Machine learning based direct method for determining status of facility control loop components

Abstract

This invention discloses a trained machine learning algorithm that processes time-series production data. The time-series production data represents control processes within a facility control loop. The trained machine learning algorithm is trained using positive training data representing normal operation of components within the facility control loop and negative training data representing abnormal operation of components within the facility control loop. The output of the trained machine learning algorithm identifies anomalies in the facility control loop.

Description

Technical Field

[0001] This disclosure relates to a direct machine learning-based method for determining the state of components of a facility control loop, and in one embodiment, but not by way of limitation, to a direct convolutional neural network (CNN)-based method for determining the state of components of an industrial control loop.

[0002] Related applications

[0003] This application claims priority to U.S. Serial Application No. 63 / 045,751, the contents of which are incorporated herein by reference in their entirety. Background Technology

[0004] Control valves are increasingly considered critical capital assets in any facility, such as an industrial plant. Due to their importance, these control valves should be monitored and maintained regularly. Properly maintained control valves help reduce process variability and improve product quality, which in turn improves the overall efficiency of the plant or other facility. Despite such monitoring and maintenance, control valves can still suffer from poor control performance due to valve nonlinearity. This nonlinearity includes viscosity, hysteresis, dead zone, and static area. These nonlinearities in the control loop cause oscillations, leading to variations in product quality, accelerated equipment wear, and instability in the control system.

[0005] In several valve nonlinearities, valve sticking is of particular concern in many cases. Valve sticking is the condition where the valve stem resists movement or fails to respond to output signals from the controller. Sticking is a specific problem in spring-diaphragm-actuated control valves, which are prevalent in processing industries (e.g., refineries). The main factors contributing to sticking are corrosion, lubricant loss, foreign matter intrusion, activation at the sliding metal surfaces at high temperatures, tight build-up around the valve stem, and / or chemical reactions between the valve's metal and the lubricant.

[0006] In industrial processing plants, control engineers typically analyze time series plots of PV (process value) - SP (setpoint) - OP (controller output) data and identify valve stickiness based on the typical shape and pattern of the PV / SP / OP plot. Figure 1A The image shows an example of a PV / SP / OP diagram where valve sticking is absent. Figure 1B The image shows an example of a PV / SP / OP diagram where viscosity exists. Point 110 indicates... Figure 1B Typical viscous patterns in [the text]. Attached Figure Description

[0007] Figure 1A An example of a PV / SP / OP diagram in which valve sticking is absent is shown.

[0008] Figure 1BAn example of a PV / SP / OP diagram in which valve stickiness exists is shown.

[0009] Figure 2 An example architecture of a convolutional neural network (CNN) is shown.

[0010] Figure 3 This is a flowchart of an exemplary implementation for training a machine learning algorithm to detect anomalies in a facility.

[0011] Figure 4A and Figure 4B This is a flowchart of another implementation scheme for training machine learning algorithms to detect anomalies in a facility.

[0012] Figure 5 This is a block diagram of a computer system on which one or more embodiments of the present disclosure may be executed. Detailed Implementation

[0013] In the following description, reference is made to the accompanying drawings, which form part of the detailed description, and in which specific embodiments that may be practiced are illustrated by way of illustration. These embodiments have been described in sufficient detail to enable those skilled in the art to practice the invention, and it should be understood that other embodiments may be utilized, and structural, electrical, and optical changes may be made, without departing from the scope of the invention. Therefore, the following description of exemplary embodiments is not restrictive, and the scope of the invention is defined by the appended claims.

[0014] One or more implementations involve machine learning-based methods for programming and improving the intelligence of control operators in industrial plants or other facilities. This intelligence is based on visual inspection of control loop trends. In one implementation, machine learning involves creating a convolutional neural network (CNN) to generate a dynamic model. This method can be used to detect process control anomalies, such as valve sticking and other control loop oscillations. The CNN can be trained to the same level as a group of trained control engineers to detect anomalies in industrial control loops. By automatically retraining the model using feedback from control engineers, the CNN can self-improve in cloud-based solutions.

