Apparatus and method for predicting physical parameter measured by sensorized rolling element

Abstract

The invention proposes a device (8) for predicting at least one physical parameter measured by a sensorized rolling element in a rolling bearing of a machine. The device (8) comprises:-a first determination means (40) configured to determine at least a set of machine operating conditions representing an operating condition of the machine,-a memory (41) storing a conditional interpolation model (42) configured to predict the machine physical parameters from the machine operating conditions, and-an implementation means (43) configured to implement the machine physical parameters based on the conditional interpolation model (42). And a processor configured to implement the conditional interpolation model (42).

Description

Technical Field

[0001] This invention relates to forecasting time series values.

[0002] More specifically, the present invention relates to a method and apparatus for predicting time-series values ​​of sensorized rolling elements implemented in rolling bearings of a machine. Background Technology

[0003] Document US10,371,206 discloses a sensorized rolling element that includes a measuring device.

[0004] Sensor-equipped rollers are embedded in the machine's bearings.

[0005] The measuring device includes sensors such as load sensors, accelerometers, and gyroscopes.

[0006] The sensor's measurement results (measurements) are wirelessly transmitted to an external receiver.

[0007] The measurement results are used to determine, for example, the contact pressure in a bearing or the remaining life of a bearing to monitor the bearing.

[0008] It is well known that physical models are implemented to monitor bearings from machine data. The simulation model outputs a prediction of the measurement results delivered by sensor-modified rollers to determine the contact pressure in the bearing or the remaining life of the bearing.

[0009] The prediction is very accurate and easy to interpret.

[0010] However, implementing a physical model is time-consuming and may require significant computing power.

[0011] Furthermore, the physical model is also sensitive to boundary conditions, making it difficult to quickly change the test conditions.

[0012] Therefore, the present invention aims to make the prediction of measurement results delivered by sensorized rolling elements in a machine more robust, without requiring significant resources. Summary of the Invention

[0013] Based on one aspect, a method for at least one machine physical parameter is proposed.

[0014] The physical parameters are measured by sensorized rolling elements in a machine that includes rolling bearings comprising stationary and mobile rings capable of rotating concentrically relative to each other, and at least one row of rolling elements located between a first raceway on the first ring and a second raceway on the second ring, wherein at least one of the rolling elements is a sensorized rolling element.

[0015] The method includes:

[0016] - Determine at least a set of machine operating conditions, wherein the machine operating conditions represent the operating status of the machines, and

[0017] - Implement a conditional imputation model, which is configured to predict the machine's physical parameters from the machine's operating conditions.

[0018] The conditional interpolation model provides a quantitative prediction of physical parameters based on the machine's operating status.

[0019] The conditional interpolation model is a digital mapping of the sensorized rolling element, which can be used to predict physical parameters in a more cost-effective and efficient manner than using a physical model.

[0020] Preferably, the conditional interpolation model includes a conditional structured state space diffusion model.

[0021] Advantageously, the machine physical parameters are the load applied to the sensorized rolling element, the temperature of the sensorized rolling element, or the speed of the sensorized rolling element.

[0022] In another respect, a method for training a conditional interpolation model is proposed, the conditional interpolation model being configured to predict machine physical parameters from a set of machine operating conditions.

[0023] The physical parameters are measured by sensorized rolling elements in a machine that includes rolling bearings, which include stationary and moving rings capable of rotating concentrically relative to each other, and at least one row of rolling elements located between a first raceway disposed on the first ring and a second raceway disposed on the second ring. At least one of the rolling elements is a sensorized rolling element, and the operating condition parameters indicate the operating condition of the machine.

[0024] The method includes:

[0025] - Determine at least one training set, said at least one training set comprising time-series values ​​of machine physical parameters measured by the sensorized rolling elements in the machine and time-series values ​​of machine operating conditions associated with the time-series values ​​of the machine physical parameters.

