System and method for predicting manufacturing process risks - Patents.com

Abstract

The present disclosure provides systems, methods, and computer program products for predicting and detecting anomalies in a subsystem of a system. One example method may include (a) determining a first plurality of tags indicative of operational performance of the subsystem. The tags can be obtained from (i) a plurality of sensors in the subsystem and (ii) a plurality of sensors in the system that are not in the subsystem. The method may further include (b) processing measurements of the first plurality of tags using an autoencoder trained to generate estimates of the first plurality of tags based on historical values ​​of the first plurality of tags; (c) determining whether a difference between the measurements and the estimates satisfies a threshold; and (d) transmitting an alert indicating that the subsystem is predicted to experience an anomaly if the difference satisfies the threshold.

Description

[Technical field]

[0001] (cross reference) This application claims priority to U.S. Provisional Patent Application No. 62 / 899,081, filed Sep. 11, 2019, which is incorporated by reference in its entirety. [Background technology]

[0002] Manufacturers may rely on a framework of scheduled and reactive maintenance to manage their operating assets, which can result in lost revenue from postponed production, excessive costs associated with unplanned maintenance actions, shortened asset life, poor product quality, and worker safety risks associated with increased on-site exposure. Summary of the Invention [Means for solving the problem]

[0003] The present disclosure provides a system, method, and computer program product for predicting and detecting anomalies in a system or subsystem of a processing pipeline and improving the reliability of the processing pipeline. The processing pipeline may be, for example, an oil and gas processing pipeline, including drilling assets, refining assets, and pipeline assets (e.g., pumps, compressors, heat exchangers, and valves). A system for predicting and detecting anomalies as described herein can be implemented on one or more computers in one or more locations. The computer can store instructions that, upon execution, cause the computer to predict anomalies in a subsystem of the system.

[0004] The system can determine a first plurality of tags indicative of the subsystem's operational performance. The first plurality of tags may include time series data from sensors in the subsystem. The first plurality of tags may additionally include time series data from sensors in the system that are not in the subsystem but are nevertheless indicative of the subsystem's operational performance. For example, the first plurality of tags may be from another subsystem that provides input to the subsystem, receives output from the subsystem, or is otherwise correlated with the subsystem. The system can identify the first plurality of tags using one or more machine learning methods. For example, the system can use a clustering algorithm to identify the first plurality of tags from all of the tags in the system. Determining the first plurality of tags may be considered a feature selection operation.

[0005] The system can process measurements of the first plurality of tags using an autoencoder. The autoencoder can have an encoder configured to generate a reduced-dimensionality or compressed output from the measurements. The autoencoder can also have a decoder configured to generate estimates of the first plurality of tags from the reduced-dimensionality or compressed output. The autoencoder can be trained based on historical values ​​of the first plurality of tags.

[0006] The system can determine whether a difference exists between the measurements of the first plurality of tags provided to the autoencoder and the estimates of the first plurality of tags generated by the autoencoder. If the measurements and estimates are identical or similar, the system can determine that the subsystem is not experiencing an anomaly. On the other hand, if the system determines that the difference between the measurements and estimates satisfies a threshold, the system can determine that the subsystem is experiencing an anomaly. The system can transmit an alert indicating that the system is experiencing such an anomaly.

[0007] While many manufacturing entities rely on scheduled or reactive maintenance to manage operational assets, the system described above can accurately predict and detect failures and anomalies in a processing pipeline before they occur by using data from hundreds to thousands of sensors and identifying correlations within the data. Because steps within a process can be highly interdependent, the correlations identified by the system described herein can be useful in identifying anomalies. In addition, the system can use machine learning models that can be retrained based on new failure events to improve prediction and detection accuracy over time. Finally, the system can predict failures and detect anomalies at the asset, system, or subsystem level, which can enable users to quickly perform scheduled maintenance on a processing pipeline, maintain safety, extend asset life, produce higher quality products, and avoid costly downtime.

[0008] The present disclosure may additionally provide applications and user interfaces that allow a user, e.g., an operations manager or engineer, to interact with and view the output of the systems described above. The applications and user interfaces may additionally allow the user to monitor, control, and perform predictive maintenance on assets in the processing pipeline.

[0009] These applications and user interfaces can provide an end-to-end solution for prioritizing, investigating, and responding to predicted or detected anomalies within a processing pipeline.

[0010] In one aspect, the present disclosure provides a method for predicting an anomaly in a subsystem of a system. The method may include determining a first plurality of tags indicative of an operational performance of the subsystem. Each tag of the first plurality of tags may include a data sequence, and the first plurality of tags may be obtained from (i) a plurality of sensors in the subsystem and (ii) a plurality of sensors in the system that are not in the subsystem. The method may further include processing measurements of the first plurality of tags with an autoencoder by (1) generating a condensed representation of the measurements and (2) decoding the condensed representation of the measurements into estimates of the first plurality of tags, the autoencoder being trained based on historical values ​​of the first plurality of tags. The method may further include determining whether a difference between the measurements and the estimates satisfies a threshold value, and transmitting an alert indicating that the subsystem is predicted to experience an anomaly if the difference satisfies the threshold value.