[0015] As known to those skilled in the art, a convolutional network (CNN) is a neural network that uses convolutions instead of general matrix multiplications in at least one layer. CNNs are widely used in image and video recognition, image classification, medical image analysis, and natural language processing. CNNs excel at replicating human intelligence used in image processing and are therefore a very powerful and proven method for recognizing and interpreting images. A typical CNN architecture is... Figure 2 The diagram shows a CNN consisting of three convolutional layers, two multilayer perceptrons, and an output layer.

[0016] Convolutional neural network (CNN)-based learning techniques offer the ability to learn and tolerate noise in data, enabling applications in nonlinear systems and providing data-driven solutions where the system's behavior is unknown or complex. In one implementation, CNNs are used to process time-series facility production data. Specifically, the use of CNN-based learning techniques to detect valve stickiness and other nonlinearities in facility control loops has advanced the field in terms of prediction accuracy and computation time. Therefore, implementations can achieve the same level of accuracy as, or even surpass, that of human control engineers in detecting valve stickiness and other anomalies.

[0017] By incorporating machine learning, the implementation uses shape pattern matching of time-series production data to detect facility anomalies, such as valve sticking in industrial plants. In industrial plants, control engineers typically analyze time-series plots of PV (process value) - SP (setpoint) - OP (controller output) data and identify sticking based on typical shape patterns in the PV / SP / OP plots. See also Figure 1A and Figure 1B The implementation plan trains machine learning algorithms to identify anomalies using shape-patterning methods, such as those used by control engineers.

[0018] The foundation of viscous detection technology lies in the qualitative description of viscous phenomena and how closed-loop variables vary with viscosity and process parameters. Due to the integral action of the controller in an industrial process, valve viscosity and other nonlinearities leave different shapes in PV, SP, and OP data. Oscillation shapes can generally be classified as square, triangular, sawtooth, or sinusoidal. When the oscillation is not caused by viscosity, the shapes of OP and PV are inherently more sinusoidal. For a typical flow control loop with viscosity, OP exhibits a perfect triangular wave, while PV exhibits a square wave. Figure 1B The diagram shows a typical viscous pattern of a flow control loop.

[0019] In one implementation, the machine learning algorithm can be tested on simulation and industrial control loop data. When using industrial control loop data, the machine learning algorithm uses data signals directly on the time series data; that is, the algorithm does not work with shape-based data (such as triangles and trapezoids like CNNs in image processing). When the time series data is provided to a trained control engineer, the time series data is labeled using visual inspection of trends without any feature extraction. Therefore, the algorithm does not need to actually create a PV-SP-OP plot. In contrast, existing systems that use feature-based CNNs in all possibilities require data preprocessing. Such feature-based use also requires a PV-SP-OP plot or some other plot like a PV-OP bilabelled plot to identify shapes like ellipses, butterflies, or other shapes to confirm the presence of anomalies. However, in the implementation of this disclosure, no data preprocessing is required, and direct PV-SP-OP data from the plant or other facility is used instead of a plot.

[0020] It should be noted that sluggishness in control valves can be physically detected by means such as valve stroke testing and / or crash testing. However, these methods are practically impossible because there are thousands of control valves in a typical plant and these tests disrupt the plant processes. Furthermore, such existing methods can take hours to execute. Therefore, one implementation is a significantly faster method for detecting valve sluggishness. Once a CNN or other machine learning algorithm is trained, input tensors, including entire batches (i.e., all loops in the facility) or small batches (i.e., only some loops in the facility due to memory limitations), can be processed simultaneously in a shorter amount of time (e.g., within minutes). And unlike valve stroke testing and crash testing, trained machine learning algorithms are a non-invasive and automated method for detecting valve sluggishness and other plant anomalies.

