Training machine learning models for fault classification based on synthetic data

Abstract

The present disclosure relates to training a machine learning model for fault classification based on synthetic data, and discloses a method comprising: generating synthetic tagged training data based on a tagged sensor data set associated with one or more industrial automation devices; and training one or more machine learning models for fault classification associated with the one or more industrial automation devices based at least on the synthetic tagged training data.

Description

Technical Field

[0001] The following example embodiments relate to industrial automation and artificial intelligence. Background Art

[0002] In supervised learning, the accuracy of a machine learning model depends on the amount of labeled training data used to train the machine learning model. If the amount of available labeled training data is small, then there are challenges in how the machine learning model can be trained. Summary of the invention

[0003] The independent claims define the scope of protection sought by various example embodiments.Example embodiments and features described in this specification that do not fall within the scope of the independent claims, if any, are to be construed as examples for understanding the various embodiments.

[0004] According to one aspect, a device is provided, comprising at least one processor and at least one memory, wherein the at least one memory stores instructions which, when executed by the at least one processor, cause the device to at least: generate synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and train one or more machine learning models for fault classification associated with the one or more industrial automation devices based at least on the synthetic labeled training data.

[0005] According to another aspect, an apparatus is provided, comprising a component for generating synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and a component for training one or more machine learning models for fault classification associated with one or more industrial automation devices based at least on the synthetic labeled training data.

[0006] According to another aspect, a method is provided, comprising: generating synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and training one or more machine learning models for fault classification associated with the one or more industrial automation devices based at least on the synthetic labeled training data.

[0007] According to another aspect, a computer program is provided, comprising instructions that, when executed by an apparatus, cause the apparatus to perform at least the following: generate synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and train one or more machine learning models for fault classification associated with one or more industrial automation devices based at least on the synthetic labeled training data.

[0008] According to another aspect, a computer-readable medium is provided, comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: generate synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and train one or more machine learning models for fault classification associated with one or more industrial automation devices based at least on the synthetic labeled training data.

[0009] According to another aspect, a non-transitory computer-readable medium is provided, comprising program instructions which, when executed by an apparatus, cause the apparatus to perform at least the following: generate synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and train one or more machine learning models for fault classification associated with one or more industrial automation devices based at least on the synthetic labeled training data. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Various example embodiments are described in more detail below with reference to the accompanying drawings, in which

[0011] Figure 1 illustrates a communication system to which some example embodiments may be applied;

[0012] Figure 2 illustrates a schematic block diagram according to an example embodiment;

[0013] Figure 3 illustrates a flow chart according to an example embodiment;

[0014] Figure 4 illustrates a flow chart according to an example embodiment; and

[0015] Figure 5 An example of an apparatus is illustrated. DETAILED DESCRIPTION

[0016] The following embodiments are illustrative. Although the specification may refer to "one", "an" or "some" embodiments in several places in the text, this does not necessarily mean that each reference refers to the same embodiment or that a particular feature applies only to a single embodiment. Individual features of different embodiments may also be combined to provide other embodiments.

[0017] Different embodiments and examples are described below using a single unit, model, equipment and memory, without limiting the embodiments and examples to such solutions. Concepts known as cloud computing and / or virtualization can be used. Virtualization can allow a single physical computing device to host one or more instances of a virtual machine that appears and operates as an independent computing device, so that a single physical computing device can dynamically create, maintain, delete or otherwise manage the virtual machine. Device operations may also be distributed between multiple servers, nodes, devices or hosts. In cloud computing network devices, computing devices and / or storage devices provide shared resources. Some other technological advances (such as software defined networks (SDN)) may cause one or more of the functions described below to migrate to any corresponding abstraction or device or equipment. Therefore, all vocabulary and expressions should be interpreted in a broad form, and they are intended to illustrate rather than constrain example embodiments.

[0018] Figure 1 The present invention illustrates a communication system in which some example embodiments may be applied. Figure 1 , some example embodiments may be based on wireless communications such as 3G (third generation), 4G (fourth generation), LTE (Long Term Evolution), LTE-A (Advanced Long Term Evolution), 5G (fifth generation), 5G NR (New Radio), 6G (sixth generation), UMTS (Universal Mobile Telecommunications System), EDGE (Enhanced Data Rates for GSM Evolution), WCDMA (Wideband Code Division Multiple Access), Bluetooth, WLAN (Wireless Local Area Network), Wi-Fi, Li-Fi (Light Fidelity), Ethernet or any other mobile network or wireless network or wired network. Communication may also occur between nodes belonging to different but compatible systems (such as LTE and 5G).

