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
  • G01M13/00—Testing of machine parts
  • G01M13/04—Bearings
  • G01M13/045—Acoustic or vibration analysis
• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
  • G01M15/00—Testing of engines
  • G01M15/14—Testing gas-turbine engines or jet-propulsion engines
• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
  • G01M13/00—Testing of machine parts
  • G01M13/02—Gearings; Transmission mechanisms
  • G01M13/028—Acoustic or vibration analysis
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N3/00—Computing arrangements based on biological models
  • G06N3/02—Neural networks
  • G06N3/04—Architecture, e.g. interconnection topology
  • G06N3/0464—Convolutional networks [CNN, ConvNet]
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N3/00—Computing arrangements based on biological models
  • G06N3/02—Neural networks
  • G06N3/08—Learning methods
  • G06N3/084—Backpropagation, e.g. using gradient descent
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N3/00—Computing arrangements based on biological models
  • G06N3/02—Neural networks
  • G06N3/08—Learning methods
  • G06N3/09—Supervised learning

Definitions

• the technical field of the invention is that of the diagnosis of parts and more particularly that of the automatic diagnosis of parts.
• the present invention relates to an automatic diagnostic method of a part of a rotating machine from a non-stationary time signal generated by the rotating machine.
• the present invention also relates to a computer, a computer program product and a recording medium.
• the diagnosis of a system is conventionally carried out by analyzing physical signals generated by the system, which can be measured by sensors, for example electric, magnetic, thermal, acoustic or even vibration.
• vibration signals are commonly used to determine the operating state of components of rotating machines, for example their bearings: during the production or maintenance phases, high frequency vibration signals are thus acquired when the rotating machine is in operation, in order to detect weak signals characteristic of bearing damage, called signatures, and thus prevent a failure of the rotating machine.
• the acquired vibration signals are analyzed by calculating statistical indicators from the vibration signals in the domain of the orders of a rotation speed.
• the calculated statistical indicators are then compared with statistical indicators representative of a population of rotating machines without damage, the damage detection thresholds, i.e. the deviation values considered as synonymous with damage, being determined empirically and subjectively by experts.
• the diagnosis is not reliable since it can vary from one expert to another and requires the intervention of specialists.
• the invention offers a solution to the problems mentioned above, by making it possible to provide a reliable and automatic diagnosis of a part of a rotating machine.
• a first aspect of the invention relates to an automatic diagnostic method for a part of a rotating machine made from a non-stationary time signal generated by the rotating machine during at least one phase during which a rotational speed of the rotating machine varies as a function of time, the process comprising the following steps:
• Construction of a diagram from the signal comprising the following sub-steps: o Splitting of the signal into a plurality of sub-signals, each sub-signal corresponding to a time interval associated with at least one rotational speed of the rotating machine and being quasi-stationary over the time interval; o For each sub-signal, calculation of the Fourier transform of the sub-signal in order to obtain a vibrational energy for each frequency of the Fourier transform of the sub-signal; o Construction of the diagram, the diagram being a matrix presenting a plurality of rows each corresponding to a rotational speed of the rotating machine, ordered in ascending order, and a plurality of columns each corresponding to a frequency of the Fourier transform divided by a rotational speed of the rotating machine, ordered in ascending order, the matrix comprising for each line and each column, the vibratory energy of the sub-signal corresponding to the speed of rotation of the rotating machine of the line for the frequency of the Fourier transform of the column;
• Supervised training of an artificial neural network to obtain a trained artificial neural network capable of providing, from a diagram, a class of operation included in a set of classes of operation comprising at least one class of nominal operation and one class faulty operation, the artificial neural network being trained on a training database comprising a plurality of training diagrams, each training diagram being constructed from a non-stationary temporal signal generated by a rotating machine of training and being associated with a class of operation among the set of classes of operation;
• the diagnosis of a part of a rotating machine is automated, since once the operating classes of the training database have been obtained, no human intervention is necessary.
• the construction of a diagram representing the vibratory energy of the temporal signal according to the frequency and the speed of rotation of the rotating machine makes it possible to obtain an image which can be supplied to an artificial neural network specialized in image processing, to detect in the diagram, signatures synonymous with damage and thus to determine an operating class for the part, that is to say to determine whether the part is defective or not.
• the time signal is processed for different speeds of rotation because the manifestation of damage can occur on only part of the speeds of rotation taken by the rotating machine.
• the trained artificial neural network provides results that are 96.25% accurate with a faulty engine detection rate of 95% and a false alarm rate of 2.5% and bases its decision on the zones of the diagram corresponding to the signatures conventionally used by the experts to make their diagnosis.
• the diagnosis provided by the method according to the invention can therefore be considered reliable.
• the method according to the invention may have one or more additional characteristics from among the following, considered individually or according to all the technically possible combinations:
• the signal is a vibration signal .
• the part is a bearing included in the rotating machine.
• the rotating machine is a motor.
• the diagram construction step includes a sub-step for applying a logarithmic scale to the constructed diagram;
• the diagram construction step includes a substep of applying a reduction in the size of the constructed diagram by a predetermined factor.
