Method for identifying a main damage mode of a machine component and associated device
A statistical digital twin model accurately identifies the main damage mode of machine components by analyzing operating parameters, improving maintenance strategies and component longevity through precise damage history matching.
Patent Information
- Application Number
- PCT/EP2026/063644
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-14
- Filing Date
- 2026-05-11
- Publication Date
- 2026-11-19
AI Technical Summary
Existing condition monitoring systems for machine components like bearings fail to accurately identify the root cause of damage, relying heavily on technician experience which is often insufficient.
A method and device utilizing a statistical simulation model, known as a 'statistical digital twin', to analyze operating parameters and build a damage history, matching simulated and real damage histories to determine the main damage mode of components across multiple machines, excluding root causes with higher certainty.
Accurately identifies the main damage mode of machine components without relying on technician experience, enabling more precise maintenance strategies and enhancing component performance by simulating scenarios to improve longevity.
Smart Images

Figure EP2026063644_19112026_PF_FP_ABST
Abstract
Description
[0001] 202500044
[0002] 1
[0003] Method for identifying a main damage mode of a machine component and associated device
[0004] The present invention is directed to the determination of the most likely damage mode of a component, and more particularly to a method for identifying a main damage mode of the component which is arranged in a plurality of same type of machines.
[0005] The invention is also related to a device for identifying a main damage mode of said component.
[0006] Components of a machine which withstand heavy and / or cyclical stresses are monitored by a condition monitoring system to detect a damage of the components.
[0007] Such components may be bearings.
[0008] Generally, to monitor bearings, each bearing may be provided with a vibration sensor delivering a signal representative to the vibrations of said bearing.
[0009] The signal is processed by the condition monitoring system to detect a damage of the bearing, for example a damage on the raceway of the inner or outer ring of the bearing.
[0010] The condition monitoring system indicates a damage which has occurred and generally does not point out the cause which initiates the root cause.
[0011] A root cause of the damage may be determined empirically by an experienced technician using tools estimating rating life and lubricant life of the bearing.
[0012] However, the determination of the root cause depends on the experience of the technician which is often not enough to identify the real root cause of the damage.
[0013] Consequently, the present invention intends to enhance the identification of the damage mode of a component arranged in a machine.
[0014] According to an aspect, a method for identifying a main damage mode of a component is proposed. The component is arranged in a plurality of same type of machines.
[0015] The method comprises:202500044
[0016] 2
[0017] (a) for each machine of the plurality of same type of machines, collecting at least a first operating parameter of the component representative of a duration between the installation of the component in the machine and the occurrence of a damage of the component, (b) aggregating the first operating parameter values of the plurality of machines to determine a real damage history of the component, the real damage history associating each damage of said component to a damage duration,
[0018] (c) determining at least a first statistical model representing the progression of damages of a first predetermined damage mode depending on the usage of the component,
[0019] (d) performing variation of at least one parameter of the first statistical model so that a simulated damage history determined from the first statistical model matches with the real damage history within a predetermined range, the simulated damage history associating a plurality of damages to durations,
[0020] (e) when the simulated damage history matches with the real damage history within the predetermined range, determining the contribution of each statistical model in the simulated damage history, the main damage mode of the component being the predetermined damage mode represented by the statistical model having the most significant contribution in the simulated damage history.
[0021] The method builds and implements a statistical simulation model of the component as a “statistical digital twin”.
[0022] The most relevant damage mode of the component may be identified from the digital twin and damage statistics collected on machines comprising the component without implementing known condition monitoring devices.
[0023] A series of root causes of damage can be excluded with higher certainty to determine why the component works and fails.
[0024] The statistical digital twin may further be implemented to simulate different scenarios and compare those to each other to defined countermeasures to enhance the performance of the component, for example the lifetime of the component arranged in the machine.202500044
[0025] 3
[0026] The statistical simulation model may further represent the damage progression from initiating event to detected failure of the component, where the damage progression depends on the operating parameters of the component.
[0027] Preferably, step (c) comprises determining at least a second statistical model representing the progression of a second predetermined damage mode and step (d) further comprises performing variation of at least one parameter of the second statistical model so that the simulated damage history determined from the first statistical model and the second statistical model match with the real damage history within the predetermined range.