[0019] It should be pointed out that Figure 1 A simplified system architecture is illustrated which shows only some elements and functional entities, all of which are logical units whose implementation may differ from what is shown. Figure 1 The connections shown are logical connections; the actual physical connections may be different. Data collection may use a so-called master protocol, where a master network node subscribes to data from a slave device (the device whose data it wants to have), and the slave device or network node automatically sends its data to the receiver / master device based on a query or based on a subscription. It will be apparent to those skilled in the art that the system may also include other functions and structures. It should be appreciated that the functions, structures, elements, and protocols used in or for communication are not relevant to the example embodiments. Therefore, they need not be discussed in more detail here.

[0020] refer to Figure 1The system may include at least one variable speed drive 101 , at least one motor 102 , one or more sensor devices 103 , a gateway device 106 , a cloud platform 104 , and a user device 105 .

[0021] Variable speed drive 101 can also be referred to as variable frequency drive. Variable speed drive 101 can be used to run or rotate motor 102 at different speeds, wherein motor 102 can run pump or any other rotating equipment (e.g., fan, gearbox, belt drive, etc.). Variable speed drive 101 can be electrically connected to motor 102. Variable speed drive 101 can include or be connected to a controller, e.g., a proportional integral derivative (PID) controller. The controller can be configured to send a control signal to variable speed drive 101. Variable speed drive 101 can control highly dynamic industrial processes, wherein, for example, the speed or torque applied to motor 102 must change according to the needs of the industrial process.

[0022] The one or more sensor devices 103 can be configured to measure the value of one or more variables associated with the motor 102 to obtain motor sensor data. For example, the one or more sensor devices 103 can measure the value of at least one of the following: the speed, torque, vibration, temperature, magnetic flux density and / or current of the motor 102 over time. There may be separate sensor devices for measuring each variable (e.g., separate sensors for speed and temperature, respectively), or the variables can be measured by the same sensor device. At least one of the one or more sensor devices 103 can measure the value of environmental conditions over time, such as the external temperature and / or humidity outside the motor 102. The one or more sensor devices 103 can be constituted by the motor 102, or attached to the motor 102, or very close to the motor 102.

[0023] The one or more sensor devices 103 may be configured to communicate with the cloud platform 104 via a gateway device 106. The gateway device 106 may include, for example, an edge gateway or a user device (which may be the same as the user device 105 or may be different from the user device 105). For example, a mobile phone (or any other user device) may be used as a gateway device 106 to read measurements from one or more sensor devices 103 and push these measurements to the cloud platform 104 via an Internet connection. The user device used as a gateway device 106 may also be used to view historical measurement information from one or more sensor devices 103 (stored at the cloud platform 104) via an Internet connection.

[0024] In order to achieve the connection between the one or more sensor devices 103 and the gateway device 106, the one or more sensor devices 103 and the gateway device 106 can be equipped with a short-range communication interface, such as Bluetooth, Ethernet, ZigBee, Li-Fi, Wi-Fi, wireless mesh network, NFC or any other wireless connection or wired connection. The one or more sensor devices 103 can also be configured to communicate with the motor 102 and / or the variable speed drive 101 via the short-range communication interface. For example, the one or more sensor devices 103 can be configured to transmit the measured motor sensor data to the gateway device 106 and / or the motor 102 or the variable speed drive 101 via the short-range communication interface.

[0025] The gateway device 106 may be connected to the Internet via a network interface, such as 3G, 4G, LTE, LTE-A, 5G, 5G NR, 6G, UMTS, EDGE, WCDMA, WLAN, Wi-Fi, Li-Fi, Ethernet, or any other mobile network, wireless network, or wired network. The gateway device 106 may be connected to the cloud platform 104 via a network interface. One or more sensor devices 103 may be connected to the cloud platform 104 via the gateway device 106. The one or more sensor devices 103 may be configured to transmit the measured motor sensor data to the cloud platform 104 via the gateway device 106.