[0026] - Implement the conditional interpolation model to determine the output set of machine physical parameter time series values ​​from the machine operating time series values ​​in the training set.

[0027] - Compare the time series values ​​of the machine physical parameters in the training set with the time series values ​​of the machine physical parameters in the output set, and

[0028] - Based on the comparison results, the conditional interpolation model is tuned to capture long-term dependencies in the time series values ​​and machine operating time series values.

[0029] Preferably, the conditional interpolation model includes a conditional structured state-space diffusion model.

[0030] Advantageously, the conditional structured state-space diffusion model includes a neural network, and the step of tuning the conditional interpolation model includes tuning the weights of the neural network based on the result of the first comparison.

[0031] According to another aspect, an apparatus is proposed for predicting at least one physical parameter of a machine measured by a sensorized rolling element in the machine.

[0032] The machine includes a rolling bearing comprising a stationary ring and a moving ring (or motion ring) capable of rotating concentrically relative to each other, and at least one row of rolling elements located between a first raceway disposed on a first ring and a second raceway disposed on a second ring, wherein at least one of the rolling elements is the sensorized rolling element.

[0033] The device includes:

[0034] - A first determining component is configured to determine at least a set of machine operating conditions, the machine operating conditions representing the overall operating status of the machine.

[0035] - A memory storing a conditional interpolation model configured to predict the machine's physical parameters from its operating conditions, and

[0036] - The implementation component is configured to implement the conditional interpolation model.

[0037] Advantageously, the apparatus includes a second determining component configured to determine at least one training set comprising time-series values ​​of machine physical parameters measured by sensorized rolling elements in the machine and time-series values ​​of machine operating conditions associated with the time-series values ​​of the machine physical parameters. The implementing component is further configured to implement the conditional interpolation model to determine an output set of machine physical parameter time-series values ​​from the machine operating condition time-series values ​​in the training set.

[0038] The device further includes:

[0039] - A comparison unit is configured to compare the time-series values ​​of machine physical parameters in the training set with the time-series values ​​of machine physical parameters in the output set, and

[0040] - A tuning component is configured to tune the conditional interpolation model based on the result of the first comparison to capture long-term correlations between time series values ​​and machine operating condition time series values.

[0041] According to another aspect, a system is proposed for predicting at least a set of time-series values ​​of at least one physical parameter measured by a sensorized rolling element in a machine.

[0042] The machine includes a rolling bearing comprising a stationary ring and a moving ring capable of rotating concentrically relative to each other, and at least one row of rolling elements, the at least one row of rolling elements being located between a first raceway disposed on the first ring and a second raceway disposed on the second ring, at least one of the rolling elements being the sensorized rolling element, and the system comprising:

[0043] -Based on the previously defined apparatus, and

[0044] - The machine includes the rolling bearing. Attached Figure Description

[0045] Other advantages and features of the invention will become apparent upon examination of the non-limiting detailed description and accompanying drawings of the embodiments, in which:

[0046] Figure 1 An example of a machine according to the invention is shown schematically;

[0047] Figure 2 An example of a roller bearing according to the invention is shown schematically;

[0048] Figure 3 An example of a sensorized rolling element according to the present invention is illustrated schematically;

[0049] Figure 4 An example of an apparatus according to the invention for predicting at least a set of time series values ​​(or at least a group of time series values) of at least one physical parameter is illustrated schematically.

[0050] Figure 5 An example of a conditional interpolation model based on the prior art is illustrated schematically; and

[0051] Figure 6 An example of a method for training a conditional interpolation model according to the present invention is illustrated schematically.

[0052] Figure 7 An example of a method for training a conditional interpolation model is shown. Detailed Implementation

[0053] refer to Figure 1 This schematically represents an example of a machine 1 including a rolling bearing 5.

[0054] Machine 1 may be a wind turbine, which includes a generator 2, a propeller 3, a shaft 4 connecting the shaft of the generator 2 to the propeller 3, and a roller bearing 5 supporting the shaft 4.