[0021] In addition to training and learning algorithms based on the intelligence and knowledge of control engineers, the implementation can integrate feedback from control engineers by being deployed in a distributed computing environment, such as in the cloud. In a cloud-based implementation, CNNs or other machine learning algorithms can be trained to the same level as a set of trained control engineers, and the performance of the CNN can be improved by retraining via the cloud based on feedback from control engineers. An implementation is also significantly more robust in detecting valve stickiness and other anomalies. As explained in more detail here, trained control engineers label the training data. This allows the implementation to reach the same level of accuracy as the control engineers. In fact, with continued training and machine learning, the system can sometimes surpass the accuracy of the control engineers. This robustness in detection allows the system to better handle problems such as excessive sensor measurement noise, and it helps the system avoid being contaminated by such noise. Furthermore, robustness allows the implementation to perform well in the presence of poor controller tuning and excessive sensor measurement noise. Moreover, the implementation can distinguish between oscillations caused by stickiness, poor controller tuning, and excessive sensor measurement noise. By appropriately labeling training data relevant to such problems and training the algorithm based on that appropriately labeled data, this implementation can distinguish between valve stickiness and other anomalies, performs well even with poor controller tuning and excessive measurement noise, and can differentiate between different causes of oscillation. Therefore, if the training data has been properly labeled, the final output layer of the CNN can perform multi-class classification to determine whether oscillations in the facility control loop are caused by valve stickiness, component tuning problems, process noise, or measurement noise.

[0022] The embodiments disclosed herein offer several advantages. First, the same network can be applied to detect oscillations and tuning problems in control loops, provided the industrial processes are structurally similar. Second, vectorization can be used to process thousands of loops in a very short time (e.g., within seconds). Third, as mentioned earlier, the final output layer of the CNN can perform multi-class classification. Fourth, the various embodiments are unit-agnostic, i.e., they only require process values ​​(PV), setpoints (SP), and outputs (OP).

[0023] Several development challenges were overcome in developing the implementation scheme disclosed herein. The first development challenge involves data augmentation. CNNs are primarily used for processing image data, where methods for data augmentation are well-established. However, in one implementation, CNNs are used for time series data, where methods for data augmentation prove more complex. Specifically, data augmentation in the control loop PV / SP / OP data is not as straightforward as in images. For example, each “Z-shape” in the PV or SP data should correspond to a symmetrical “Z-shape” in the OP, as the system needs to adhere to the physical and chemical rules governing the system. This correspondence had to be resolved by the inventors to verify the correct operation of the system.

[0024] Another development challenge faced by the inventors involved the fact that different control loop types have different characteristics. While control engineers can visually classify viscous issues in fast loops (i.e., rapid flow of ingredients and / or products), it is not always possible to visually classify viscous issues in loops with significant delays, dynamic loops, and / or integrated loops. To overcome this problem, the inventors created simulated data to supplement real data, and then used the real data augmented with simulated data to train the network. The resulting trained model can then be used to process and interpret current real production data.

[0025] After overcoming these obstacles, the following high-level implementation scheme was developed. Data representing the normal operation of the facility and data representing one or more anomalies of the facility are collected. This data can be referred to as labeled or annotated data. For example, in an industrial facility, data related to valve stickiness can be labeled as anomaly. A machine learning algorithm is trained using the labeled data. As mentioned above, the machine learning algorithm can be a convolutional neural network (CNN), and in a specific implementation, the CNN can be a CNN with 3 to 4 convolutional layers followed by 3 to 4 fully connected layers. After initially training the CNN with labeled data, the hyperparameters of the CNN are tuned and the CNN is evaluated. Then, if necessary, the CNN is retuned.

[0026] As described above, both real and simulated data are used to train the machine learning algorithm, and the resulting model is then used to process and interpret real production data. More specifically, in one implementation, the overall process of using machine learning algorithms such as CNNs to classify anomalies in industrial plants, such as valve stickiness, is now discussed in detail. Process variables (PV), setpoints (SP), and controller output (OP) signals are selected for a specific number of simulated valve flow control loops. For example, 3,000 simulated flow loops can be selected. Furthermore, a certain number of samples, for example, 2,000 samples, are selected from each time-series control loop. For each loop, the data is classified / labeled as representing valve stickiness data or valve data representing no valve stickiness.