[0026] Cloud platform 104 (also referred to as cloud computing platform) refers to a fully integrated collection of cloud-based services and resources that enable users to build, deploy, manage and scale various applications and services on the Internet. It provides a flexible and scalable infrastructure for hosting applications, storing data and performing computing tasks without the need for local hardware and infrastructure.

[0027] The cloud platform 104 may be configured to receive motor sensor data from one or more sensor devices 103 via the gateway device 106. The cloud platform 104 may also be configured to store the received motor sensor data in at least one memory or database.

[0028] The cloud platform 104 may also be connected to the user device 105 to enable a user of the user device 105 to monitor the condition of the motor 102 via a user interface.

[0029] The user device 105 may include user equipment such as a smartphone, mobile phone, tablet computer, laptop computer, desktop computer, or any other computing device. The user device 105 may be a remote device located at a different location from the motor 102. Alternatively, the user device 105 may be a local device located on-site near the motor 102.

[0030] The user device 105 can be connected to the Internet via a network interface, such as 3G, 4G, LTE, LTE-A, 5G, 5G NR, 6G, UMTS, EDGE, WCDMA, WLAN, Wi-Fi, Li-Fi, Ethernet or any other mobile network, wireless network or wired network. The user device 105 can be configured to exchange information with the cloud platform 104 via the network interface, that is, transmit and / or receive data.

[0031] Industrial automation equipment, such as motors and variable speed drives, can be associated with analytical solutions to provide condition monitoring tools to users using sensor data. For example, it may be desirable to provide a fault classification tool for motors that can detect faults associated with the motor and predict actual problems. This may help minimize downtime associated with the motor.

[0032] In order to develop a fault classifier for electric motors, a large amount of labeled training data for various machines and faults may be required. In this article, the label refers to the health of the motor. Some examples of such labels may include (but are not limited to): "healthy", "broken rotor bars", "motor eccentricity", "bearing inner race fault", "shaft misalignment", or "rotor unbalance". For example, labels for different motors and faults can be collected in a laboratory environment. However, collecting these labels may require a lot of manual work, making the label collection process impractical and time consuming.

[0033] To address the above-mentioned issues regarding the lack of labeled training data, some example embodiments may utilize generative artificial intelligence (AI) to synthesize new training data under various conditions by utilizing a small amount of initial labeled data. In other words, some example embodiments require significantly less initial labeled real data to learn to synthesize new data under various conditions. The initial labeled data can be obtained from various sources such as laboratory measurements, simulated data, and / or synthetic data obtained using domain knowledge. In this way, a generative AI model can be trained to learn the characteristics of a signal under various conditions. These conditions may include, for example, different fault conditions, machine sizes, loading points, load types, and speeds.

[0034] The trained generative AI model can then be used to generate synthetic labeled training data. Synthetic data refers to artificially generated data rather than data generated by real-world events. The synthetic training data can be used to train a machine learning model to obtain an accurate fault classifier. Therefore, some example embodiments enable the development of accurate fault classifiers while avoiding the time-consuming and impractical manual label collection process. In other words, some example embodiments allow the development of a fault classifier for an electric motor without the need for a large amount of manually labeled data.

[0035] Generative AI is a subfield of artificial intelligence that focuses on creating models and systems that can generate new data that is similar to the input data they were trained on. These models are able to generate new samples that exhibit similar characteristics to the data they learned from, allowing them to produce entirely new data points that were not part of the original dataset. The main goal of generative AI is to generate artificial data that is indistinguishable from real data.

[0036] Variational autoencoders and generative adversarial networks are some examples of techniques that can be used for generative AI.

[0037] A variational autoencoder is a type of artificial neural network-based generative model that learns to encode and decode data. It consists of an encoder network that maps the input data to a latent space where it is represented by a probability distribution. The decoder network then draws samples from this distribution and reconstructs the original data. A variational autoencoder is used to generate new samples from the learned data distribution.

[0038] Generative adversarial networks are another type of generative model based on artificial neural networks, which uses a generator network to create fake samples and a discriminator network to distinguish between real and fake samples. The generator and discriminator are trained in an adversarial process, where the generator tries to produce samples that are indistinguishable from real data and the discriminator tries to better distinguish between real and fake. This back-and-forth training leads to a generator that can produce highly realistic data samples.

