Classifications

• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
  • G01M7/00—Vibration-testing of structures; Shock-testing of structures
• • B—PERFORMING OPERATIONS; TRANSPORTING
  • B64—AIRCRAFT; AVIATION; COSMONAUTICS
  • B64F—GROUND OR AIRCRAFT-CARRIER-DECK INSTALLATIONS SPECIALLY ADAPTED FOR USE IN CONNECTION WITH AIRCRAFT; DESIGNING, MANUFACTURING, ASSEMBLING, CLEANING, MAINTAINING OR REPAIRING AIRCRAFT, NOT OTHERWISE PROVIDED FOR; HANDLING, TRANSPORTING, TESTING OR INSPECTING AIRCRAFT COMPONENTS, NOT OTHERWISE PROVIDED FOR
  • B64F5/00—Designing, manufacturing, assembling, cleaning, maintaining or repairing aircraft, not otherwise provided for; Handling, transporting, testing or inspecting aircraft components, not otherwise provided for
  • B64F5/60—Testing or inspecting aircraft components or systems
• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
  • G01M5/00—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings
  • G01M5/0066—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings by exciting or detecting vibration or acceleration

Definitions

• the present invention relates to a method and a device for frequency analysis of data. It applies, in particular, to the analysis of aircraft flight range opening test data.
• the present invention applies in particular, in the aeronautical field, to the flight controls, for example the analysis and the flight control of the vibratory modes of the structure, in the automotive field, to the studies and checks of the vibrations of the vehicles, in electrodynamics (control of electricity generating machinery), particularly in the nuclear field, vibration monitoring of the reactor core, mechanics (study and control of moving parts), seismic (study of signals used in oil prospecting) and zoology (study of sounds emitted by animals).
• the object of the present invention is to estimate, during the test (during flight, in the case of an aircraft) the characteristics of the vehicle and, in particular, the resonance frequencies and the spectral characteristics. In other words, the very large amount of information that comes from the sensors installed on the vehicle, it is to extract the relevant signatures very quickly, or even in real time.
• the signals that one wishes to analyze are composed of a noise plus one or more sinusoidal signals whose frequencies and amplitudes are likely to vary over time. It is a question of estimating these frequencies and these amplitudes in real time.
• Modal parameter identification methods are used to extract, in near real-time, the frequency and damping values and to study their evolution in the flight domain.
• the analysis of time data from floating tests is complex: the data is tainted with noise and requires formatting by signal processing (including filtering and downsampling).
• signal processing including filtering and downsampling.
• Several sensors are - WWWQJJ
• WO03005229 discloses a system for frequency analysis of signals from a sensor. However the resolution of this analysis is limited.
• the present invention aims to remedy these disadvantages.
• the present invention aims, in a first aspect, a frequency analysis method of a system, characterized in that it comprises:
• a step of signal inputs from a first sensor a step of inputting signals from at least one second sensor, each second sensor being positioned near the first sensor so that the signals from each second sensor; sensor are strongly correlated with the signals from the first sensor,
• a model considers the signal coming from a sensor as the output of a filter excited by a white noise.
• the structural properties include, for example, the spectral properties, frequencies, amplitudes, phases at the origin, damping, modes.
• the representative model of the structural modes is considered linear. Thanks to the implementation of the present invention, a real-time processing is carried out by performing online monitoring of the frequency / damping pairs.
• the present invention makes it possible to ascertain in real time that the behavior of the system, for example of the aircraft, is satisfactory since the structural properties of the system are available in real time. This improves the methods of analysis used while meeting the growing constraints of time saving and therefore cost reduction.
• the signals coming from the sensors are considered as polynomials. Thanks to these provisions, the representation of the signals is compact, since the number of polynomial coefficients is much smaller than the number of signal samples used.
• the estimation step comprises:
• each step of inputting signals from sensors comprises a step of real-time reduction of the noise level of the signals from sensors preceding the adaptive modeling step.
• the mode estimation step includes a step of extracting parameters from the model as a function of the result of the adaptive modeling step.
• the step of extracting parameters from the model comprises a step of inverting a matrix of polynomials of order N and of dimension equal to the number of sensors.
• the mode estimation step is adapted to provide the parameters of each of the models constituting a set of redundant information that makes it possible to reduce the variance of the estimated modes.
• the adaptive modeling step performs an ARMA type of modeling ("autoregressive to adjusted average"). According to particular characteristics, said type-modeling
• the estimation step comprises a step of inverting a polynomial matrix which is a symmetric inter-spectral matrix representing, on its main diagonal, the power spectral density of each of the sensors and, in other words, the inter-spectra.
• the method which is the subject of the present invention comprises a step of classifying the modes resulting from the step estimating modes by implementing one of two constraints:
• the present invention relates to a computer program loadable in a computer system, said program containing instructions for implementing the method object of the present invention, as briefly described above. Since the advantages, aims and particular characteristics of this program are similar to those of the process that is the subject of the present invention, as briefly described above, they are not recalled here.
• FIG. 1 represents, schematically, an aircraft comprising a device capable of implementing the method that is the subject of the present invention
• FIG. 2 represents signals from two sensors of the device illustrated in FIG. 1;
• FIG. 3 represents, in the form of a logic diagram, steps implemented in a first embodiment of the method that is the subject of the present invention,
• FIG. 4 represents, in the form of a logic diagram, the steps implemented in a second embodiment of the method that is the subject of the present invention