[0027] In this specific example, a 2,000 x 3,000 vector is then created using the control loop data from the valve viscosity classification, and this vector is used to train a one-dimensional CNN. The data is split into a training set for first training the CNN and a test set for then testing the trained CNN. The split between the training and test sets can be a 90:10 ratio. The CNN is then trained using five convolutional layers, four fully connected layers, and a rectified linear unit (ReLU) activation function. After the training phase is complete, the test dataset is fed to the trained CNN to determine if it can correctly identify and predict valve viscosity. Control engineers are typically the ones who interpret the outputs and results of the CNN that processes the test data.

[0028] After training, testing, and tuning, the trained CNN can be used on production data from an industrial plant or other facility. First, a specific number of PV, SP, and OP signals are selected from real-world plant flow control loops. In one implementation, the number of selected signals could be approximately 10,000, and the number of samples in each time-series control loop data could be approximately 5,000. The trained CNN can then use production data to classify each loop, i.e., whether it exhibits or does not exhibit stickiness. Subsequently, the CNN can be retrained using real-world plant data, which requires further interpretation of the results and retuning of the CNN.

[0029] In another implementation, the real-world factory data can be augmented by using a number of consecutive samples after every certain number of samples in each loop. For example, 2,000 consecutive samples could be used after every 100 samples in each loop. Subsequently, the CNN can be retrained and retuned, the real-world factory data can be data-balanced so that the number of viscous and non-viscous valve loops is approximately equal, and the CNN can then be retrained and retuned for final timing.

[0030] The above process of selection, training, and retraining is in Figure 3 , Figure 4A and Figure 4B It is shown in graphic form. Figure 3 , Figure 4A and Figure 4B This is a block diagram illustrating the operation and characteristics of systems and methods used to train machine learning algorithms to identify anomalies in facility control loops. Figure 3 , Figure 4A and Figure 4B These include multiple boxes 310 to 329 and 410 to 472, respectively. Although Figure 3 , Figure 4A and Figure 4BIn the examples, the boxes are arranged substantially sequentially; however, other examples may use multiple processors organized into two or more virtual machines or subprocessors, or a single processor, to reorder the boxes, omit one or more boxes, and / or execute two or more boxes in parallel. Furthermore, other examples may implement these boxes as one or more specific interconnect hardware or integrated circuit modules, where associated control and data signals are transmitted between and through modules. Therefore, any process flow is applicable to software, firmware, hardware, and hybrid implementations.

[0031] See now Figure 3 The process can be divided into a development phase 310 and a deployment phase 320. The deployment phase further comprises an online phase 321 and an offline phase 328. At 311, annotated / labeled loop data is received from simulated data and sample factory data. As noted, these data in the valve viscosity example can represent PV, SP, and OP data. At 312, the received data is augmented. This data augmentation may include data resampling and data pruning. The data is placed in a vector format. At 313, the vectorized data is used to train a one-dimensional CNN using five convolutional layers, four fully connected layers, and a ReLU activation layer. Operation 313 produces the trained CNN or model 314.

[0032] The trained model 314 is provided to the deployment phase 320, which begins at model execution 322. At 323, real-time production loop data is provided to the trained model 314. As noted, this production data may include PV, SP, and OP data. Model execution 322 generates a decision regarding whether the control loop discussed at 324 is experiencing valve stickiness. At 325, the control engineer can provide feedback on the accuracy and correctness of valve stickiness / non-stickiness through the trained CNN, which can be used to update the model at 329, and the updated model can be used at model execution 322.

[0033] Figure 4A and Figure 4B Another implementation scheme for training a machine learning algorithm to identify anomalies in facility control loops is shown. Figure 4A and Figure 4B In a more specific implementation, machine learning and CNNs are used directly on time-series data (not feature-based shape-based machine learning) to learn the operation of industrial process control loop systems and to identify and react to anomalies within these systems. See now for details. Figure 4A and Figure 4BAt 410, the computer system accesses the time-series production data. This time-series production data represents the control processes in the control loops of a facility. Facilities may include industrial plants such as refineries, manufacturing plants, pulp mills, and steel mills, and may also include buildings such as offices, museums, theaters, and schools. Control loops may refer to, for example, hybrid loops in an industrial facility or heating, ventilation, and air conditioning (HVAC) loops in an office building. This time-series production data may include process variables, setpoints, and controller outputs. At 415, the time-series production data is placed in a vector.