[0039] The example embodiments described below may provide various advantages. One such advantage is the development of a fault classifier that can indicate when a motor actually needs maintenance. This may reduce the downtime of the motor. Another advantage is that some example embodiments reduce the burden on experts to manually examine each anomaly to determine the type of fault. Another advantage is that some example embodiments reduce the burden on data scientists to manually train fault classification models.

[0040] Figure 2 A schematic block diagram according to an example embodiment is illustrated. Figure 2 In the example, generative artificial intelligence (AI) 210, fine-tuning 220, and fault classifier training 230 can be implemented together to train a fault classifier 240 in an automatic manner. For example, training of the fault classifier 240 (and the generative AI model 211) can be performed in the cloud platform 104 or in any computing device with sufficient computing resources for training.

[0041] refer to Figure 2, generative AI 210 is responsible for developing generative AI model 211. Available labeled data from various sources 202, 203, 204 can be used to develop (train) generative AI model 211. The labeled data sources may include a synthesizer database (DB) 202, which includes synthetic data generated by a domain knowledge-based data synthesizer 201; a real data database 203, which includes real measured sensor data 202 (time series data) of one or more motors 102; and / or a simulated data database 204, which includes simulated data of one or more motors 102. The simulated data may be based on a digital twin of one or more motors 102.

[0042] The synthesizer 201 is a statistical model based on domain knowledge. In a given spectrum, it is known which frequencies exist under various conditions of the motor. The amplitudes of these frequency components are not easily known in advance. For a faulty motor, the amplitudes of the fault-related frequencies can be statistically determined by analyzing real available data from laboratory experiments. For healthy motors, the amplitudes of the frequencies can be determined from available (anonymous) customer data. However, these models alone cannot add sufficiently realistic noise to simulate real data with acceptable errors. Furthermore, these models will not add frequencies that are not directly related to the motor, which means that the equipment that the motor is driving (e.g., pumps, fans, gearboxes, belt drives, etc.) will produce vibrations.

[0043] If there is a sufficient amount of labeled real data with faults (in the real data database 203), the generative AI model 211 can be trained based solely on this real data to generate new training data, which is the best scenario. In other words, in this best scenario, the synthesizer database 202 and the simulation database 204 may not be needed. However, in the absence of sufficient labeled real data for fault conditions, the synthetic data 202 and / or simulated data 204 generated by the domain knowledge-based data synthesizer 201 can be utilized. This can be achieved by collecting synthesized data of available labeled real data. In other words, by using the domain knowledge-based synthesizer 201, any available real measured data can be synthesized. In addition, real data of newly connected (or recently connected) motors can be measured, which are considered healthy, in which case data for "healthy" conditions of these motors can be synthesized.

[0044] Therefore, by using a synthesizer 201 based on domain knowledge and / or simulation, a data set including real data 203 with some labels and its synthesized version can be obtained. In this case, the generative AI model 211 can be trained to learn how to convert the synthesized noise-free data into real data under various operating conditions. After achieving this, the generative AI model 211 can be used to convert any synthesized data into real data, which can be used to train one or more fault classifier models 240. In this scenario, the generative AI can also be replaced by any other technology that can convert the synthesized noise-free data into real data under various operating conditions.

[0045] The purpose of generative AI is to generate one or more signals, such as at least one of the following: speed, temperature, vibration, magnetic flux, and / or current. These signals are affected by various conditions of one or more motors 102, such as load, motor rating, and health. Therefore, these signals are the conditions (or inputs) that the generative AI model 211 uses to generate new data.

[0046] As a non-limiting example, the generative AI model 211 can be used to generate the following data (the numbers in this example are arbitrary and more conditions can be added to this): Load type: pump; Load amplitude: 10kW; Speed: 1000rpm; Motor rated power: 15kW; Motor rated speed: 1500rpm; Health status: "Bearing outer race fault stage 3".

[0047] As for the model architecture of the generative AI model 211, various approaches can be used. For example, the generative AI model 211 may include a variational autoencoder or a generative adversarial network. The generative artificial intelligence model 211 can be trained on as much data as possible, and the generative ability of the model 211 can be evaluated by using appropriate metrics and / or domain experts.