• the method according to the invention comprises a step of construction of the training database, comprising the following sub-steps: o For each non-stationary time signal generated by a rotating training machine, construction of an initial diagram to from the signal; o For each initial diagram constructed, standard normalization of the initial diagram constructed to obtain a training diagram. Standard normalization makes it possible to homogenize the training database.
• the artificial neural network is a convolutional artificial neural network.
• a second aspect of the invention relates to a computer configured to implement the steps of the method according to the invention.
• a third aspect of the invention relates to a computer program product comprising instructions which, when the program is executed on a computer, lead the latter to implement the steps of the method according to the invention.
• a fourth aspect of the invention relates to a computer-readable recording medium comprising instructions which, when executed by a computer, lead the latter to implement the steps of the method according to the invention.
• FIG. 1 shows a plot of a non-stationary temporal signal as a function of time.
• FIG. 2 shows a block diagram of a method according to the invention.
• FIG. 3 is a diagram obtained at the end of a first step of the method according to the invention.
• Figure 4 shows a confusion matrix of the results provided by an artificial neural network trained during a third step of the method according to the invention.
• FIG. 5 shows a diagram obtained at the end of the first step of the method according to the invention, on which has been represented via a heat map, the importance of areas in the decision making of the trained artificial neural network.
• the present invention relates to a method for diagnosing a part of a rotating machine, the diagnosis being automatic, that is to say that it does not require any human intervention.
• Diagnosis of a part of a rotating machine means the determination of the operating state of the part of the rotating machine, an operating state being for example a nominal operating state or a state faulty operation.
• the rotating machine is preferably an engine, for example a piston engine.
• the rotating machine is preferably included in an aircraft, such as for example a turbojet engine.
• the part is, for example, one or more gears, a rotor-stator assembly, a rotating shaft, an oil imbalance, a blading wake, a moving wheel wake or even one or more bearings.
• the rotating machine is a motor, and the part is a bearing.
• the method according to the invention is carried out from a non-stationary time signal generated by the rotating machine.
• non-stationary time signal is meant a time-based physical signal whose frequency content varies over time.
• non-stationary time signal time signal or “signal” will be used interchangeably.
• the time signal is generated during at least one phase during which a rotational speed of the rotary machine varies as a function of time.
• the fact that the time signal is non-stationary is due to the fact that the speed of rotation of the rotating machine varies during the generation phase of the time signal.
• the signal is for example measured using a sensor, possibly embedded in the rotating machine, for example an electric, magnetic, thermal, acoustic or vibration sensor.
• the senor is a vibration sensor, in particular an accelerometer.
• Figure 1 shows an example of a non-stationary time signal, the abscissa axis representing the time and the ordinate axis representing the amplitude of the signal.
• the signal is a vibration signal.
• the method according to the invention can be carried out as part of a check at the exit of the production line of the rotating machine or during maintenance of the rotating machine.
• the signal is for example a vibratory signal generated during an acceleration and deceleration phase of the motor in operation.
• the method 100 according to the invention comprises a plurality of steps, the sequence of which is shown in Figure 2.
• a first step 101 of the method 100 according to the invention is a step 101 of constructing a diagram from the signal.
• the first step 101 includes a plurality of sub-steps.
• a first sub-step 1011 of the first step 101 is a sub-step of splitting the signal into a plurality of sub-signals.
• the cutting is realized in such a way that each sub-signal is quasi-stationary over the corresponding time interval.
• Each time interval, and therefore each sub-signal is associated with at least one rotational speed of the rotating machine corresponding to the rotational speed assumed by the rotating machine over the time interval.
• a second sub-step 1012 of the first step 101 is a sub-step of calculating, for each sub-signal, the Fourier transform of the sub-signal, which makes it possible to obtain, for each frequency, the transform of Calculated Fourier, that is to say for each frequency of the bandwidth of the sub-signal, a vibrational energy associated with the frequency in the sub-signal.
• the vibrational energy therefore corresponds to the frequency amplitude of the sub-signal.
• a third sub-step 1013 of the first step 101 is a diagram construction sub-step.
• the diagram is a matrix having a plurality of rows and a plurality of columns.
• Each line corresponds to a speed of rotation taken by the rotating machine during the generation phase of the time signal and each column corresponds to a frequency of the Fourier transform calculated in the second sub-step 1012, divided by a speed of rotation taken by the rotating machine on the time signal generation phase.
• the plurality of lines is ordered in ascending order, that is to say that a first line corresponds to a rotational speed lower than a rotational speed of a second line.
• the plurality of columns are ordered in ascending order.
• the diagram For each row of the plurality of rows and each column of the plurality of columns, the diagram includes the vibration energy of the sub-signal associated with the rotational speed of the rotating machine corresponding to the row for the frequency of the Fourier transform corresponding to the column.
• the constructed diagram is therefore two-dimensional and can therefore be likened to an image in grayscale or in color.
• FIG. 3 An example of a constructed diagram is represented in FIG. 3.
• Each point or pixel of the constructed diagram corresponds to a given rotational speed, a given frequency divided by the given rotational speed, and to a given vibrational energy whose intensity is represented by a color defined by a color scale.
• the intensity of the vibrational energies can be represented by a level of gray defined by a scale of gray levels.
• the constructed diagram comprises for example a number of pixels comprised between two million and twelve million.
