Search

EP-4738039-A1 - CONCEPT RELATING TO THE DETECTION OF A PRESENCE OR IMMANENCE OF A FAILURE OR DEGRADATION OF A MOTORIZED SYSTEM

EP4738039A1EP 4738039 A1EP4738039 A1EP 4738039A1EP-4738039-A1

Abstract

The training or optimization of a failure detector for detecting a presence or immanence of a failure or a degradation of a motorized system which comprises one or more closed-loop controlled electro motors may be achieved more efficiently by a more efficient generation of the dataset underlying this training, in that the more efficient generation of this dataset is achievable by providing a set of control parameters of a controller component arranged in one or more control loops of the one or more closed-loop controlled electro motors, with the settings representing different deviations from an in-use setting of the control parameters, and/or providing a set of set point signals of the one or more control loops of the one or more closed-loop controlled electro motors, the set of set point signals representing different variations of a base set point signal.

Inventors

  • GITTLER, Thomas
  • Jaquemet, Claude
  • BACIGALUPO, Carlos
  • LORENZ, Timo
  • JOSHI, Mihir Yogendra

Assignees

  • maxon international ag

Dates

Publication Date
20260506
Application Date
20241104

Claims (15)

  1. Method for generating a data set for training a failure detector (80) configured to detect a presence or immanence of a failure or degradation of a motorized system (120) comprising one or more closed-loop controlled electro motors (122), the method comprising providing (50) a set (52) of settings (54;56) of controller parameters (131) of a controller component (130) arranged in one or more control loops (124) of the one or more closed-loop controlled electro motors (122), the settings representing different deviations from an in-use setting (54) of the controller parameters (131), for each of the settings (56;(54,56)) of the controller parameters (131), measure (60) one or more signals (84) within the one or more control loops (124) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122) with the controller parameters (131) set according to the respective setting (56;(54,56)) so as to obtain a measured signal, and/or providing a set of setpoint signals (10) of the one or more control loops (12) of the one or more closed-loop controlled (124) electro motors (122), the set of setpoint signals representing different variations of a base setpoint signal, for each of the setpoint signals (10), measure one or more signals within the one or more control loops (124) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122) with applying the respective setpoint signal (10) onto the one or more control loops (124) so as to obtain a measured signal.
  2. Method according to claim 1, further comprising for each of the settings (56;(54,56)) of the controller parameters (131), attributing the deviation of the respective setting (56;(54,56)) from the in-use setting (54) of the controller parameters (131) to a target detector output value, wherein, for each of the settings (56;(54,56)) of the controller parameters, a combination of the measured signal obtained for the respective setting and the target detector output value attributed to the deviation of the respective setting (56;(54,56)) from the in-use setting (54) of the controller parameters (131) forms a data item of the data set.
  3. Method according to any of claims 1 to 2, further comprising subjecting (70) the data set (62) formed by the measured signals obtained for the settings (56;(54,56)) of the controller parameters (131) and the deviations of the settings (56;(54,56)) from the in-use setting (54) of the controller parameters (131) to an optimization scheme for optimizing a parametrizable mapping ought to map the one or more signals (84) within the one or more control loops (124) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122) with the controller parameters (131) set according to the in-use setting (54) onto a detector output value (86) informing on the presence or immanence of the failure or degradation of the motorized system(120).
  4. Method according to any of claims 1 to 3, wherein the one or more signals comprise one or more of a) one or more feedback signals (16) within the one or more control loops (124), b) one or more output signals (14) output by the one or more control loops (124), c) one or more following error signals within the one or more control loops (124), d) one or more signals which depend on one or more of a), b) and c).
  5. Method according to any of claims 1 to 4, wherein the controller parameters comprise weights for weighting one or more of A) output signals (14) of the one or more control loops (124), B) following error signals within the one or more control loops (124), C) affine combinations thereof, D) a transformed version of one or more of A, B, and C, such as an integral and a derivative, and E) one or more signals which depend on one or more of A) to D).
  6. Method according to any of claims 1 to 5, wherein the settings (56;(54,56)) deviate from the in-use setting (54) with respect to weights in a manner resulting in, compared to the in-use setting (54), one or more of different differential gain, different integral gain, different proportional gain.
  7. Method according to any of claims 1 to 6, wherein the parametrizable mapping is a multivariate function, a machine learning model.
  8. Method according to any of claims 1 to 7, wherein the motorized system (120) is a self-driving vehicle, a robot, or a machine having at least one movable part.
  9. Method according to any of claims 1 to 8, wherein, for each of the settings (56;(54,56)) of the controller parameters (131), the measuring (60) the one or more signals (84) within the one or more control loops (124) or influenced by the one or more control loops (124) during the operation of the one or more closed-loop controlled electro motors (122) with the controller parameters (131) set according to the respective setting (56;(54,56)) is performed by operating the one or more closed-loop controlled electro motors (122) using a setpoint signal (10) which is equal for all settings or results, at equal measuring conditions in terms of measurement environment and degree of wear of the motorized system, in less than 10% variation of the measured signals.