[0028] Advantageously, step (d) comprises:
[0029] (g) determining the simulated damage history from a set of parameter values of each statistical model,
[0030] (h) performing pattern recognition of the simulated damage history and real damage history to determine a similarity score,
[0031] if the similarity score is within the predetermined range, the simulated damage history matches with the real damage history,
[0032] if the similarity score is outside the predetermined range, steps (g) and (h) are repeated with another set of parameter values of each statistical model.
[0033] Preferably, step (e) comprises:
[0034] - for each statistical model, determining an auxiliary damage history from said statistical model, the auxiliary damage history associating a plurality of damages to durations and performing pattern recognition of the auxiliary damage history and the simulated damage history to determine an auxiliary similarity score,
[0035] - when the auxiliary similarity score is determined for each statistical model, determining the auxiliary similarity score having the biggest value, the contribution of each mode being equal to the auxiliary similarity score of said statistical model.
[0036] Advantageously, step (a) further comprises collecting at least a second operating parameter of the machines representative of the use of the machines, step (d) comprising defining range of variation of at least one parameter of at least one statistical model.202500044
[0037] 4
[0038] Preferably, the component is a bearing.
[0039] Advantageously, the first operating parameter comprises the time duration or the cumulative number of rotations of the bearing.
[0040] Preferably, the first statistical model comprises a rating life model or a grease life model or a corrosion model.
[0041] According to another aspect, a device for identifying a main damage mode of a component is proposed. The component is arranged in a plurality of same type of machines.
[0042] The device comprises:
[0043] - collecting means configured to collect for each machine of the plurality of same type of machines, at least a first operating parameter of the component representative of a duration between the installation of the component in the machine and the occurrence of a damage of the component,
[0044] - aggregating means configured to aggregate the first operating parameter values of the plurality of machines to determine a real damage history of the component, the real damage history associating each damage of said component to a damage duration,
[0045] - first determining means configured to determine at least a first statistical model representing the progression of damages of a first predetermined damage mode depending on the usage of the component, - performing means configured to perform variation of at least one parameter of the first statistical model so that a simulated damage history determined from the first statistical model matches with the real damage history within a predetermined range, the simulated damage history associating a plurality of damages to durations,
[0046] - second determining means configured to determine the contribution of each statistical model in the simulated damage history when the simulated damage history matches with the real damage history within the predetermined range and to determine the main damage mode of the component which is the predetermined damage mode represented by the statistical model having the most significant contribution in the simulated damage history.
[0047] Preferably, the component is a bearing.202500044
[0048] 5
[0049] Other advantages and features of the invention will appear on examination of the detailed description of embodiments, in no way restrictive, and the appended drawings in which:
[0050] Figure 1 illustrates schematically a machine;
[0051] Figure 2 illustrates schematically an example of a device for identifying a main damage mode of a component of the machine according to the invention;
[0052] Figure 3 illustrates schematically an example of a method for identifying a main damage mode of the component of the machine according to the invention;
[0053] Figure 4 illustrates schematically a first example of a simulated damage history according to the invention;
[0054] Figure 5 illustrates schematically a second example of a simulated damage history according to the invention; and
[0055] Figure 6 illustrates schematically a third example of a simulated damage history according to the invention.
[0056] Reference is made to figure 1 which represents schematically a partial longitudinal cross section of a machine 1.
[0057] The machine 1 comprises a housing 2 and a shaft 3 supported in the housing 2 by a rolling bearing 4 (e.g. roller bearing or ball bearing).
[0058] The rolling bearing 4 is provided with a rotating ring 5 mounted on the shaft 3, and with a stationary ring 6 mounted into the bore of the housing 2. The stationary ring 6 radially surrounds the rotating ring 5. The rotating and stationary rings 5, 6 rotate concentrically relative to one another.
[0059] The rolling bearing 4 is further provided with a row of rolling elements 7 radially interposed between inner and outer raceways of the rotating and stationary rings 5, 6. In the illustrated example, the rolling elements 7 are balls. Alternatively, the rolling bearing may comprise other types of rolling elements 7, for example rollers. In the illustrated example, the rolling bearing comprises one row of rolling elements 7. Alternatively, the rolling bearing may comprise several rows of rolling elements.