[0048] The load of a motor refers to the mechanical resistance or work that the motor needs to perform or overcome while operating. In other words, it is the force or torque applied to the motor shaft that resists the motor's rotation. When the motor is connected to a mechanical system, it needs to provide the necessary torque to move or rotate the load. The load can vary depending on the application of the motor. For example, in an industrial application, the load may be a conveyor that needs to transport heavy materials, a pump that needs to lift water, or a fan that needs to circulate air. The performance and behavior of the motor may vary depending on the load it is operating under.

[0049] The rating of a motor includes information such as rated power, rated voltage, rated current, rated speed, and shaft height. It indicates the design specifications and capabilities of the motor. The rating information may help understand the expected performance and operating range of the motor.

[0050] Fine-tuning 220 involves updating the model weights of the generative AI model 211 to obtain new data whenever new data is received. The data source 206 of the new data can include, for example, new sensor data from a new type of motor 205 and / or raw data from motor sensors in the field with confirmed health conditions (e.g., "healthy" or certain fault conditions). For example, fine-tuning 220 can be performed by applying transfer learning.

[0051] Raw data refers to measured time series waveforms (signals) from one or more sensors 103. For example, if a new type of motor is connected to the system, it is likely to be a "healthy" motor. Therefore, the health status of this new type of motor can be labeled as "healthy", and thus, the measured data from the new type of motor can be labeled. This labeled data can then be supplied to the generative AI model 211 along with the defined health status (e.g., "healthy"). As explained above, the generative AI model 211 can be trained in different ways (e.g., to learn to make simulated data realistic or directly learn to generate data only under certain conditions).

[0052] Transfer learning is a machine learning technique in which a pre-trained model developed for one task or domain is reused as a starting point for training a new model on a different but related task or domain. Instead of training a model from scratch, transfer learning leverages the knowledge and learned representations from a pre-trained model to improve the learning process and performance on a new task.

[0053] After fine-tuning 220, the generative AI model is then used to synthesize large amounts of labeled training data 207 for different machine sizes, ratings, and under different fault and operating conditions.

[0054] For example, in the synthesized training data 207, in addition to static data such as load type and / or motor rating, the input can also include unprocessed time series waveforms of vibration, magnetic flux, temperature, speed and / or load, and the labels refer to health conditions (e.g., "healthy", "bearing inner ring failure", "bearing outer ring failure", and "rotor bar fracture", etc.). For example, the training data can be viewed as a table where the columns include vibration, magnetic flux, temperature, speed, load, load type, motor rated power, motor rated speed, shaft height, and label (health condition). Then, the rows in the table include time steps (some values ​​remain unchanged, such as motor rating).

[0055] During fault classifier training 230, the synthesized training data 207 is used to train one or more supervised fault classifier models 240 (i.e., machine learning models). The purpose of the one or more fault classifier models 240 is to determine the health of one or more motors 102 based on input signals (e.g., vibration, magnetic flux, temperature, motor rating, load type, load magnitude, and / or speed). In other words, the goal of the one or more fault classifier models 240 is to identify a health condition based on the input signals.

[0056] For example, the fault classifier model may be trained by using one of the following algorithms: Extreme Gradient Boosting (XGBoost), Random Forest, or Support Vector Machine, to which the synthetic training data 207 may be fed.

[0057] XGBoost can handle tabular data well. XGBoost is an implementation of gradient boosting, which is a boosting technique that builds multiple weak learners (e.g., decision trees) in sequence and combines their predictions to create a more powerful and accurate prediction model. In other words, XGBoost can build a powerful model by iteratively adding weak learners and adjusting their weights based on the errors of previous learners. XGBoost may combine regularization techniques such as L1 (Lasso) and L2 (Ridge) regularization to prevent overfitting, which penalizes complex models. XGBoost also has a built-in mechanism for handling missing values ​​in the dataset during training and prediction. XGBoost is designed to be computationally efficient and can take advantage of parallel processing, making it suitable for large-scale datasets. XGBoost supports a variety of loss functions depending on the type of problem, such as mean squared error for regression tasks and logarithmic loss (cross entropy) for binary classification. XGBoost provides feature importance scores that help understand the relative importance of different features in prediction.