[0055] In other embodiments, machine 1 may be a tunnel boring machine, a mining extraction machine, or a big offshore crane.

[0056] Machine 1 may also include sensor 6 to measure values ​​representing the machine operating condition.

[0057] The machine's operating status can be, for example, wind speed, shaft 4 speed, machine temperature, or power generated by wind turbine 1.

[0058] The roller bearing 5 includes at least one sensorized rolling element 7.

[0059] The sensorized rolling element 7 is designed to measure at least one machine physical parameter and deliver a set of time-series values ​​of the machine physical parameter, the at least one machine physical parameter including, for example, a load applied to the sensorized rolling element 7 or the temperature or speed of the sensorized rolling element 7.

[0060] In a variant not shown, the wind turbine also includes a gearbox that connects the shaft of the generator 2 to the shaft 4 of the wind turbine. The rolling bearings of the gearbox may include sensorized rolling elements 7. An example of a rolling bearing 5 is described in detail below.

[0061] Sensor 6 and sensorized rolling element 7 communicate with device 8, which is designed to predict a set of time-series values ​​of physical parameters measured by sensorized rolling element 7.

[0062] An example of device 8 is described in detail below.

[0063] Machine 1 and device 8 form a system for predicting the physical parameters of the machine as measured by the sensorized rolling element 7.

[0064] Figure 2 An example of roller bearing 5 is shown schematically.

[0065] The bearing 5 includes an outer ring or stationary ring 9, which has a tapered first outer raceway for a first row of rolling elements 10 and a tapered second outer raceway for a second row of rolling elements 11. The rolling elements include tapered rollers. The bearing also includes a mobile ring, which has a first inner ring or mobile ring 12 and a second inner ring or mobile ring 13 stacked axially. The first inner ring or mobile ring 12 has a tapered first inner raceway for the first row of rollers 10, and the second inner ring or mobile ring 13 has a tapered second inner raceway for the second row of rollers 11. Furthermore, the bearing 5 includes a first cage 14 for holding the rollers of a first group of rollers (or roller rows) and a second cage 15 for holding the rollers of a second group of rollers. Typically, the cage can be formed from segments that abut against each other in the circumferential direction.

[0066] To provide the necessary stiffness and ensure a long service life, the bearings are preloaded. The axial positions of the moving rings 12 and 13 relative to the stationary ring 9 are set such that the first roller group 9 and the second roller group 11 have negative internal clearance.

[0067] In the modified version, the bearing was not preloaded.

[0068] In the depicted bearing, at least one rolling element in either the first roller row 10 or the second roller row 11 is replaced with a sensorized rolling element 7. The shaft 4 is surrounded by and fixed to the moving rings 12 and 13.

[0069] Rolling bearing 5 includes tapered rollers.

[0070] In another embodiment, the rolling bearing 5 may include other types of rolling elements, such as cylindrical rollers or spherical rollers. The rolling bearing 5 may also include only one row of rolling elements or more than two rows of rolling elements, the number of cages being determined according to the number of rows.

[0071] The rolling bearing 5, which includes a row of rolling elements, includes a single inner ring.

[0072] In another embodiment, the outer ring 9 is a moving ring, and the inner rings 12 and 13 are stationary rings.

[0073] Figure 3 An example of a sensor-based rolling element 7 is shown schematically.

[0074] The sensor-controlled rolling element 7 includes a roller body 16 and a sensor unit 18. The roller body 16 includes a central hole 17, and the sensor unit 18 extends through the roller body 16 within the central hole 17.

[0075] The sensor unit 18 includes a housing 19 formed by two semi-cylindrical housings, which are fixed together by a first end cap 20 and a second end cap 21. The first end cap 20 and the second end cap 21 are screwed onto corresponding first threaded portions 22 and second threaded portions 23 at opposite axial ends of the housings. The sensor unit housing is integrally formed to fit within a roller bore 17 and is mounted into and located within the bore 17 by a first sealing element 24 and a second sealing element 25.