• FIG. 5 represents a filter arrangement implemented during one of the steps illustrated in FIG. 4,
• FIG. 6 represents, schematically, the successive functions implemented in one embodiment of a noise reduction system
• FIG. 7 represents, schematically, at each moment, the samples coming from the sensors constituting the inputs of the algorithm of the embodiment illustrated in FIG. 4;
• FIG. 8 schematically represents recursions implemented in the second embodiment illustrated in FIG. 4;
• FIG. 9 represents, schematically, an evolution of classes of a non-supervised classification method of "dynamic clouds" type and
• FIG. 10 gives an illustration of a validation window implemented in the second embodiment illustrated in FIG. 4.
• FIG. 1 shows an aircraft 105 provided with two sensors close to each other 110 and 115 at the front of the wing 120 and two sensors close together 125 and 130, at the rear of the wing 120.
• near here refers to sensors that receive signals strongly correlated with each other.
• the close sensors receive substantially the same vibrations, shifted in time and damped differently but according to a substantially linear transfer function.
• the sensors in question are, for example, accelerometers.
• FIG. 2 shows that the signal 205 coming from a first sensor of a pair of sensors comprises noise 210, and two peaks 215 and 220 and that the signal 255 coming from the second sensor of the same pair of sensors comprises 260 and two peaks 265 and 270.
• the peak 265 corresponds to the peak 215 damped and shifted in time.
• the peak 220 corresponds to the peak 270 damped and shifted in time.
• the present invention allows an analysis of the structural properties of the aircraft.
• These structural properties include, for example, the spectral properties, frequencies, amplitudes, phases at the origin, damping, modes.
• FIG. 3 shows, in a first embodiment, the method of frequency analysis that is the subject of the present invention comprises, first of all, a step 305 of positioning, on the structure of a mechanical system subject to vibrations, groups of a plurality of sensors. In each sensor group, at least one so-called “second” sensor is positioned near a so-called “first” sensor.
• a step 310 is made of signal inputs from a first sensor of a said group of sensors and a signal input stage 315 from at least a second sensor of the same group of sensors.
• Each signal input step from a sensor comprises a step of reducing the noise level of the signals from the sensor.
• This noise reduction can be carried out sensor by sensor, in known manner or on a vector having, for each of its coordinates, a signal from a sensor.
• this denoising function is provided by a decomposition on a wavelet basis (Algorithm of Stéphane Mallat).
• the signals from the sensors are sampled at a much higher frequency, for example 256 Hz in FIG. the use of wavelets allows a simple and fast treatment.
• the transfer functions are extracted by processing the signals from the first sensor and each second sensor.
• the representative model of the structural modes is considered to be linear and the frequency / damping pairs are tracked in line without taking into account the input, that is to say the injected excitation.
• the signals coming from the sensors are considered as polynomials and the information is extracted from the signals coming from the sensors, after taking into account the phase and value changes of the said information between the signals coming from the different sensors. different groups of sensors.
• step 320 a step 325 of solving a linear recurring equation with coefficients varying slowly over time is performed in order to be able to estimate the models over a sufficiently stable time interval.
• Step 325 comprises: a step 330 of ARMA recursive adaptive modeling over time (at each instant), the order (for each order considered) and the sensor space (for each sensor) and
• Step 330 carries out parametric type modeling of ARMA type (self-adjusting to adjusted average).
• the samples from the sensors are assembled into a vector whose number of components is the number of sensors considered (see Figure 7). For example, if there are four sensors, the vectors considered are of dimension four. More generally, in the following description, is called "p" the number of sensors of a group of nearby sensors.
• Step 330 performs so-called "recursive over time” modeling because it uses the latest estimates obtained to update its parameters.
• the present invention implements relationships between two consecutive instants because they are considered to be correlated and coherent. In embodiments, it is the two previous instants that serve for time recursion.
• step 330 comprises a step 335 comprising a step of processing a symmetrical inter-spectral matrix representing, on its main diagonal, the power spectral density of each of the sensors and, in extra diagonal terms, inter-spectra.
• Step 350 provides parameters of each of the models constituting a set of redundant information that reduces the variance of the estimated modes.
• Step 350 includes a step 355 of extracting model parameters based on the result of the adaptive modeling step.
• Step 355 includes a step 360 of inverting a matrix of polynomials of order N and of dimension equal to the number of sensors.
• Step 360 includes step 365 of Cholesky decomposition.
• the method that is the subject of the present invention follows a real-time procedure, described with reference to FIG. 4, dedicated to the analysis of flight domain opening test data.
• This embodiment makes it possible to process each information before the appearance of the next one, without taking into account the excitation injected into the structure.
• the characteristics of the method fully meet the safety and cost reduction constraints stated in the preamble and make it possible to improve the procedure for opening the flight envelope by providing a more efficient modal analysis.
• the signals to be analyzed are acceleration measurements made on the primary structure of the aircraft.
• the first step 405 of the analysis method consists in performing a "denoising" of the p signals from the p sensors of the same group of nearby sensors, for example using the pyramidal algorithm proposed by S. Mallat, in using orthonormal wavelet bases.
• This algorithm has its origin in the work of Burt and Adelson dating back to 1983, which focused on vision and compression of images.
• This algorithm of great simplicity implementation, has a computing load proportional to the number of samples to be processed.
• the non-linear character of the treatment complicates the implementation, especially since the filter bank is not causal.
• the filters are arranged in the manner illustrated in FIG. 5, using wavelets.
• Mallat's algorithm is absolutely real-time, it is generalized to the simultaneous processing of p samples at every moment.