[0060] Figure 2 illustrates schematically an example of a device 9 for identifying a main damage mode of a component of the machine 1.202500044
[0061] 6
[0062] It is assumed that the component of the machine 1 is the rolling bearing 4.
[0063] Of course, the component of the machine 1 may be another component of the machine, for example a seal of the machine 1, an electric motor of the machine 1 or an electronic component of the machine 1.
[0064] The device 9 comprises collecting means 10, aggregating means 11, first determining means 12, performing means 13 and second determining means 14.
[0065] The device 9 further comprises a processing unit 15 implementing the collecting means 10, the aggregating means 11, the first determining means 12, the performing means 13 and the second determining means 14.
[0066] The device 9 comprises a memory 16 storing a plurality of statistical models 17, 18, 19.
[0067] Each statistical model 17, 18, 19 represents the progression of damages of a predetermined damage mode of the component depending on the usage of the component.
[0068] Of course, each statistical model 17, 18, 19 represents the progression of a different predetermined damage mode.
[0069] Figure 3 illustrates schematically an example of a method for identifying a main damage mode of the rolling bearing 4 of the machine 1 implementing the device 9.
[0070] The rolling bearing 4 is arranged in a plurality of machines of the same type and design as the machine 1 so that N rolling bearings 4 are analysed, N being the number of machines.
[0071] The machine 1 may be a pump, the rolling bearing 4 supporting the shaft 3 driving an impeller of the pump.
[0072] In a step 20, the collecting means 10 collects, for each machine out of N machines, a first operating parameter of the rolling bearing 4 representative of a duration between the installation of the rolling bearing 4 in said machine and the occurrence of a damage of the rolling bearing 4.
[0073] The collecting means 10 collects damage statistics of the rolling bearings 4 arranged in the N machines.202500044
[0074] 7
[0075] The first operating parameter comprises the time duration between the installation of the rolling bearing 4 in said machine and the occurrence of a damage of the rolling bearing 4.
[0076] In another embodiment, the first operating parameter comprises the cumulative number of rotations of the rolling bearing 4 between the installation of the rolling bearing 4 in said machine and the occurrence of a damage of the rolling bearing 4.
[0077] In the following, it is assumed that the first operating parameter comprises the time duration between the installation of the rolling bearing 4 in said machine and the occurrence of a damage of the rolling bearing 4.
[0078] It is assumed that the rolling bearings 4 are new when they are arranged in said machines.
[0079] The collecting means 10 may further collect a second operating parameter of the machines representative of the use of the machines.
[0080] The second operating parameter is for example representative of initiating conditions leading to a damage of the component which may be stand-still periods of the machines or an instant of relubrication of the rolling bearing.
[0081] The first operating parameter values and the second operating parameter values may be transmitted by the users of the machines.
[0082] In a step 21, the aggregating means 11 aggregates the first operating parameter values of the N machines to determine a real damage history of the rolling bearing 4. The real damage history associates N damages of the rolling bearing 4 to N damage durations.
[0083] In a step 22, the first determining means 12 determines statistical models representing the progression of predetermined damage modes of the rolling bearing 4 depending on the usage of the rolling bearing 4 from statistical models stored in the memory 16.
[0084] For example, the first determining means 12 determines a first statistical model 17 representing the progression of damages of a first predetermined damage mode depending on the usage of the rolling bearing 4, the first predetermined damage mode being associated to fatigue, a second statistical model 18 representing the progression of damages of a second predetermined damage mode depending on the202500044
[0085] 8
[0086] usage of the rolling bearing 4, the second predetermined damage mode being associated to aging of grease lubricating the rolling bearing 4, and a third statistical model 19 representing the progression of damages of a third predetermined damage mode depending on the usage of the rolling bearing 4, the third predetermined damage mode being associated to corrosion of the rolling bearing 4.
