns’ to calculate the error by using only the measurements of the secondary sensors; in fact, the accelerometer output is not one of the algorithm inputs, but the desired output. During the subsequent operating phase on the field, once determined the ML algorithm, also the variations in the measurand will be present, and the accelerometer output will consist of a term due to the error and one to the measurand. The first term will be calculated by the ML algorithm and it could be subtracted from the accelerometer output to obtain the ‘correct’ acceleration measurement.

[0084] It is observed that this variant is not exhaustive like the preceding one, illustrated in Figure 4, and it does not generate training data capable of compensating the interaction non-linear effects between the measurand and the disturbances. However, in some applications, when the non-linear effects are negligible, it could be sufficient. Moreover, it has the great advantage of simplifying considerably the setup for acquiring the training data. In fact, it not necessary to change the measurand and it can be assumed that it is always null, without the need for an independent measurer.

[0085] Substantially, a method for performing acceleration measurements in a target site using a multi-sensor system 1 according to the present invention, comprises the following steps: at first instants of time: to cause in the target site disturbances which generate a variation in the physical quantities corresponding to the environmental conditions and / or in the acceleration with respect to an initial or reference condition by activating the actuation means 8, whereas one contextually measures: the punctual values of such physical quantities and of the acceleration subjected to such disturbances, and reference values of the acceleration not subjected to such disturbances, for each one of the first instants of time; storing the so-measured acceleration punctual values in association with the corresponding variations in the above-mentioned physical quantities, and the corresponding reference values of acceleration; processing the punctual values of acceleration, of the corresponding variations in physical quantities and the corresponding reference values of acceleration by means of Machine Learning techniques, to determine a correlation algorithm between the punctual values of acceleration, the corresponding variations in the physical quantities and the corresponding reference values of acceleration, and once this algorithm is determined: at a second instant of time, acquiring a punctual value of the acceleration in the target site, with the actuation means 8 deactivated, and contextually acquiring one or more values corresponding to the physical quantities at the target site; applying the algorithm, in such a way as to correct the acceleration punctual value measured according to the values corresponding to the physical quantities measured in the second instant of time at the target site; and providing the so corrected acceleration value as output.

[0086] The present invention has been sofar described with reference to preferred embodiments. It is to be meant that other embodiments belonging to the same inventive core may exist, as defined by the protective scope of the herebelow reported claims.

Claims

CLAIMS1. A multi-sensor system (1) for measuring acceleration, including: one or more primary sensors (4) for measuring acceleration according to one or more axes of a spatial reference system (X, Y, Z); one or more secondary sensors (5), each one configured to measure one or more physical quantities or parameters associated with the environmental conditions wherein said primary sensors (4) operate, comprising at least one of:- temperature;- pressure;- rotation around one or more axes of said spatial reference system (X, Y, Z);- position with respect to said spatial reference system (X, Y, Z); wherein said multi-sensor system (1) comprises: one or more removable actuation means (8), configured to cause a variation in the acceleration measured by said primary sensors (4) and / or, for each one of said one or more secondary sensors (5), configured to be activated such as to cause a variation in the corresponding physical quantity or parameter measured by the respective secondary sensor (5); a central unit (6) configured to store the data measured by said one or more primary sensors (4) and said one or more secondary sensors (5), in first predetermined instants of time in correspondence of which one or more actuation means (8) are activated, wherein said central unit (6) is further configured: to process said data detected by said primary sensors (4) and said one or more secondary sensors (5) by means of Machine Learning techniques, to determine a correlation algorithm between: the data detected at said first predetermined instants of time by said one or more primary sensors (4) and secondary sensors (5), wherein, once this algorithm has been determined, to carry out subsequentacceleration measurements at a second predetermined instant of time, the central unit (6) is further configured: to acquire the values measured by said one or more primary sensors (4) at the second predetermined instant of time, wherein said actuation means (8) is deactivated; to acquire the values corresponding to said one or more physical quantities or parameters associated with external disturbances, detected by said one or more secondary sensors (5) at said second predetermined instant of time; to apply said algorithm, in such a way as to correct the acceleration value detected by said one or more primary sensors (4) according to the values detected by said one or more secondary sensors (5); and to provide the so corrected acceleration value as output.