• the principle adopted is based on the oversampling of the signal in order to allow "denoising” by analysis and synthesis.
• a step 406 of decomposing the signals in sub-bands is first effected.
• the sequence formed by the sampling of the continuous signal is considered, as a first step, as the approximation of this signal to a certain scale related to the discretization (the sampling corresponds in fact at the finest scale).
• the approximation coefficients (/ ⁇ jk j and details ⁇ / ⁇ J k j on the scale y are calculated from those obtained on the scaley -i by a simple filtering operation by the filters ô (f ) and // (/) followed by decimation. Then, a step 407 is performed, a thresholding of the coefficients of the decomposition, during which only the coefficients of the first subband are kept.
• the reconstruction is a dual operation of the previous one. It is obtained by digital filtering preceded by an interpolation on the coefficients of approximation and details resulting from the decomposition.
• the general structure of the "denoiser” is thus that illustrated in FIG. 6. It is noted that the maximum frequency considered is half of the sampling frequency.
• the desired signal is in the lower frequencies. After decomposing the signal in frequency bands ranging from 0 to f / 16, from f / 16 to f / 8, from f / 8 to f / 4 and from f / 4 to f / 2, thresholding is carried out by setting zero coefficients of the wavelets. Then, a synthesis is performed to provide a denoised signal comprising the desired signal.
• an ARM-type adaptive modeling (“AutoRegressive Moving Average”) is performed.
• This step of modeling the signal is of parametric type, which makes it possible to obtain a spectral analysis of the signal studied, and recursively in time, in order and on the space of the sensors.
• Step 410 comprises a step 411 for determining a forward linear prediction vector and a backward linear prediction vector.
• each determination of a prediction vector consists in expressing x ⁇ in a linear combination of the last N vectors of samples, with the representation illustrated in FIG. 7 which represents the forward linear prediction in vector form over the space of the p sensors.
• the vector x n has as components the p current samples of the p sensors.
• ⁇ esA k are matrices of dimension p which correspond to the number of sensors, N being the order of modeling.
• e n is the output of a RIF filter excited by the vector sequence of samples X n .
• the linearity property makes it possible to invert the process: X n then appears as the output of a filter excited by e n .
• This filter is obtained by inverting the polynomial matrix ⁇ (z), it is stable and of type RII (acronym of "infinite impulse response").
• the properties are similar to those obtained for forward linear prediction.
• the vector N ⁇ _ annum calculated directly through an "ESA" algorithm is the product of a matrix by a vector.
• the matrix is the weighted sum up to the moment n + 1 of the dyadic product of the vector ⁇ f- N a, n and the vector is this
• This matrix is symmetrical, on its main diagonal appears the power spectral density of each of the sensors. Extra diagonal terms are inter-spectra.
• the extraction of the parameters of the models consists in making at each moment and for each order the inversion of the matrix of polynomials of order N and of dimension p: ⁇ (z).
• the polynomials F represent the numerators of the different transfer functions, the denominators (in fact the only denominator, according to the superposition theorem) are the eigenvalues.
• the numerator coefficients are the square roots of the N + 1 elements of the N + 1 th column of the inter-spectral matrix (without the sign).
• the processing of adaptive modeling can be represented as in FIG. 8.
• FIG. 8 In this diagram of the processing of multisensor modeling, we see the nested three recursions 805, 810 and 815.
• a systematic search is performed for each data item of the best class; calculating the distance of the data to the centroids and assigning the element to the class whose center is closest to it (for example by using a Euclidean distance or Kullback-Leibler) and
• a construction of the trajectories of the modes is carried out.
• the problem is to follow in real time the trajectories of a set of targets corresponding to the frequencies of modes whose number changes over time.
• the structure of the algorithm is built around a Kalman filter per track followed.
• a set of "measures” is provided by the models, one seeks to correlate them to the existing tracks.
• the objective here is to select, from the received measurements, those that are likely to come from the target from which the measurement is predicted.
• a principle often used is to define a window, commonly called “gating" around the prediction made.
• the general processing architecture is based on the following principles:
• the measurement-target combination consists of comparing the measurements with those predicted from the known trajectories. This treatment must not only maintain the existing tracks but also initialize new ones, and possibly eliminate those that correspond to targets having left the observation space. It is the quality of these functions that the performance of the track follower depends. - the filtering with the update of the state, the gain of Kalman as well as matrices of covariance.
• the Kalman filter allows, in this procedure, to pursue several targets where the prediction plays a fundamental role. It provides, for each target, a filtered estimation of the state in the sense of the minimum variance, predicts the state and allows the calculation of the "gating".
• the solution consists of the set of two systems of prediction and filtering equations, namely:
• the validation window allows, for each target, to select the measures likely to belong to the target.
• the principle is to define an area, a volume in the space of the observations, around the predicted measure.
• the size of this zone is defined thanks to the statistical properties of the predicted measurement (Gaussian in this case).
• the "size" of this volume must be carefully chosen. From him, indeed, depends the sorting of measurements and the probability that the measurement coming from the target is inside the surface delimiting this volume.
• Figure 10 gives an illustration of the validation window.
• the measurement-target association technique is the central part of the target tracking procedure. Many techniques exist among which some do not manage the appearance and the disappearance of the tracks. It is therefore necessary to provide an additional mechanism to achieve this management. A simple way is to adopt the following rules:
• a track is confirmed (detected) if at least Nd consecutive measurements have been associated with it.