[0076] Sensor unit 18 is designed to measure physical parameters and deliver a set of time-series values ​​of the physical parameters.

[0077] Sensor unit 18 also includes a load sensor 26 for measuring the load value applied to the sensorized rolling element 7.

[0078] The sensor unit 18 may also include a speed sensor 27 for measuring the rotational speed of the sensorized rolling element 7 in the bearing 5, and may also include a temperature sensor 28 for measuring the temperature of the sensorized rolling element 7.

[0079] The sensor unit 18 includes a wireless transmitter 29 for transmitting a set (or multiple sets of measurement results) of multiple measurements from sensors 26, 27, and 28; a sampler 30 for sampling the signals delivered by the sensors; and a battery 31 for powering the sensors 26, 27, 28, and the wireless transmitter.

[0080] Figure 4 An example of device 8 is illustrated schematically.

[0081] The device 8 includes a first determining component 40, which is designed to determine a set of machine operating conditions from values ​​delivered by the sensor 6.

[0082] Device 8 also includes a memory 41 storing the conditional imputation model 42, an implementation component 43, a comparison component 44, a tuning component 45, and a second determination component 46.

[0083] The conditional interpolation model 42 may include a conditional structured state space diffusion model, such as a neural network.

[0084] Figure 5 An example of a conditional interpolation model 42, which includes a conditional structured state space diffusion model, is illustrated schematically.

[0085] A paper titled "Diffusion-based time series interpolation and prediction using structured state-space models" by Juan Miguel Lopez Alcaraz and Niels Stratotoff of the University of Oldenburg, Germany, discloses an example of a conditional structured state-space diffusion model.

[0086] Diffusion-based time series interpolation and prediction using structured state-space models are methods for handling missing input data in machine learning applications.

[0087] The conditional structured state-space diffusion model 50 includes a neural network containing 1D convolutional layers and S4 layers.

[0088] The conditionally structured state-space diffusion model 50 includes a first input (end) 51 designed to receive time series data, a second input (end) 52 designed to receive an interpolation mask, and an output (end) 53.

[0089] Note that the conditional interpolation model 42 is trained to minimize at least one cost function associated with, for example, mean squared error, mean absolute error, or binary cross-entropy.

[0090] The conditional interpolation model 42, which is stored in memory 41 and implemented by implementation component 43, predicts machine physical parameters from the set of machine operating conditions of the first determining component 40.

[0091] The first input 51 is a set of time series values ​​of the machine's operating status, the second input 52 is set to zero, and the output 53 is a set of time series values ​​of the predicted machine physical parameters.

[0092] Figure 6 An example of a set of time-series values ​​of machine physical parameters predicted by the trained conditional interpolation model 42 is illustrated.

[0093] Machine physical parameters include load.

[0094] Curve C1 represents the load measured by the sensorized rolling element 7 over time t, including missing values.

[0095] Curve C2 represents the load predicted by the trained conditional interpolation model 42 from a set of time-series values ​​of machine operating parameters associated with the load measured by the sensorized rolling element 7.

[0096] Figure 7 An example of a method for training conditional interpolation model 42 is shown.

[0097] In step 70, the second determining component 46 determines at least one training set, which includes time series values ​​of machine physical parameters measured by the sensorized rolling element 7 in the machine 1 and time series values ​​of machine operating conditions associated with the time series values ​​of the machine physical parameters.

[0098] In step 71, the conditional interpolation model 42 is trained from the training set or multiple training sets.

[0099] For each training set, implementation component 43 implements conditional interpolation model 42 to determine an output set of machine physical parameter time series values ​​from the machine operating time series values ​​of the training set.

[0100] The machine operating status time series values ​​of the training set are input into the first input 51 of the conditional structured state space diffusion model 50, and the second input 52 of the conditional structured state space diffusion model 50 is set to zero.