• the first step 101 of the method 100 further comprises a sub-step 1014 of applying a logarithmic scale to the constructed diagram obtained at the end of the third sub-step 1013.
• the first step 101 further comprises a sub-step 1015 of reducing the size of the constructed diagram obtained at the end of the third sub-step 1013 of a predetermined factor.
• the predetermined factor is for example equal to ten.
• the reduction in the size of the constructed diagram can be achieved by means of a sub-sampling operation by maximum value, called "maxpooling".
• the method further comprises a third step 103 of supervised training of an artificial neural network on a training database, to obtain a trained artificial neural network capable of providing from a diagram, an operating class.
• the operating class is included in a set of operating classes comprising at least one nominal operating class and one faulty operating class.
• the set of operating classes may additionally include an at-risk operating class.
• Supervised training otherwise called supervised learning, makes it possible to train an artificial neural network for a predefined task, by updating its parameters so as to minimize a cost function corresponding to the error between the data of output provided by the artificial neural network and the real output datum, i.e. what the artificial neural network should output to fulfill the predefined task on a certain input datum.
• the training database therefore includes input data, each associated with output data.
• the input data are training diagrams, each training diagram being constructed from a non-stationary time signal generated by a rotating training machine and each associated with a class of operation among the set of operating classes, the operating classes therefore being the output data.
• Each rotating drive machine is of the same type as the rotating machine, that is to say that if the rotating machine is an aircraft engine, each rotating drive machine is also an aircraft engine. .
• the supervised training of the artificial neural network therefore consists in updating the parameters of the artificial neural network so as to minimize a cost function corresponding to the error between the prediction of the class of operation provided by the artificial neural network from a training diagram in the training database and the class of operation associated with said training diagram in the training database.
• the cost function results for example from the composite of the binary cross-entropy function by the sigmoid function.
• the minimization of the cost function is for example carried out by a stochastic gradient descent algorithm with backpropagation (“Back-Propagation Through Time” in English or BPTT).
• the artificial neural network is preferably a convolutional artificial neural network.
• the artificial neural network is for example the artificial neural network VGG19.
• the method 100 comprises a second step 102 of building the training database, carried out prior to the third step 103, comprising a plurality of sub-steps.
• a first sub-step 1021 of the second step 102 consists, for each signal generated by a rotating drive machine, in constructing an initial diagram from the signal.
• the first sub-step 1021 of the second step 102 is carried out in the same way as the first step 101, that is to say by dividing the signal into sub-signals, by calculating the Fourier transform of each sub-signal to obtain vibrational energies and constructing the initial diagram using the obtained vibrational energies.
• a second sub-step 1022 of the second step 102 consists, for each initial diagram obtained at the end of the first sub-step 1021, in applying a standard normalization, otherwise called z-score normalization, to the initial diagram to obtain a training diagram.
• a standard normalization otherwise called z-score normalization
• the method comprises a fourth step 104 of using the artificial neural network trained on the constructed diagram obtained at the end of the first step 101, to obtain an operating class and therefore a diagnosis for the part of the machine. rotating.
• Each engine from a set comprising 1665 engines performed an identical maneuver on the test bench. This maneuver is an acceleration of the engine followed by a deceleration of the engine. During this maneuver, a non-stationary temporal vibration signal was measured by an accelerometer included in the motor.
• a database was formed according to the second step 102 of the method 100 according to the invention from the 1665 measured vibration signals, to obtain 1665 diagrams and a class of nominal operation or a class of faulty operation was associated manually to each diagram obtained by experts.
• the set of 1665 diagrams and their associated classes constitute an annotated database which has been divided into three parts: a training database, a validation database and a test database according to proportions represented in the table below:
• the artificial neural network used is the artificial neural network VGG19.
• the artificial neural network was trained in a supervised manner on the training database and the training was validated using the validation database.
• the diagrams included in the test database were provided to the trained artificial neural network and several quality criteria were evaluated by comparison for each diagram of the database. of test data, between the class of operation provided by the artificial neural network for said diagram and the class of operation associated with said diagram in the test database.
• the rate of false alarms a false alarm being in this case the classification of a defective motor in the category of motors exhibiting nominal operation.
• the quality criteria presented in the previous table can also be represented in the form of a confusion matrix, as shown in Figure 4.
• Each column of the matrix corresponds to the true class of operation, that is, i.e. the class of functioning included in the test database, and each row of the matrix corresponds to the class of functioning provided by the artificial neural network.
• the confusion matrix then reads as follows: vertically, for the first column, of the 40 motors in nominal operation, 39 were estimated by the artificial neural network as motors in nominal operation and 1 motor was estimated as faulty (i.e. a false negative), vertically, for the second column, out of the 40 faulty motors, 2 were estimated by the neural network as motors in nominal operation (i.e.
• FIG. 5 represents a diagram on which the Grad-Cam module has been applied.
• the highlighted areas correspond well to engine bearing damage signatures, identified manually by human operators, which means that the artificial neural network correctly identifies the engine bearing damage signatures to make the decision to assign a malfunction class.