[0058] The one or more trained fault classifier models 240 may then be deployed, for example, in the cloud platform 104 to predict the health status of at least one of the one or more motors 102 using its sensor data (ie, real data).

[0059] The health condition indicates whether one or more motors 102 are healthy or in a fault condition. A fault condition indicates a specific fault or anomaly that has occurred in the motor. For example, a fault condition may include at least one of the following: bearing failure, rotor imbalance, phase imbalance, overheating, etc. Each fault condition represents a unique problem that needs to be identified and resolved.

[0060] It should be noted that some example embodiments may also be extended to other areas, such as drive condition monitoring (in addition to motor condition monitoring).

[0061] Figure 3 A flow chart according to an example embodiment of a method performed by one or more cloud services of a cloud platform 104 or an apparatus 500 or any other computing device including or consisting of a cloud platform 104 is illustrated.

[0062] refer to Figure 3 In block 301 , synthetic labeled training data is generated based at least on a set of labeled sensor data associated with one or more industrial automation devices.

[0063] In block 302 , one or more machine learning models are trained for fault classification associated with one or more industrial automation devices based at least on synthetic labeled training data.

[0064] Figure 4 A flow chart according to an example embodiment of a method performed by one or more cloud services of a cloud platform 104 or an apparatus 500 or any other computing device including or consisting of a cloud platform 104 is illustrated.

[0065] refer to Figure 4 In box 401, a generative artificial intelligence model is trained based on a set of labeled sensor data associated with one or more industrial automation devices, wherein the set of labeled sensor data includes at least one of the following: real sensor data, synthetic sensor data, or simulated sensor data associated with one or more industrial automation devices.

[0066] For example, generative AI models could include variational autoencoders or generative adversarial networks.

[0067] The tagged sensor data set may include multiple health conditions and at least one of the following: speed data, temperature data, vibration data, load data, rating data, magnetic flux data, current data of one or more industrial automation devices. These measured quantities (e.g., speed, temperature, vibration, magnetic flux, current) are affected by the health conditions (tags) of one or more industrial automation devices.

[0068] The plurality of health conditions may include at least one of: health state, bearing failure, rotor imbalance, phase imbalance, overheating, stator eccentricity, rotor eccentricity, rotor bar breakage, shaft misalignment, bearing misalignment, or stator winding failure.

[0069] The one or more industrial automation devices may include at least one of: one or more electric motors or one or more variable speed drives.

[0070] In block 402, a trained generative artificial intelligence model is updated (fine-tuned) based on new sensor data associated with at least one industrial automation device having a confirmed health condition (e.g., a newly installed industrial automation device having a "healthy" status). After the generative artificial intelligence model is trained, new sensor data may be received.

[0071] In box 403, synthetic labeled training data is generated by using the updated generative artificial intelligence model.

[0072] The synthetic labeled training data may include multiple health conditions of multiple different sizes and ratings of one or more industrial automation devices under multiple operating conditions. For example, the synthetic labeled training data may include multiple health conditions and at least one of the following: synthetic speed data, synthetic temperature data, synthetic vibration data, synthetic load data, synthetic rating data, synthetic magnetic flux data, or synthetic current data of the one or more industrial automation devices.

[0073] In block 404 , one or more machine learning models are trained based at least on the synthetic labeled training data for fault classification associated with one or more industrial automation devices.

[0074] For example, one or more machine learning models may be trained by using an Extreme Gradient Boosting (XGBoost) algorithm.

[0075] In block 405, one or more health conditions associated with at least one of the one or more industrial automation devices are detected and classified using the trained one or more machine learning models. New sensor data (e.g., real-time sensor data) associated with the one or more industrial automation devices may be provided as input to the trained one or more machine learning models to detect and classify the one or more health conditions.

[0076] The above article uses Figure 3 to Figure 4 There is no absolute time order for the functions and / or blocks described, in which some functions and / or blocks may be executed simultaneously or in an order different from the described order. Other functions and / or blocks may also be executed between them or within them.

