Method of training machine learning model to detect one or more faults

Abstract

The present disclosure describes a method of training a machine learning model to detect one or more faults associated with at least one component of a driveline. The method includes providing a simulation model associated with at least one component of the driveline, configuring the simulation model using a predefined configuration to generate training data for a machine learning model, generating training data using the configured simulation model, and training a machine learning model using the generated training data to detect health of at least one component of the driveline. The simulation model is configured to simulate a plurality of conditions based on a predefined configuration to generate training data comprising a plurality of data sets, where at least one data set of the plurality of data sets is associated with a corresponding fault of at least one component of the driveline.

Description

Technical Field

[0001] The present disclosure relates to condition monitoring and more particularly to drivetrain condition monitoring using machine learning models. Typically, in order to detect failures of engine and related drivetrain applications or loads, such as pumps or fans, machine learning models require training with complex measurements and considerable expertise. This requires a lot of time and work before deploying the trained model. Background Art

[0002] US2020 / 0103894 A1 discloses a deep learning system configured to train a computer vision system using a training data set to identify operating characteristics of mechanical equipment. Summary of the invention

[0003] The disclosure relates to monitoring of the condition of a transmission system. Machine learning models are often used to detect the condition of components of the drivetrain based on various sensor data associated with the drivetrain components. However, the use of machine learning methods requires a large amount of measurement data to train machine learning models.

[0004] This can be particularly challenging because accumulating data, especially for critical defects of the application, may not be feasible for example for safety reasons, or may be very expensive. Therefore, the transfer of such models is only possible for similar applications (i.e. of the same type and same size). In order to transfer the machine learning model to a different application and achieve the same accuracy, time-consuming retraining of the existing model would be required. Furthermore, in order to use the machine learning model in a specific configuration, data under reference conditions (a "good" healthy state with no faults) of the drive train is required. This is often difficult or impossible for applications operating in brownfields. Therefore, if the application encounters a fault when the machine learning model is initialized, the data may not be available. Therefore, there is a need for a method and apparatus that allows training of a machine learning model without the need for extensive collection of measurement data.

[0005] Thus, the present disclosure describes a method for training a machine learning model to detect one or more faults associated with at least one component of a transmission system. The method includes providing a simulation model associated with at least one component of the transmission system, the simulation model being used to simulate the operation of at least one component of the transmission system, configuring the simulation model using a predefined configuration to generate training data for the machine learning model, using the configured simulation model to generate training data, and using the generated training data to train the machine learning model to detect the health of at least one component of the transmission system. The trained machine learning model is configured to receive sensor data associated with at least one component of the transmission system and detect one or more faults associated with at least one component of the transmission system based on the received sensor data. The simulation model is configured to simulate multiple conditions based on the predefined configuration. The training data includes multiple data sets, wherein at least one of the multiple data sets is associated with a corresponding fault of at least one component of the transmission system.

[0006] Thus, the present disclosure discloses the use of synthetic labeled data from simulation models to train machine learning models. Thus, this allows machine learning models to be easily and quickly trained for condition monitoring and fault detection without the need to extensively collect measurement data associated with various conditions, faults, and configurations of a drive train. Thus, machine learning models can be quickly trained and deployed using simulation models associated with components of a drive train. In one example, a drive train includes an engine connected to a load using a gearbox.

[0007] In addition, the generated training data can cover a variety of fault conditions, such as misalignment, cavitation, imbalance, etc. The simulation model can be configured to simulate multiple critical faults associated with the application of interest. Different fault severities (tolerable, minor, severe, etc.) can be simulated under different operating conditions (speed, load, etc.). In addition, different driveline configurations can be simulated. For example, various foundations can be used to mount the driveline on different types of couplings. The simulation model subcomponents can be parameterized to cover different configurations and sizes.

[0008] In one embodiment, the predefined configuration includes a plurality of parameters associated with at least one component of the drivetrain, each parameter further including a plurality of values ​​associated with the corresponding parameter. Therefore, the simulation is configured to simulate a plurality of conditions specified based on the values ​​of the plurality of parameters. Thus, the generated training data covers a comprehensive set of scenarios associated with at least one component of the drivetrain.

[0009] In one embodiment, the plurality of data sets includes at least one additional data set, wherein the at least one additional data set is associated with normal operation of at least one component of the powertrain. Thus, the machine learning model is trained using data sets reflecting fault conditions and health conditions. In one embodiment, the sensor data includes vibration data associated with at least one component of the powertrain. Thus, the machine learning model is configured to receive vibration data from one or more vibration sensors and detect a condition of the at least one component based on the vibration data.