Landscapes

• Physics & Mathematics (AREA)
• Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Computational Linguistics (AREA)
• General Health & Medical Sciences (AREA)
• Life Sciences & Earth Sciences (AREA)
• Artificial Intelligence (AREA)
• Biomedical Technology (AREA)
• Biophysics (AREA)
• Software Systems (AREA)
• Data Mining & Analysis (AREA)
• Evolutionary Computation (AREA)
• Health & Medical Sciences (AREA)
• Molecular Biology (AREA)
• Computing Systems (AREA)
• General Engineering & Computer Science (AREA)
• Mathematical Physics (AREA)
• Acoustics & Sound (AREA)
• Chemical & Material Sciences (AREA)
• Combustion & Propulsion (AREA)
• Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
• Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=84689025

Family Applications (1)

Country Status (5)

Family Cites Families (6)

• 2021
  • 2021-12-02 FR FR2112840A patent/FR3130028B1/fr active Active
• 2022
  • 2022-12-01 US US18/713,562 patent/US20250020545A1/en active Pending
  • 2022-12-01 WO PCT/FR2022/052216 patent/WO2023099849A1/fr not_active Ceased
  • 2022-12-01 CN CN202280080166.1A patent/CN118302757A/zh active Pending
  • 2022-12-01 EP EP22834691.2A patent/EP4441663A1/de active Pending

Also Published As

Similar Documents

Legal Events