[0101] In step 71, for each training set, the comparison unit 44 performs a first comparison between the output set of machine physical parameter time series values ​​and the machine physical parameter time series values ​​of the training set.

[0102] Comparison component 44 can implement the mean absolute error algorithm.

[0103] The tuning means 45 tunes the conditional interpolation model 42 based on the result of the first comparison to capture the long-term dependencies between the time series values ​​and the machine operating condition time series values.

[0104] The tuning component 45 tunes the conditional interpolation model 42 based on the result of the first comparison in order to minimize the cost function associated with, for example, the mean absolute error.

[0105] When the conditional interpolation model 42 is composed of a neural network, the tuning component 45 tunes the weights of the neural network according to the result of the first comparison.

[0106] The tuning component 45 tunes the weights of the neural network based on the result of the first comparison in order to minimize the cost function associated with, for example, the mean absolute error.

[0107] The method may also include a verification step 72 and a verification step 73 of the trained conditional interpolation model 42 to check whether the accuracy of the time series values ​​of the machine physical parameters determined by the conditional interpolation model 42 is sufficient.

[0108] Suppose we have obtained a validation set.

[0109] Each validation set includes time-series values ​​of machine physical parameters measured by sensorized rolling elements 7 in machine 1, and time-series values ​​of machine operating conditions associated with those time-series values.

[0110] The verification set can be obtained through the second determining component 46.

[0111] In step 72, for each verification step, the implementation component 43 implements the conditional interpolation model 42 based on the machine operating status time series values ​​of the verification set to determine a second output set of time series values.

[0112] In step 73, for each validation set, the comparison unit 44 performs a second comparison between the second output set of machine physical parameter time series values ​​and the machine physical parameter time series values ​​of the validation set.

[0113] The tuning component 45 tunes the conditional interpolation model 42 based on the result of the second comparison to minimize the cost function.

[0114] Device 8 provides a quantitative prediction of the set of time series values ​​of physical parameters based on the set of time series values ​​of machine operating status.

[0115] Conditional interpolation model 42 is a digital mapping ( / twin) of sensorized rolling element 7.

[0116] Conditional interpolation model 42 can be used to predict physical parameters in a more cost-effective and efficient manner than using a physical model.

[0117] Conditional interpolation model 42 can be used to identify potential problems and inefficiencies in machine 1 before they occur, in order to optimize the design and operation of machine 1.

[0118] Conditional interpolation model 42 can provide real-time data on the performance of machine 1, so as to run machine 1 in a more efficient manner.

[0119] In cases where the sensorized rolling element loses its ability to collect data, such as when the battery powering the sensorized rolling element 7 is depleted, or when the roller bearing 5 does not include the sensorized rolling element 7, the conditional interpolation model 42 can be used as a virtual sensor.

[0120] Conditional interpolation model 42 allows simulation of machine 1's operation under different conditions to identify potential failure points and take measures to prevent them from occurring on machine 1, thereby improving the reliability of machine 1.

Claims

1. A method for predicting at least one physical parameter of a machine (1) measured by a sensorized rolling element (7), the machine comprising a rolling bearing (5) including a stationary ring (9) and a moving ring (12, 13) rotatable concentrically relative to each other, and at least one row of rolling elements (7, 10, 11) located between a first raceway disposed on a first ring and a second raceway disposed on a second ring, at least one of the rolling elements (7, 10, 11) being the sensorized rolling element (7), the method comprising: - Determine at least a set of machine operating conditions, wherein the machine operating conditions represent the operating status of the machine, and - Implement a conditional interpolation model (42), which is configured to predict the physical parameters of the machine from the machine's operating conditions.

2. The method according to claim 1, characterized in that, The conditional interpolation model (42) includes a conditional structured state space diffusion model.

3. The method according to claim 1 or 2, characterized in that, The physical parameters of the machine are the load applied to the sensorized rolling element (7), the temperature of the sensorized rolling element (7), or the speed of the sensorized rolling element (7).