[0034] At 420, a trained machine learning algorithm is used to process the time-series production data. The trained machine learning algorithm processes the vector as time-series production data is added. The trained machine learning algorithm is trained using positive training data representing normal operation of components within the facility control loop and negative training data representing abnormal operation of components within the facility control loop. Positive training data may include, for example, data related to process values ​​(PV), setpoints (SP), and controller outputs (OP), and may indicate the absence of valve stickiness, oscillating loops, poorly performing loops, overperforming loops, data spikes, and frozen data. Negative training data may include the presence of valve stickiness, oscillating loops, poorly performing loops, overperforming loops, data spikes, and frozen data. An example of a component being analyzed is a control valve. In one embodiment, the positive and negative training data are labeled data representing the operation of components within the facility control loop. Multiple labels are considered positive or negative depending on the absence or presence of a combination of potential problems (e.g., valve stickiness, oscillating loops, poorly performing loops, overperforming loops, data spikes, and frozen data).

[0035] Furthermore, positive and negative training data can undergo data balancing and data augmentation. Data balancing refers to structuring the positive and negative training data so that the data are approximately equal in quantity. For example, there should be approximately 1 megabyte of positive training data and 1 megabyte of negative training data. In one implementation, data augmentation increases the size and diversity of the data used to train the machine learning algorithm without actually collecting new data. Data augmentation can be achieved through upsampling, downsampling, windowing, and using simulated data. In windowing, a constant window of data is obtained using different starting samples to increase the data size. In upsampling, interpolation is used to reduce the sampling time, and in downsampling, the sampling time is increased by maintaining a constant value for every nth sample. Therefore, the trained machine learning algorithm can process multiple sampling periods, regardless of whether the duration of the sampling period is 1 second, 2 seconds, 5 seconds, or longer.

[0036] At 430, one or more anomalies in the facility control loop are identified based on the output of a trained machine learning algorithm. These anomalies may include valve nonlinearity and, as described above, may include valve stickiness, oscillating loops, poorly performing loops, overperforming loops, data spikes, and frozen data. The trained machine learning algorithm can classify several different anomalies based on time-series production data. This can be referred to as multi-classification. For example, the trained machine learning algorithm can classify one or more of the following using a specifically labeled time-series production dataset: component failure, component tuning problems, noise from the facility control loop, and / or noise from measuring components (e.g., temperature sensors) in the facility control loop. As shown at 435, one or more anomalies can be identified using the output of the trained machine learning algorithm on the processed vector. Furthermore, this implementation can improve the system's processing speed. At 440, a signal indicating one or more anomalies is transmitted to a computer display device. For example, the computer display device may be a terminal in a control room and / or a mobile device of a controller engineer. As shown at 440A, the transmitted signal may suggest or instruct the control engineer to take one or more actions to resolve the anomaly. For example, the suggestion may include adjusting a specific process of the facility, adjusting one or more valves, adjusting one or more temperatures, and / or adjusting one or more pressures.

[0037] As noted at 421, the trained machine learning algorithm can be a convolutional neural network (CNN). Many variations of CNNs can be used. However, in one implementation, a CNN may include three or four convolutional layers and three or four fully connected layers. In another implementation, a CNN may be a one-dimensional CNN including five convolutional layers, and these five convolutional layers may include four fully connected layers and a rectified linear unit (ReLU) activation function.

[0038] At point 422, the trained machine learning algorithm can determine that the oscillation control loop is caused by external factors. External factors are those factors outside the configuration and function of the components in the facility control loop. For example, an external factor could be the complete malfunction of a specific piece of equipment. Of course, as mentioned above, the trained machine learning algorithm can also determine that the oscillation control loop is caused by internal factors such as valve sticking.

[0039] At point 431, the trained machine learning algorithm identifies data spikes in the time series production data. An example of a data spike is a temperature outside an acceptable set range. At point 431A, the trained machine learning algorithm further identifies the location of the data spike in the time series production data. When training the machine learning algorithm, the presence and location of data spikes can be identified by tagging historical time series data for the presence or absence of data spikes and their locations.