Landscapes

• Engineering & Computer Science (AREA)
• Aviation & Aerospace Engineering (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Manufacturing & Machinery (AREA)
• Transportation (AREA)
• Feedback Control In General (AREA)
• Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
• Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
• Complex Calculations (AREA)
• Radar Systems Or Details Thereof (AREA)
• Navigation (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=40673306

Family Applications (1)

Country Status (9)

Families Citing this family (16)

Family Cites Families (12)

• 2008
  • 2008-07-07 FR FR0854622A patent/FR2933513B1/fr active Active
• 2009
  • 2009-07-06 CA CA2730039A patent/CA2730039C/en active Active
  • 2009-07-06 CN CN200980126699.3A patent/CN102105771B/zh active Active
  • 2009-07-06 EP EP09794007A patent/EP2297564A1/fr not_active Withdrawn
  • 2009-07-06 US US13/002,911 patent/US8725468B2/en active Active
  • 2009-07-06 WO PCT/FR2009/000833 patent/WO2010004133A1/fr active Application Filing
  • 2009-07-06 RU RU2011104085/28A patent/RU2503938C2/ru not_active IP Right Cessation
  • 2009-07-06 JP JP2011517192A patent/JP5480255B2/ja not_active Expired - Fee Related
  • 2009-07-06 BR BRPI0910510A patent/BRPI0910510A2/pt not_active Application Discontinuation

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events