[0077] As used herein, “at least one of: ” and “at least one of ” and similar expressions, where a list of two or more elements is connected by “and” or “or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

[0078] Figure 5 An example of an apparatus 500 including components for performing one or more of the example embodiments described above is illustrated. The apparatus 500 may include or be composed of a cloud platform 104 or a server of the cloud platform 104. For example, the component may include one or more cloud services of the cloud platform 104, or the component may include a virtualized infrastructure. Cloud services refer to a wide range of on-demand computing resources and applications that the cloud platform 104 can provide to users via the Internet. Alternatively, the apparatus 500 may include any other computing device.

[0079] The apparatus 500 may include, for example, a circuit system or a chipset that may be suitable for implementing one or more of the example embodiments described above. The apparatus 500 may be an electronic device or a computing system that includes one or more electronic circuit systems. The apparatus 500 may include a control circuit system 510 (such as at least one processor) and at least one memory 520, which stores instructions 522 that, when executed by at least one microprocessor, cause the apparatus 500 to perform one or more of the example embodiments described above. For example, such instructions 522 may include computer program code (software). At least one processor and at least one memory storing instructions may provide a means for providing or causing the execution of any of the methods and / or blocks described above.

[0080] The processor is coupled to the memory 520. The processor is configured to read data from the memory 520 and write data to the memory 520. The memory 520 may include one or more memory cells. The memory cell may be volatile or non-volatile. It should be noted that there may be one or more non-volatile memory cells and one or more volatile memory cells, or alternatively, one or more non-volatile memory cells, or alternatively, one or more volatile memory cells. The volatile memory may be, for example, a random access memory (RAM), a dynamic random access memory (DRAM), or a synchronous dynamic random access memory (SDRAM). The non-volatile memory may be, for example, a read-only memory (ROM), a programmable read-only memory (PROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, an optical storage device, or a magnetic storage device. In general, the memory may be referred to as a non-transient computer-readable medium. The term "non-transient" as used herein is a restriction on the medium itself (i.e., tangible, not a signal), rather than a restriction on data storage persistence (e.g., RAM and ROM). The memory 520 stores computer readable instructions executed by the processor. For example, non-volatile memory stores computer readable instructions, and the processor executes the instructions using volatile memory for temporarily storing data and / or instructions.

[0081] The computer readable instructions may have been pre-stored in the memory 520, or alternatively or additionally, they may be received by the device via an electromagnetic carrier signal and / or may be copied from a physical entity (such as a computer program product). Execution of the computer readable instructions causes the device 500 to perform one or more of the functions described above.

[0082] Memory 520 may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and / or removable memory.

[0083] The device 500 may also include or be connected to a communication interface 530, which includes hardware and / or software for implementing a communication connection according to one or more communication protocols. The communication interface 530 may include at least one transmitter and at least one receiver, which may be integrated into the device 500 or the device 500 may be connected to the at least one transmitter and at least one receiver. The communication interface 530 may provide a component for executing some of the blocks of one or more of the example embodiments described above. The communication interface 530 may include one or more components controlled by a corresponding control unit, such as: a power amplifier, a digital front end (DFE), an analog-to-digital converter (ADC), a digital-to-analog converter (DAC), a frequency converter, a modulator (demodulator) and / or an encoder / decoder circuit system.

[0084] The communication interface 530 provides the apparatus with a communication capability to communicate in a wireless communication network. The communication interface 530 may, for example, provide a network interface for communicating with one or more sensor devices 103 and / or gateway device 106.

[0085] It should be noted that the apparatus 500 may also include Figure 5 Various components not shown in the figure. The various components can be hardware components and / or software components.

[0086] The term "circuitry" as used in this application may refer to one or more or all of the following: a) a pure hardware circuit implementation (such as an implementation in an analog and / or digital circuit system only); and b) a combination of hardware circuitry and software, such as (as applicable): i) a combination of one or more analog and / or digital hardware circuits with software / firmware and ii) one or more hardware processors with software (including one or more digital signal processors), software, and any portion of one or more memories that work together to enable a device such as a mobile phone to perform various functions); and c) one or more hardware circuits and / or one or more processors, such as one or more microprocessors or a portion of one or more microprocessors that requires software (e.g., firmware) to operate, but the software may not be present when it is not required for operation.