[0040] In one implementation, as shown at 450, in a distributed computing resource system, such as in the cloud, the processing of time-series production data and the identification of anomalies in the time-series production data are performed using a trained machine learning algorithm. When executed in the cloud, as shown at 451, input regarding the identified anomalies can be collected from a control engineer, and at 452, the trained machine learning algorithm can be retrained based on the control engineer's input. For example, retraining may include tuning the hyperparameters associated with the machine learning algorithm and then re-evaluating the machine learning algorithm.

[0041] Figure 5 This is a block diagram of a machine in the form of a computer system, within which a set of instructions can be executed to cause the machine to perform any one or more of the methods discussed herein. In alternative embodiments, the machine operates as a standalone device or can be connected (e.g., networked) to other machines. In a networked deployment, the machine may operate as a server or client machine in a client-server network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. In a preferred embodiment, the machine will be a server computer; however, in alternative embodiments, the machine may be a personal computer (PC), tablet PC, set-top box (STB), personal digital assistant (PDA), mobile phone, network device, network router, switch or bridge, or any machine capable of (sequentially or otherwise) executing instructions specifying the actions to be taken by the machine. Furthermore, although a single machine is shown, the term "machine" should also be considered to include any collection of machines that individually or jointly execute one or more sets of instructions to perform one or more of the methods discussed herein.

[0042] An exemplary computer system 500 includes a processor 502 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), or both) communicating with each other via a bus 508, main memory 501, and static memory 506. The computer system 500 may further include a display unit 510. The computer system 500 may additionally include a storage device 516 (e.g., a drive unit), a signal generation device 518 (e.g., a speaker), a network interface device 520, and one or more sensors 528, such as a GPS sensor, a compass, an accelerometer, or other sensors.

[0043] The drive unit 516 includes a machine-readable medium 522 on which one or more sets of instructions and data structures (e.g., software 524) embodying or utilized by any one or more methods or functions described herein are stored. The software 524 may also reside wholly or at least partially within main memory 501 and / or processor 502 during execution by computer system 500, which also constitute the machine-readable medium.

[0044] Although machine-readable medium 522 is shown as a single medium in the exemplary embodiment, the term "machine-readable medium" can include a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) storing one or more instructions. The term "machine-readable medium" should also be considered to include any tangible medium capable of storing, encoding, or carrying instructions for machine execution and any or more methods for causing a machine to perform the present invention, or any tangible medium capable of storing, encoding, or carrying data structures utilized by or associated with such instructions. Therefore, the term "machine-readable medium" should be considered to include, but is not limited to, solid-state memory as well as optical and magnetic media. Specific examples of machine-readable media include: non-volatile memory, including, for example, semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.

[0045] Software 524 can further transmit or receive data via network interface device 520 using a transmission medium through communication network 526 using any of a number of well-known transmission protocols (e.g., HTTP). Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet, mobile phone networks, conventional telephone (POTS) networks, and wireless data networks (e.g., and (Network). The term "transmission medium" should be considered to include any intangible medium capable of storing, encoding, or carrying instructions for machine execution, and includes digital or analog communication signals or other intangible media to facilitate communication of such software.

[0046] It should be understood that other variations and modifications of the invention and its various aspects exist, which may be apparent to those skilled in the art, and the invention is not limited to the specific embodiments described herein. The above features and embodiments can be combined with each other in different combinations. Therefore, it is contemplated to cover any and all modifications, variations, combinations, or equivalents falling within the scope of this invention.

[0047] The summary of the specification is provided to comply with 37 C. FR § 1.72(b) and will allow the reader to quickly determine the nature and essence of the disclosure. It should be understood that it is not intended to interpret or limit the scope or meaning of the claims.

[0048] In the foregoing description of the embodiments, various features are grouped together in a single embodiment for the purpose of simplifying this disclosure. This approach of the disclosure should not be construed as reflecting that the claimed embodiments have more features than expressly stated in each claim. Rather, as reflected in the following claims, the subject matter of the invention lies in fewer than all features of a single disclosed embodiment. Therefore, the following claims are hereby incorporated into the detailed description, wherein each claim exists independently as a separate exemplary embodiment.