[0087] The first statistical model 17 comprises a rating life model for randomly initiated fatigue damage which are developed in the material of one of the two rings of the rolling bearing 4 or in the material of washers of the rolling bearing 4 or in the material of one of the rolling elements of the rolling bearing 4.
[0088] An example of the rating life model is the basic rating life model L10, defined in ISO 281, equal to:
[0089] ^
[0090]
[0091] io = (f)p(1)
[0092] where:
[0093] L10: the basic rating life at 90% probability in millions of revolutions
[0094] C: the dynamic load rating in kN
[0095] P: the equivalent dynamic bearing load in kN
[0096] p: a geometrical constant (3 for ball bearings, 10 / 3 for roller bearings).
[0097] The parameters C, P and p of the rating life model may be determined from the design data of the machine 1.
[0098] In variant, the rating life model may take into account more than one parameter representative of the conditions of use of the bearing. The rating life model may for example take into account one or more of the load distributions acting on the bearing, the centrifugal loads acting on the bearing, lubrication condition of the bearing, contamination of the bearing, clearance and preload of the bearing, misalignment and moment load, profiles and form deviations of the bearing.
[0099] In variant, the rating life model may comprise another model, for example comprise a stress-based Generalized Bearing Life Model (GBLM) disclosed in the document entitled “A Model for Rolling202500044
[0100] 9
[0101] Bearing Life with Surface and Subsurface Survival — Tribological Effects” G. E. Morales-Espejel, A Gabelli, A. J. C. de Vries, Tribology Transactions 58 :894-906, 2015.
[0102] The parameters of the first statistical model are the number N of rolling bearings 4, the start instant which is generally equal to zero corresponding to the installation of the N new bearings 4 in the N machines and the rating life for example at 10% probability of expected damages of the N rolling bearings 4 at a predetermined time duration between the start instant and for example the damage of 10% of the N rolling bearings 4 (L10).
[0103] Of course, another value of probability of damage may be used. Figure 4 illustrates schematically an example of a simulated damage history determined from the first statistical model wherein N is equal to 100 and L10 is equal to 500. In figures 4 to 6, the horizontal time axis is marked with t and the two vertical axes are marked with ff for the frequency of failures respectively frequency of damages and with cf for the cumulative failures respectively cumulative damages.
[0104] The continuous line represents the frequency of damage, and the dotted line represents the cumulative damages depending on the time duration.
[0105] The second statistical model 18 comprises a grease life model modelling the aging of grease lubricating the rolling bearing 4.
[0106] The grease life model comprises a uniform random distribution split in bins placed into a series of successively larger bins, where the bin size is given by predetermined aging constants to match rating life values for example rating values LI, L10 and L50 of the rolling bearing 4.
[0107] LI is the rating life at 1% probability of expected damages of the N rolling bearings 4 at a predetermined time duration between a start instant and the damage of 1% of the N rolling bearings 4, L10 is the rating life at 10% probability of expected damages of the N rolling bearings 4 at a predetermined time duration between the start instant and the damage of 10% of the N rolling bearings 4 and L50 is the rating life at 50% probability of expected damages of the N rolling bearings 4202500044
[0108] 10
[0109] at a predetermined time duration between a start instant and the damage of 50% of the N rolling bearings 4.
[0110] Time factors for the grease life model are Ll = l, L10=2.7 and L50=5.4 so that L10 is equal to 2.7L1 and L50 is equal to 5.4L1.
[0111] The bin sizes are then re-distributed over equally sized bins so that the probability of damage increases with the time duration.
[0112] The parameters of the first statistical model is the start instant of aging of the grease and a rating life value for example LI .
[0113] Figure 5 illustrates schematically an example of a simulated damage history with the first statistical model and the second statistical model superimposed.
[0114] For the first statistical model, L10 is equal to 500.
[0115] For the second statistical model, LI is equal to 370, L10 is equal to 1000 and L50 is equal to 2000.
[0116] The continuous line represents the frequency of damage, and the dotted line represents the cumulative damages depending on the time duration.
[0117] The third statistical model 19 comprises a corrosion model modelling the damage of corrosion in the rolling bearing 4.
[0118] A corrosion damage is associated with a propagation time before it becomes detectable, where the propagation time can be different for different types of corrosion damages and environments.