2. The multi-sensor system (1) according to claim 1 , which includes an operative chamber (10) comprising said one or more primary sensors (4).

3. The multi-sensor system (1) according to claim 2, wherein said operative chamber (10) comprises one or more of said secondary sensors (5).

4. The multi-sensor system (1 ) according to one of the preceding claims, comprising an independent device configured to measure a reference acceleration value, which is not modified by the activation of said actuation means (8), wherein said central unit (6) is configured to store the data detected by said one or more primary sensors (4), said one or more secondary sensors (5) and said independent device, in first predetermined instants of time at which one or more actuation means (8) are activated, wherein said central unit (6) is further configured: to process said data detected by said one or more primary sensors (4), said one or more secondary sensors (5) and said independent device, by means of Machine Learning techniques, to determine a correlation algorithm between: the data detected at said first predetermined instants of time by said one or more primary sensors (4), said one or more secondary sensors (5) and the accelerationvalue detected by said independent device, wherein, once this algorithm has been determined, to carry out subsequent acceleration measurements, the central unit (6) is further configured: to acquire the values measured by said one or more primary sensors (4) at the second predetermined instant of time, wherein said actuation means (8) is deactivated; to acquire the values corresponding to said one or more physical quantities or parameters associated with external disturbances, detected by said one or more secondary sensors (5) at said second predetermined instant of time; to apply said algorithm in such a way as to correct the acceleration value detected by said one or more primary sensors (4) according to the values of said one or more secondary sensors (5); and to provide the so corrected acceleration value as output.

5. The multi-sensor system (1 ) according to one of the preceding claims, wherein said one or more secondary sensors (5) comprise one or more of the following: accelerometers (41 , 42, 43), thermometers (51 ); triaxial gyroscope (57, 58, 59); position sensor (53); barometer (52).

6. The multi-sensor system (1 ) according to one of the preceding claims, wherein said actuation means (8) comprises one or more of the following: heating means (81), cooling means, linear actuators (82) configured to vary the elevation and / or angle of inclination of said one or more primary sensors (4), sliders (83) configured to move said one or more primary sensors (4) horizontally, rotary devices (84) configured to rotate said one or more primary sensors (4).

7. The multi-sensor system (1 ) according to one of the preceding claims, wherein said actuation means (8) is included in a training platform (11), configured to be removed from said multi-sensor system (1 ).

8. The multi-sensor system (1) according to claim 7 when it depends also upon claim 2, wherein said platform (11) is configured to be mechanically and reversibly connected to said operative chamber (10).

9. The multi-sensor system (1 ) according to one of claims 2 to 8, wherein said operative chamber (10) comprises hooking means configured to implement a mechanical connection with a land, submarine, air or space transport vehicle.

10. A vehicle for land, submarine, air or space transport, comprising a multisensor system (1 ) according to one of the preceding claims.

11. A method for performing acceleration measurements in a target site using a multi-sensor system (1) according to one of the preceding claims, comprising the following steps: at first instants of time, causing in the target site disturbances which generate a variation in the acceleration and / or in the physical quantities corresponding to the environmental conditions with respect to an initial or reference condition by activating actuation means (8), and contextually measuring in the target site: the punctual values of such physical quantities and of the acceleration, and reference values of the acceleration not subject to such disturbances, for each one of the first instants of time; storing said punctual acceleration values in association with the corresponding variations in such physical quantities and the corresponding reference values of acceleration; processing said punctual values of acceleration, of the corresponding variations in such physical quantities and the corresponding reference values of acceleration by means of Machine Learning techniques, to determine a correlation algorithm between them, and once this algorithm has been determined: at a second instant of time, acquiring a punctual value of the acceleration in the target site, with said actuation means (8) deactivated, and contextually acquiring one or more values corresponding to said physical quantities at the target site; applying said algorithm in such a way as to correct said acceleration punctual value measured as a function of said values corresponding to the physical quantities detected in the second instant of time at the target site; andproviding the so corrected acceleration value as output.

Images