[0087] This definition of circuitry applies to all uses of this term in this application, including in any claims. As another example, as used in this application, the term circuitry also covers an implementation of only a hardware circuit or processor (or multiple processors) or a portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example, a baseband integrated circuit or processor integrated circuit for a mobile device, or a similar integrated circuit in a server, cellular network device, or other computing or network device, if applicable to the particular claim element.

[0088] The techniques and methods described herein can be implemented by various components. For example, these techniques can be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or a combination thereof. For hardware implementation, one or more devices of the example embodiments can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital information processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), graphics processing units (GPUs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof. For firmware or software, it can be implemented by a module (e.g., a process, a function, etc.) of at least one chipset that performs the functions described herein. The software code can be stored in a memory unit and executed by a processor. The memory unit can be implemented in or outside a processor. In the latter case, it can be coupled to the processor via various components known in the art. Additionally, it should be appreciated by those skilled in the art that the components of the system described herein can be rearranged and / or supplemented by additional components to facilitate the implementation of the various aspects described therein, etc., and they are not limited to the precise configurations listed in a given figure.

[0089] It is obvious to those skilled in the art that, with the advancement of technology, the inventive concept can be implemented in various ways within the scope of the claims. The embodiments are not limited to the example embodiments described above, but can be varied within the scope of the claims. Therefore, all words and expressions should be interpreted in a broad form, and they are intended to illustrate rather than limit the embodiments.

Claims

1. An apparatus for training one or more machine learning models for fault classification, the apparatus comprising at least one processor and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to at least: generating synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and The one or more machine learning models are trained based at least on the synthetic labeled training data for fault classification associated with the one or more industrial automation devices.

2. The apparatus of claim 1 , wherein the synthetic labeled training data is generated using a generative artificial intelligence model.

3. The apparatus according to claim 2, further configured to: The generative artificial intelligence model is trained based on the set of labeled sensor data.

4. The apparatus according to claim 3, further configured to: updating the generative artificial intelligence model based on new sensor data associated with at least one industrial automation device having a confirmed health condition, Wherein the new sensor data comprises sensor data received after the training of the generative artificial intelligence model.

5. The apparatus of claim 2, wherein the generative artificial intelligence model comprises a variational autoencoder or a generative adversarial network.

6. The apparatus according to claim 1 or 2, wherein the one or more machine learning models are trained by using an extreme gradient boosting (XGBoost) algorithm.

7. The apparatus of claim 1 or 2, wherein the set of labeled sensor data comprises at least one of: real sensor data, synthetic sensor data, or simulated sensor data.

8. The apparatus of claim 1, wherein the tagged sensor data set comprises a plurality of health conditions and at least one of: speed data, temperature data, vibration data, load data, rating data, magnetic flux data, or current data of the one or more industrial automation devices.

9. The apparatus of claim 1 or 2, wherein the synthetic labeled training data comprises a plurality of health conditions of the one or more industrial automation devices under a plurality of operating conditions at a plurality of different sizes and ratings.

10. The apparatus of claim 8, wherein the plurality of health conditions comprises at least one of: health state, bearing fault, rotor imbalance, phase imbalance, overheating, stator eccentricity, rotor eccentricity, rotor bar breakage, shaft misalignment, bearing misalignment, or stator winding fault.

11. The device according to claim 1 or 2, further configured to: One or more health conditions associated with at least one of the one or more industrial automation devices are detected and classified by using the one or more machine learning models trained based at least on the synthetic labeled training data.

12. The apparatus of claim 1 or 2, wherein the one or more industrial automation devices include at least one of the following: one or more electric motors or one or more variable speed drives.

13. The device according to claim 1 or 2, wherein the device comprises a cloud platform or is included in a cloud platform.

14. A method for training one or more machine learning models for fault classification, the method comprising: generating synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; as well as The one or more machine learning models are trained based at least on the synthetic labeled training data for fault classification associated with the one or more industrial automation devices.

15. A computer-readable medium comprising program instructions, which, when executed by an apparatus for training one or more machine learning models for fault classification, causes the apparatus to at least perform the following: generating synthetic labeled training data based on a set of labeled sensor data associated with one or more industrial automation devices; and The one or more machine learning models are trained based at least on the synthetic labeled training data for fault classification associated with the one or more industrial automation devices.

Images