[0119] A normal distribution is used to model corrosion damages and is shifted with respect to the installation of the rolling bearing in the machine by a delay to model the propagation time.
[0120] The parameters of the third statistical model are the delay, the mean value of the normal distribution, the variance of the normal distribution and a number of damages.
[0121] The second operating parameter representative of the stand-still periods of the machines permits for example to determine the offset time duration.
[0122] Figure 6 illustrates schematically an example of a simulated damage history with the first statistical model, the second statistical model and the third statistical model superimposed.
[0123] For the first statistical model, L10 is equal to 500.202500044
[0124] 11
[0125] For the second statistical model, LI is equal to 370, LIO is equal to 1000 and L50 is equal to 2000.
[0126] For the third statistical model, the mean value of the normal distribution is equal to 1000, the variance of the normal distribution is equal to 100 and the number of damages is equal to 15 and the delay is equal to 200.
[0127] The continuous line represents the frequency of damage, and the dotted line represents the cumulative damages depending on the time duration.
[0128] Of course, the first determining means 12 may determine other statistical models to represent other predetermined damage modes, for example to represent damages caused by passages of electrical current in the rolling bearing 4 or ingress of third bodies in the rolling bearing 4 causing wear.
[0129] The statistical models stored in the memory 16 depend on the component object of said method.
[0130] In a step 23, the performing means 13 performs variation of parameters of the first statistical model, the second statistical model and the third statistical model so that a simulated damage history matches with the real damage history within a predetermined range.
[0131] The simulated damage history associates a plurality of damages to durations.
[0132] The simulated damage history is determined by the performing means 13 from the first statistical model, second statistical model and third statistical model defined in step 22.
[0133] The performing means 13 determines the simulated damage history from a set of parameter values of the first statistical model, the second statistical model and the third statistical model, and performs pattern recognition of the simulated damage history and real damage history to determine a similarity score.
[0134] If the similarity score is within the predetermined range, the simulated damage history matches with the real damage history,
[0135] If the similarity score is outside the predetermined range, the performing means 13 determines the simulated damage history from another set of parameter values of the first statistical model, the second202500044
[0136] 12
[0137] statistical model and the third statistical model and perform pattern recognition of the simulated damage history and real damage history to determine a similarity score until the similarity score is within the predetermined range.
[0138] The similarity score is for example equal to 70% of similarity between the simulated damage history and the real damage history.
[0139] The performing means 13 implements for example pattern recognition algorithms known from the prior art.
[0140] In a step 24, when the simulated damage history matches with the real damage history within the predetermined range, the second determining means 14 determines the contribution of each statistical model in the simulated damage history. The main damage mode of the rolling bearing 4 is the predetermined damage mode represented by the statistical model having the most significant contribution in the simulated damage history.
[0141] For each statistical model of the first statistical model, the second statistical model and the third statistical model, the second determining means 14 determines an auxiliary damage history from said statistical model, the auxiliary damage history associating a plurality of damages to durations and second determining means 14 performs pattern recognition of the auxiliary damage history and the simulated damage history to determine an auxiliary similarity score.
[0142] The contribution of each mode being equal to the auxiliary similarity score of said statistical model.
[0143] When the auxiliary similarity score is determined for each statistical model, the second determining means 14 determines the auxiliary similarity score having the biggest value.
[0144] The method builds and implements a statistical simulation model of the component as a “statistical digital twin”.
[0145] The most relevant damage mode of the component, for example the rolling bearing 4, may be identified from the digital twin and damage statistics collected on machines comprising the component without implementing known condition monitoring devices.
[0146] A series of root causes of damage can be excluded with higher certainty to determine why the component works and fails.202500044
[0147] 13
[0148] The statistical digital twin may further be implemented to simulate different scenarios and compare those to each other to defined countermeasures to enhance the performance of the component, for example the lifetime of the component arranged in the machine.
[0149] In the case of the rolling bearing 4, countermeasures may be to propose an alternative grease to prolong the grease lifetime, to modify the design of the rolling bearing 4, to modify the material of parts of the rolling bearing 4 to eliminate corrosion damages or to update the maintenance strategy of the rolling bearing 4 to reduce the occurrence of certain damages.