4. A method for training a conditional interpolation model (42), the conditional interpolation model (42) being configured to predict machine physical parameters from a set of machine operating conditions, the machine physical parameters being measured by sensorized rolling elements (7) in a machine (1), the machine including rolling bearings (5) including stationary rings (9) and moving rings (12, 13) capable of rotating concentrically relative to each other, and at least one row of rolling elements (7, 10, 11), the at least one row of rolling elements (7, 10, 11) being located between a first raceway disposed on a first ring and a second raceway disposed on a second ring, at least one of the rolling elements (7, 10, 11) being a sensorized rolling element (7), the operating condition parameters representing the operating condition of the machine, the method comprising: - Determine at least one training set, said at least one training set comprising time series values ​​of the machine's physical parameters measured by the sensorized rolling element (7) in the machine (1) and time series values ​​of the machine's operating conditions associated with the time series values ​​of the machine's physical parameters. - Implement the conditional interpolation model (42) to determine the output set of time series values ​​of machine physical parameters from the time series values ​​of machine operating conditions in the training set. - Compare the time series values ​​of the machine physical parameters in the training set with the time series values ​​of the machine physical parameters in the output set, and - Tune the conditional interpolation model (42) based on the comparison results to capture the long-term correlation between time series values ​​and time series values ​​of machine operating conditions.

5. The method according to claim 4, characterized in that, The conditional interpolation model (42) includes a conditional structured state space diffusion model.

6. The method according to claim 5, characterized in that, The conditional structured state space diffusion model (42) includes a neural network, and the step of tuning the conditional interpolation model includes tuning the weights of the neural network based on the result of the first comparison.

7. An apparatus (8) for predicting at least one physical parameter of a machine measured by a sensorized rolling element (7) in a machine (1), the machine comprising a rolling bearing (5) including a stationary ring (9) and a moving ring (12, 13) rotatable concentrically relative to each other, and at least one row of rolling elements (7, 10, 11) located between a first raceway disposed on a first ring and a second raceway disposed on a second ring, at least one of the rolling elements (7, 10, 11) being the sensorized rolling element (7), the apparatus comprising: - A first determining component (40) is configured to determine at least a set of machine operating conditions, the machine operating conditions representing the operating condition of the machine (1). - Memory (41), storing a conditional interpolation model (42), the conditional interpolation model (42) being configured to predict the machine's physical parameters from the machine's operating conditions, and - The implementation component (43) is configured to implement the conditional interpolation model (42).

8. The apparatus according to claim 7, characterized in that: The device includes a second determining component (46) configured to determine at least one training set, the at least one training set comprising time-series values ​​of machine physical parameters measured by sensorized rolling elements (7) in the machine (1) and time-series values ​​of machine operating conditions associated with the time-series values ​​of the machine physical parameters. The implementation component (43) is also configured to implement the conditional interpolation model (43) to determine an output set of time series values ​​of machine physical parameters from the time series values ​​of machine operation in the training set. The device (8) further includes: - Comparison unit (44) is configured to compare the time-series values ​​of the machine physical parameters of the training set with the time-series values ​​of the machine physical parameters of the output set, and - Tuning component (45) is configured to tune the conditional interpolation model (42) according to the result of the first comparison to capture the long-term correlation between time series values ​​and time series values ​​of machine operating conditions.

9. A system for predicting at least a set of time-series values ​​of at least one physical parameter measured by a sensorized rolling element (7) in a machine (1), the machine comprising a rolling bearing (5) including a stationary ring (9) and a moving ring (12, 13) capable of rotating concentrically relative to each other, and at least one row of rolling elements (7, 10, 11) located between a first raceway disposed on a first ring and a second raceway disposed on a second ring, at least one of the rolling elements (7, 10, 11) being the sensorized rolling element (7), the system comprising: - The apparatus (8) according to claim 7 or 8, and -The machine (1) includes the rolling bearing (5).

Images