Claims

1. A method comprising: A time-series production dataset representing the control process within a facility control loop, including at least one control valve, is accessed by a computer processor. The time-series production dataset is processed by the computer processor using a trained machine learning algorithm, which is trained using positive training data representing normal operation of one or more control valves within the facility control loop and negative training data representing abnormal operation of one or more control valves within the facility control loop, wherein the positive training data and the negative training data undergo data balancing and data augmentation. The computer processor identifies one or more anomalies associated with at least one control valve in the facility control loop based on the output of the trained machine learning algorithm, wherein the one or more anomalies associated with at least one control valve in the facility control loop indicate at least an anomaly based on a control valve nonlinearity. as well as The computer processor transmits a signal to the computer display device indicating one or more anomalies associated with at least one control valve in the facility control loop.

2. The method according to claim 1, further comprising: Based on the identification of the one or more anomalies, a signal instructing one or more actions for resolving the one or more anomalies is transmitted to the computer display device.

3. The method according to claim 1, further comprising: The computer processor identifies data spikes in the time series production dataset and the location of the data spikes in the time series production dataset based on the output of the trained machine learning algorithm.

4. The method according to claim 1, further comprising: The computer processor identifies oscillation control loops caused by external factors based on the output of the trained machine learning algorithm.

5. The method according to claim 1, comprising: The time-series production dataset is processed by the computer processor, and the processing identifies one or more anomalies in the distributed computer resource system; Collect inputs to identify the one or more anomalies; and retrain the trained machine learning algorithm based on the inputs.

6. The method according to claim 1, further comprising: The computer processor places the time-series production dataset into a vector; The trained machine learning algorithm is used to process the vector; And use the processed vectors to identify one or more of the anomalies.

7. An apparatus comprising: One or more processors; Memory; and One or more programs stored in memory, the one or more programs including instructions for performing a process, including: A time-series production dataset representing the control process within a facility control loop, including at least one control valve, is accessed by a computer processor. The time-series production dataset is processed by the computer processor using a trained machine learning algorithm, which is trained using a training dataset comprising positive training data representing normal operation of one or more control valves within the facility control loop and negative training data representing abnormal operation of one or more control valves within the facility control loop, wherein the positive training data and the negative training data undergo data balancing and data augmentation. The computer processor identifies one or more anomalies associated with at least one control valve in the facility control loop based on the output of the trained machine learning algorithm, wherein the one or more anomalies associated with the at least one control valve in the facility control loop indicate at least an anomaly based on a control valve nonlinearity; and The computer processor transmits a signal to the computer display device indicating one or more anomalies associated with at least one control valve in the facility control loop.

8. The device according to claim 7, further comprising: The computer processor identifies data spikes in the time series production dataset and the location of the data spikes in the time series production dataset based on the output of the trained machine learning algorithm.

9. The device according to claim 7, further comprising: The computer processor identifies oscillation control loops caused by external factors based on the output of the trained machine learning algorithm.

10. A non-transitory computer-readable storage medium comprising one or more programs executable by one or more processors of a device, the one or more programs comprising instructions that, when executed by the one or more processors, cause the device to perform the following process: A time-series production dataset representing the control process within a facility control loop, including at least one control valve, is accessed by a computer processor. The computer processor processes the time-series production dataset using a trained machine learning algorithm, which is trained using a training dataset comprising positive training data representing normal operation of one or more control valves within the facility control loop and negative training data representing abnormal operation of one or more control valves within the facility control loop, wherein the training dataset includes actual training data augmented at least in part based on the augmented training dataset. The computer processor identifies one or more anomalies associated with at least one control valve in the facility control loop based on the output of the trained machine learning algorithm, wherein the positive training data and the negative training data undergo data balancing and data augmentation; and The computer processor transmits a signal to the computer display device indicating one or more anomalies associated with at least one control valve in the facility control loop.

Images