[0010] In one embodiment, the plurality of parameters include a first set of parameters and a second set of parameters, the first set of parameters is associated with one or more faults associated with at least one component of the same transmission system, and the second set of parameters is associated with the operation of at least one component of the transmission system. Therefore, the simulation model is configured to simulate multiple failure conditions in combination with multiple drivetrain configurations.

[0011] On the other hand, the present disclosure describes a computing device for training a machine learning model to detect one or more faults associated with at least one component of a transmission system. The computing device includes one or more processors connected to a memory module. The one or more processors are configured to configure a simulation model using a predefined configuration to generate training data for the machine learning model, wherein the simulation model is configured to simulate the operation of at least one component of the transmission system under multiple conditions based on the predefined configuration, use the configured simulation model to generate training data, wherein the training data includes multiple data sets, wherein at least one of the multiple data sets is associated with a corresponding fault of at least one component of the transmission system; and use the generated training data to train the machine learning model to detect the health of at least one component of the transmission system; wherein the trained machine learning model is configured to receive sensor data associated with at least one component of the transmission system, and detect one or more faults associated with at least one component of the transmission system based on the received sensor data. Similarly, the present disclosure describes a non-transitory storage medium containing a plurality of instructions that, when executed on one or more processors, cause the one or more processors to configure a simulation model using a predefined configuration to generate training data for a machine learning model, wherein the simulation model is configured to simulate operation of at least one component of a transmission system under a plurality of conditions based on the predefined configuration, use the configured simulation model to generate training data, wherein the training data includes a plurality of data sets, wherein at least one of the plurality of data sets is associated with a corresponding fault of at least one component of the transmission system; and use the generated training data to train a machine learning model to detect the health of at least one component of the transmission system; wherein the trained machine learning model is configured to receive sensor data associated with at least one component of the transmission system, and detect one or more faults associated with at least one component of the transmission system based on the received sensor data. The advantages of the method are applicable to devices and non-transitory storage media. These aspects will be combined with Figure 1-3 Further explanation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A portion of an industrial facility including a powertrain engine connected to an engine controller is shown;

[0013] Figure 2 A method of training a machine learning model to detect one or more faults associated with at least one component of a powertrain is shown; and

[0014] Figure 3 A computing device for training a machine learning model to detect one or more faults associated with at least one component of a powertrain is shown. DETAILED DESCRIPTION

[0015] Figure 1 A portion 100 of an industrial facility is shown that includes an engine 110 connected to a drive train of an engine controller 130. An industrial facility in the present invention refers to any environment in which one or more industrial processes such as manufacturing, refining, smelting, equipment assembly, etc. may be performed. Examples of industrial facilities include processing plants, refineries, automobile factories, etc. The drive train includes a mechanical load connected to the engine 110 using a gearbox. The mechanical load is associated with the process in the industrial facility and may include fans, pumps, etc.

[0016] The engine controller 130 is capable of monitoring and controlling the engine 110. Therefore, the engine controller 130 is connected to a plurality of sensors mounted on the engine 110. The plurality of sensors (e.g. Figure 1 The sensors 120 shown are used to measure multiple parameters associated with the engine 110. For example, the multiple sensors include magnetic field sensors, vibration sensors, acoustic sensors, temperature sensors, etc. The magnetic field sensor can measure the magnetic field strength along one or more axes of the engine 110. The vibration sensor can measure the vibration along one or more axes of the engine 110. The temperature sensor can measure the temperature of the area close to the engine 110. The acoustic sensor can measure the sound around the engine 110, especially the sound generated around the transmission side of the engine 110. Therefore, the engine controller 130 is configured to receive sensor data from the multiple sensors and determine the condition of the engine based on the sensor data.

[0017] Thus, in this regard, the engine controller 130 includes a trained machine learning model for determining the health or condition of the engine. The engine controller 130 is connected to the training module 150 to receive the trained machine learning model. The training module 150 may be located on an industrial cloud platform 170 outside of the industrial facility. The training module 150 is connected to the simulation model 160 to train the machine learning model 140, which is provided to the engine controller 130 after training. The machine learning model 140 is trained by the training module 150 using the simulation model 160 of the engine 110. This allows the machine learning model 140 to be trained without a large amount of real engine data, thereby making it easy and quick to deploy the machine learning model. This will refer to Figure 2 Further explanation.

[0018] Figure 2 A method 200 of training a machine learning model 140 to detect one or more failures associated with at least one component of the drivetrain is shown. In one example, method 200 is implemented by training module 150 . Although the method 200 may be applied to any component in the drivetrain, the method 200 is explained using the engine 110 .