Claims
CLAIMS1. Method for identifying a main damage mode of a component (4), the component (4) being arranged in a plurality of same type of machines (1), the method comprising:(a) for each machine (1) of the plurality of same type of machines, collecting at least a first operating parameter of the component (4) representative of a duration between the installation of the component in the machine and the occurrence of a damage of the component,(b) aggregating the first operating parameter values of the plurality of machines (1) to determine a real damage history of the component, the real damage history associating each damage of said component to a damage duration,(c) determining at least a first statistical model (17) representing the progression of damages of a first predetermined damage mode depending on the usage of the component,(d) performing variation of at least one parameter of the first statistical model (17) so that a simulated damage history determined from the first statistical model matches with the real damage history within a predetermined range, the simulated damage history associating a plurality of damages to durations,(e) when the simulated damage history matches with the real damage history within the predetermined range, determining the contribution of each statistical model in the simulated damage history, the main damage mode of the component being the predetermined damage mode represented by the statistical model (17) having the most significant contribution in the simulated damage history.
2. Method according to claim 1, wherein step (c) comprises determining at least a second statistical model (18) representing the progression of a second predetermined damage mode and step (d) further comprises performing variation of at least one parameter of the second statistical model so that the simulated damage history determined fromthe first statistical model and the second statistical model match with the real damage history within the predetermined range.
3. Method according to claims 1 or 2, wherein step (d) comprises:(g) determining the simulated damage history from a set of parameter values of each statistical model (17, 18, 19),(h) performing pattern recognition of the simulated damage history and real damage history to determine a similarity score,if the similarity score is within the predetermined range, the simulated damage history matches with the real damage history,if the similarity score is outside the predetermined range, steps (g) and (h) are repeated with another set of parameter values of each statistical model.
4. Method according to any of claims 1 to 3, wherein step (e) comprises:- for each statistical model (17, 18, 19), determining an auxiliary damage history from said statistical model, the auxiliary damage history associating a plurality of damages to durations and performing pattern recognition of the auxiliary damage history and the simulated damage history to determine an auxiliary similarity score,- when the auxiliary similarity score is determined for each statistical model (17, 18, 19), determining the auxiliary similarity score having the biggest value, the contribution of each mode being equal to the auxiliary similarity score of said statistical model.
5. Method according to any of claims 1 to 4, wherein step (a) further comprises collecting at least a second operating parameter of the machines (1) representative of initiating conditions leading to a damage of the component, step (d) comprising defining range of variation of at least one parameter of at least one statistical model.
6. Method according to any of claims 1 to 5, wherein the component is a bearing (4).
7. Method according to claim 6, wherein the first operating parameter comprises the time duration or the cumulative number of rotations of the bearing.
168. Method according to claims 6 or 7, wherein the first statistical model (17) comprises a rating life model or a grease life model or a corrosion model.
9. Device (9) for identifying a main damage mode of a component (4), the component (4) being arranged in a plurality of same type of machines (1), the device comprising:- collecting means (10) configured to collect for each machine (1) of the plurality of same type of machines, at least a first operating parameter of the component representative of a duration between the installation of the component in the machine and the occurrence of a damage of the component,- aggregating means (11) configured to aggregate the first operating parameter values of the plurality of machines (1) to determine a real damage history of the component, the real damage history associating each damage of said component to a damage duration, - first determining means (12) configured to determine at least a first statistical model (17) representing the progression of damages of a first predetermined damage mode depending on the usage of the component,- performing means (13) configured to perform variation of at least one parameter of the first statistical model (17) so that a simulated damage history determined from the first statistical model matches with the real damage history within a predetermined range, the simulated damage history associating a plurality of damages to durations,- second determining means (14) configured to determine the contribution of each statistical model (17, 18, 19) in the simulated damage history when the simulated damage history matches with the real damage history within the predetermined range and to determine the main damage mode of the component which is the predetermined damage mode represented by the statistical model having the most significant contribution in the simulated damage history.
10. Device according to claim 9, wherein the component is a bearing (4).