  10. Failure detector for detecting a presence or immanence of a failure or degradation of a motorized system (120) comprising one or more closed-loop controlled electro motors (122), configured to obtain, from one or more signals (84) within one or more control loops (124) of the one or more closed-loop controlled electro motors (122) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122), a measured signal, and map, using a mapping, the measured signal onto a detector output value (86) informing on the presence or immanence of the failure or degradation of the motorized system (120), the mapping being parametrized by use of a data set (62) gained by the method according to any of claims 1 to 9.
  11. Failure detector according to claim 10, wherein the measuring the one or more signals within one or more control loops of the one or more closed-loop controlled electro motors (122) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122) is performed with the controller parameters (131) set according to the in-use setting (54), and/or applying the base setpoint signal onto the one or more control loops (122).
  12. Apparatus for failure surveillance of a motorized system (120) comprising one or more closed-loop controlled electro motors (122), the apparatus being configured to interact with the motorized system (120), comprising a failure detector (80) for detecting a presence or immanence of a failure or degradation of the motorized system (120), and comprising an input/output interface (102), the apparatus being switchable between several operation modes including a data set generation mode, at least one effective operation mode and a failure detection mode, which may be one of the at least one effective operation mode or separate therefrom, the apparatus being configured to operate according to a) and/or b), namely, in the data set generation mode, a) receive a set (52) of settings (56;(54.56)) of controller parameters (131) of a controller component (130) arranged in one or more control loops (124) of the one or more closed-loop controlled electro motors (122), via the input/output interface (102), for each setting of the set (52) of settings (56;(54, 56)), activate the one or more closed-loop controlled electro motors (122) to operate with the controller parameters (131) set according to the respective setting, measure one or more signals (84) within the one or more control loops (124) or influenced by the one or more control loops (124) during the one or more closed-loop controlled electro motors (122) operating with the controller parameters (131) set according to the respective setting (56;(54, 56)) so as to obtain a measured signal, output the measured signal via the input/output interface (102), and/or b) receive a set of setpoint signals of the one or more control loops (124) of the one or more closed-loop controlled electro motors (122), the set of setpoint signals representing different variations of a base setpoint signal, for each setting of the set of settings, activate the one or more closed-loop controlled electro motors (122) to operate with the controller parameters (131) set according to the respective setting, measure one or more signals within the one or more control loops (124) or influenced by the one or more control loops (124) during the one or more closed-loop controlled electro motors (122) operating with the controller parameters (131) set according to the respective setting so as to obtain a measured signal, output the measured signal via the input/output interface (102), in at least one of the at least one effective operation mode, activate the one or more closed-loop controlled electro motors (122) to operate with a) the controller parameters (131) set according to the in-use setting (54), in the failure detection mode, activate the one or more closed-loop controlled electro motors (122) to operate with a) the controller parameters (131) set according to the in-use setting (54) and/or the base setpoint signal being applied onto the one or more control loops (124), measure the one or more signals (84) within the one or more control loops (124) or influenced by the one or more control loops (124) during the one or more closed-loop controlled electro motors (122) operating with a) the controller parameters (131) set according to the in-use setting (54) and/or b) the base setpoint signal being applied onto the one or more control loops (124) so as to obtain a further measured signal, map, using a parametrizable mapping, the further measured signal onto a detector output value (86) informing on the presence or immanence of the failure or degradation of the motorized system (120), the parametrizable mapping being parametrized via the input/output interface (102).
  13. Apparatus according to claim 12, configured to perform the activation of the one or more closed-loop controlled electro motors (122) in the data set generation mode, and the failure detection mode, using a setpoint signal which is equal among the data set generation mode and the failure detection mode or results, at equal measuring conditions in terms of measurement environment and degree of wear of the motorized system (120), in less than 10% variation of the measured signals.
  14. Apparatus according to claim 12 or 13, configured to present the detector output value to a user, or inform the user when the detector output value (86) fulfills a predetermined criterion which is indicative of a situation wherein the presence or immanence of the failure or degradation of the motorized system (120) is likely to occur.
  15. Method for detecting a presence or immanence of a failure or degradation of a motorized system (120) comprising one or more closed-loop controlled electro motors (122), the method comprising measuring one or more signals (84) within one or more control loops (124) of the one or more closed-loop controlled electro motors (122) or influenced by the one or more control loops (124) during an operation of the one or more closed-loop controlled electro motors (122) so as to obtain a measured signal, and mapping, using a parametrizable mapping, the measured signal onto a detector output value (86) informing on the presence or immanence of the failure or degradation of the motorized system (120), the parametrizable mapping being determined by use of a data set (62) gained by the method according to any of claims 1 to 9.