[0019] In step 210 , a simulation model 160 associated with at least one component of the transmission system is provided. The simulation model 160 is used to simulate the operation of at least one component of the transmission system. The simulation model 160 is based on the configuration of the drivetrain in an industrial facility. Therefore, the simulation model 160 is based on the configuration of the connected load, the configuration of the engine, and the configuration of the gearbox.

[0020] Then, at step 220, the training module configures the simulation model using the predefined configuration to generate training data for the machine learning model. The simulation model is configured to simulate a plurality of conditions based on the predefined configuration. The predefined configuration includes a plurality of parameters associated with at least one component of the transmission system, each parameter further including a plurality of values ​​associated with the corresponding parameter. For example, the plurality of parameters includes a first set of parameters and a second set of parameters, the first set of parameters being associated with one or more faults associated with at least one component of the transmission system, and the second set of parameters being associated with operation of at least one component of the transmission system.

[0021] The first set of parameters may indicate whether a particular fault is to be simulated, and if so, the extent of the fault that may be simulated. For example, the first set of parameters may include faults such as misalignment, cavitation, impeller fault, etc. Each parameter in the first set of parameters may further include one or more values ​​representing the extent of the corresponding fault. For example, the misalignment parameter may include values ​​of 0, 1, 2, etc. A value of 0 indicates that there is no misalignment fault. Values ​​above 0 indicate the extent of the misalignment fault.

[0022] The second set of parameters may indicate operating conditions associated with at least one component. For example, the second set of parameters may include different parameters associated with a load connected to the engine, such as the weight of the load, the diameter of the load, the revolutions per minute associated with the load, the curve characteristics of the load, etc. In addition, each parameter includes one or more values. For example, the weight of the load may include 25 kg, 30 kg, etc.

[0023] Then, in step 230, the training module generates training data using the configured simulation model. The simulation model generates training data by simulating various physical parameters that can be read by one or more sensors. For example, the simulation model can simulate vibrations, electromagnetic fields / fluxes, acoustic signals, temperatures, etc. In particular, the simulation model simulates the above-mentioned physical parameters at positions where sensors can be installed on at least one component of the transmission system. The simulations performed using the simulation model are performed using a first set of parameters and a second set of parameters. The simulation model simulates the operation of at least one component under various conditions based on a combination of values ​​and parameters from the first set of parameters and the second set of parameters. For example, the simulation model generates vibration data for an engine with a misalignment fault with a 20kg load and a .1 value, an engine with a 30kg load but no misalignment fault, and other such combinations based on the first set of parameters and the second set of parameters.

[0024] The training data generated by the simulation model includes a plurality of data sets, wherein at least one of the plurality of data sets is associated with a corresponding fault of at least one component of the drive train. As mentioned above, the simulation model generates the training data by simulating the operation of the component of the drive train (particularly physical parameters such as vibration, electromagnetic flux, temperature, etc.) based on a combination of a first set of parameters and a second set of parameters. For each combination of the first set of parameters and the second set of parameters, a corresponding data set is generated via simulation. According to the first set of parameters, the corresponding data set is labeled with the fault present in the simulation. Therefore, the generated training data is labeled with the corresponding fault from the first set of parameters.

[0025] In one example, the plurality of data sets include at least another data set, wherein the at least another data set is associated with the normal operation of at least one component of the drivetrain. Therefore, the simulation model is also used to simulate the operation of at least one component without any failures. Therefore, all first set of parameters are simulated with a value of 0.

[0026] Then, at step 240, the training module uses the generated training data to train the machine learning model to detect the health of at least one component of the drivetrain. Different data sets with markers are used to train the machine learning model to detect failures in at least one component of the drivetrain.

[0027] Thus, the trained machine learning model is configured to receive sensor data associated with at least one component of the drivetrain and detect one or more failures associated with at least one component of the drivetrain based on the received sensor data.

[0028] It can be noted that although the above-mentioned method is explained with reference to the training module 150, the above-mentioned method may be implemented using one or more devices or via one or more software modules. Therefore, the present disclosure describes a computing device 300 in conjunction with Figure 3 The computing device is further described.

[0029] Figure 3 A computing device 300 for training a machine learning model 140 to detect one or more faults associated with at least one component of a drive train is disclosed. The computing device 300 includes one or more processors 320 connected to a memory module 330. The computing device 300 further includes a network interface 310 for communicating with industrial devices in an industrial facility. The memory module 330 (also referred to as a non-transitory storage medium 330) includes a plurality of instructions that, when executed by the one or more processors, cause the one or more processors 320 to configure a simulation model using a predefined configuration to generate training data for the machine learning model, generate training data using the configured simulation model, and train the machine learning model using the generated training data to detect the health of at least one component of a drive train.