Description

The present application relates to the detection of a presence or immanence of a failure or degradation of a motorized system such as a robot or the like. In the context of prognostics and health management (PHM) of mechatronic systems, condition monitoring approaches strive to provide early (or earlier) indications of future or imminent failures. Obtaining run-to-failure, fault presence, or anomalous datasets for such approaches is a significant challenge. Due to the rarity of failures in many systems, especially those for high precision and safety-critical applications, the datasets containing faults, failures or relevant anomalies are scarce. Imitating real anomalies or faults can be costly as it often requires destructive testing or extensive engineering work. Additionally, creating run-to-failure datasets is time-consuming and therefore another cost factor besides engineering and material. Amortization of these efforts is only possible if the corresponding application is produced in large quantities. Even if we assumed that possible amortization across many produced units of a system, for which run-to-failure samples were collected and the corresponding failure mode recorded - there still remains the issue of the large variety of failure modes. Approximately 80% of occurring failures have not been recorded in this specific behavior before, making it virtually impossible to collect a broad spectrum of real application failure data. Additionally, a remaining variable is the so-called exemplary variance. Even in mass-produced systems, each unit is subject to small variations in behavior, potentially hindering the transfer of failure modes and models trained on other units. This scarcity hampers the development of robust data-driven prognostics models, which rely on extensive and representative data sets to predict failures or the remaining useful life (RUL) of components. Generally speaking, the effort required to obtain datasets incorporating fault, anomaly or failure cases hinders the development, the amortization and the affordability of PHM approaches. To address this issue, the current state of the art employs several strategies. One prominent approach is the use of synthetic data generation, such as the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) model developed by NASA, which simulates realistic run-to-failure trajectories under various operational conditions. This approach has required extensive simulation and engineering of a generic jet engine, which cannot be applied directly to any specific jet engine model. Generally speaking, simulation data of mechatronic systems are often used to simulate failures or anomalous behavior. However, this does not apply to systems produced in small quantities or one-off, and it does not consider exemplary variance. Additionally, transfer learning techniques are increasingly utilized, allowing models trained on available datasets to be adapted to new, similar systems with limited data. This means that a model is trained on application X and then used on application Y, or the model is trained on unit 1 of application X, and then applied to unit 2 (which is identical to unit 1) of application X. This is often the case if large engineering efforts are undertaken to model a specific application that is produced in large quantities, and where the engineering and (often physical) modeling effort can be subject to an amortization across the large fleet of produced units. Also, in many cases models are trained on failures recorded on a fleet of units, e.g. in some automotive components. Another innovative method involves the integration of domain knowledge into machine learning models to enhance their predictive capabilities despite the paucity of failure data. These approaches can vary, and are usually simulation based, or depend on the setting of specific thresholds e.g. for vibrations in specific frequency ranges. They usually require very specific failure modes and extensive, deep domain knowledge. Thus, there is a need for a concept which allows the detection of the presence or immanence of failure or degradation of a motorized system in a more efficient manner so that an amortization of the detection is easier to achieve. Accordingly, it is the object of the present invention to provide concepts which allow for more efficient detection of a presence or immanence of a failure or degradation of a motorized system. This object is achieved by the subject matter of the independent claims. The present invention is based on the finding that a training or optimization of a failure detector for detecting a presence or immanence of a failure or a degradation of a motorized system which comprises one or more closed-loop controlled electro motors may be achieved more efficiently by a more efficient generation of the dataset underlying this training, and that the more efficient generation of this dataset is achievable by providing a set of control parameters