[0030] For the purposes of this specification, a computer-usable or computer-readable non-transient storage medium may be any device capable of containing, storing, communicating, propagating or transmitting a program for use by or in conjunction with an instruction execution system, device or apparatus. The medium may be an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system (or device or apparatus), or a propagation medium that is not included in the definition of a physical computer-readable medium as a signal carrier itself, including semiconductor or solid-state memory, magnetic tape, removable computer floppy disk, random access memory (RAM), read-only memory (ROM), hard disk and optical disk, such as compact disk read-only memory (CD-ROM), compact disk read / write and DVD. As known to those skilled in the art, both the processing unit and the program code used to implement each aspect of the present technology may be centralized or distributed (or a combination thereof).

[0031] Although the present disclosure has been described with reference to a few industrial equipment, multiple industrial equipment may be utilized in the context of the present disclosure. Although the disclosure has been described in detail with reference to certain embodiments, it is to be understood that the disclosure is not limited to those embodiments.

[0032] In view of this disclosure, many modifications and variations will occur to those skilled in the art without departing from the scope of the various embodiments of the disclosure as described herein. Therefore, the scope of the disclosure is specified by the following claims rather than by the foregoing description. All changes, modifications and variations within the equivalent meaning and range of the claims are considered to be within their scope. All advantageous embodiments claimed in the method claims may also be applied to the device / non-transient storage medium claims.

Claims

1. A method of training a machine learning model to detect one or more faults associated with at least one component of a powertrain, the method comprising: a. providing a simulation model associated with at least one component of the transmission system, the simulation model being used to simulate the operation of at least one component of the transmission system; b. configuring the simulation model using a predefined configuration to generate training data for the machine learning model, wherein the simulation model is configured to simulate multiple conditions based on the predefined configuration, c. generating the training data using the configured simulation model, wherein the training data comprises a plurality of data sets, wherein at least one data set of the plurality of data sets is associated with a corresponding fault of the at least one component of the powertrain; d. using the generated training data to train the machine learning model to detect the health of at least one component of the powertrain; Wherein, the trained machine learning model is configured to receive sensor data associated with the at least one component of the transmission system, and detect the one or more faults associated with the at least one component of the transmission system based on the received sensor data.

2. The method according to claim 1, wherein: The predefined configuration includes a plurality of parameters associated with the at least one component of the powertrain, each parameter further including a plurality of values ​​associated with the corresponding parameter.

3. The method according to claim 1, wherein: The plurality of data sets includes at least one further data set, wherein the at least one further data set is associated with normal operation of the at least one component of the drive train.

4. The method according to claim 1, wherein: The sensor data includes vibration data associated with the at least one component of the powertrain.

5. The method according to claim 1, wherein: The drive train includes an engine connected to a load using a gearbox.

6. The method according to claim 2, wherein: The plurality of parameters includes a first set of parameters associated with one or more faults associated with the at least one component of the powertrain and a second set of parameters associated with operation of the at least one component of the powertrain.

7. A computing device for training a machine learning model to detect one or more faults associated with at least one component of a powertrain, the computing device comprising: a. One or more processors connected to the memory module, the one or more processors being configured to: i. configuring a simulation model using a predefined configuration to generate training data for the machine learning model, wherein the simulation model is configured to simulate operation of at least one component of the powertrain under a plurality of conditions based on the predefined configuration, ii. generating the training data using the configured simulation model, wherein the training data comprises a plurality of data sets, wherein at least one data set of the plurality of data sets is associated with a corresponding fault of the at least one component of the powertrain; iii. using the generated training data to train the machine learning model to detect the health of at least one component of the powertrain; Wherein, the trained machine learning model is configured to receive sensor data associated with the at least one component of the transmission system, and detect the one or more faults associated with the at least one component of the transmission system based on the received sensor data.

8. A non-transitory storage medium comprising a plurality of instructions which, when executed on one or more processors, cause the one or more processors to: a. Configure the simulation model using a predefined configuration to generate training data for the machine learning model, where: The simulation model is configured to simulate operation of at least one component of the powertrain under a plurality of conditions based on the predefined configuration, b. generating the training data using the configured simulation model, wherein the training data comprises a plurality of data sets, wherein at least one data set of the plurality of data sets is associated with a corresponding fault of the at least one component of the powertrain; c. using the generated training data to train the machine learning model to detect the health of at least one component of the powertrain; Wherein, the trained machine learning model is configured to receive sensor data associated with the at least one component of the transmission system, and detect the one or more faults associated with the at least one component of the transmission system based on the received sensor data.

Images