[0011] In some implementations, (a) includes using an unsupervised learning algorithm to process data from sensors in the system to generate a plurality of populations of tags, the plurality of populations of tags including a first plurality of tags indicative of an operational performance of a subsystem. In some implementations, each tag of the first plurality of tags is correlated with at least one other tag of the first plurality of tags or with the operational performance of the subsystem. In some implementations, the unsupervised learning algorithm is a clustering algorithm. The clustering algorithm may be a k-means clustering algorithm, a cosine similarity algorithm, a topological data analysis algorithm, or a hierarchical density-based spatial clustering for applications with noise (HDB-SCAN).

[0012] In some implementations, (a) is performed offline.

[0013] In some implementations, the method further includes applying a weighting process to each of the measurements and the estimates prior to (c).

[0014] In some implementations, the method further includes dynamically adjusting the threshold value during operation of the system.

[0015] In some implementations, the method further includes determining whether a difference between the individual tag measurement value and the individual tag estimate satisfies a tag threshold and transmitting an alert if the difference satisfies the tag threshold.

[0016] In some implementations, the autoencoder is a deep neural network or a one-class support vector machine (SVM).

[0017] In some implementations, the historical data includes data series collected during periods when no faults occurred within the subsystem.

[0018] In some implementations, the system comprises multiple subsystems.

[0019] In some implementations, the measurements of the first plurality of tags include a data series collected during a predetermined lookback horizon. In some implementations, the method further includes dynamically adjusting the lookback horizon.

[0020] In some implementations, the method further includes generating a risk score for the subsystem based on a magnitude of a difference between the measurement values ​​of the first plurality of tags and the estimated values ​​of the first plurality of tags.

[0021] In some implementations, the alert includes an email or text message. In some implementations, the alert can include a notification within a mobile or web application. In some implementations, the mobile or web application is configured to allow a user to implement one or more corrective actions regarding the subsystem.

[0022] In some implementations, the method further includes determining actions to correct the anomaly in the subsystem using a trained machine learning algorithm. The trained machine learning algorithm can be trained based on historical human actions and corresponding responses by the subsystem. In some implementations, the method further includes determining actions to correct the anomaly in the subsystem using rules-based logic based on engineering principles to correlate observed patterns in the tag measurements to known failure mechanisms for processes and equipment in the subsystem.

[0023] In some implementations, the alert includes a numeric risk score indicating the likelihood that an anomaly will occur. The numeric risk score can be based, at least in part, on the magnitude of the difference between the measurements of the first plurality of tags and the estimates of the first plurality of tags. In some implementations, the alert includes a numeric risk score indicating the likelihood that an anomaly will occur. The numeric risk score can be based, at least in part, on the amount by which the difference exceeds a threshold. In some implementations, the alert includes a list of tags from the first plurality of tags that contributed most to the alert.

[0024] Another aspect of the present disclosure provides a non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements any of the methods described above or elsewhere herein.

[0025] Another aspect of the present disclosure provides a system comprising one or more computer processors and a computer memory coupled thereto, the computer memory comprising machine executable code that, upon execution by the one or more computer processors, implements any of the methods described above or elsewhere herein.

[0026] Additional aspects and advantages of the present disclosure will become readily apparent to those skilled in the art from the following detailed description, in which only illustrative embodiments of the present disclosure are shown and described. As will be realized, the present disclosure is capable of other and different embodiments, and its several details are capable of modification in various obvious respects, all without departing from the present disclosure. Thus, the drawings and description should be regarded as illustrative in nature, and not restrictive. The present invention provides, for example, the following: (Item 1) 1. A method for predicting anomalies in a subsystem of a system, comprising: (a) determining a first plurality of tags indicative of operational performance of the subsystem, each tag of the first plurality of tags including a data sequence, the first plurality of tags being obtained from (i) a plurality of sensors in the subsystem and (ii) a plurality of sensors in the system that are not in the subsystem; (b) processing measurements of the first plurality of tags with an autoencoder by (1) generating compressed representations of the measurements and (2) decoding the compressed representations of the measurements into estimates of the first plurality of tags, the autoencoder being trained based on historical values ​​of the first plurality of tags; and (c) determining whether a difference between the measured value and the estimated value satisfies a threshold; and (d) transmitting an alert indicating that the subsystem is predicted to experience an anomaly if the difference satisfies the threshold. A method comprising: (Item 2) 2. The method of claim 1, wherein (a) includes using an unsupervised learning algorithm to process data from sensors in the system to generate a plurality of populations of tags, the plurality of populations of tags including the first plurality of tags indicative of operational performance of the subsystem. (Item 3) 3. The method of claim 2, wherein each tag of the first plurality of tags is correlated with at least one other tag of the first plurality of tags or with operational performance of the subsystem. (Item 4) 3. The method of claim 2, wherein the unsupervised learning algorithm is a clustering algorithm. (Item 5) 5. The method according to claim 4, wherein the clustering algorithm is a k-means clustering algorithm, a cosine similarity algorithm, a topological data analysis algorithm, or a hierarchical density-based spatial clustering for noisy applications (HDB-SCAN). (Item 6) 2. The method according to claim 1, wherein (a) is performed offline. (Item 7) 2. The method of claim 1, further comprising, prior to (c), applying a weighting process to each of the measured values ​​and the estimated values. (Item 8) 2. The method of claim 1, further comprising dynamically adjusting the threshold during operation of the system. (Item 9) 2. The method of claim 1, further comprising: determining whether a difference between the measurement value of an individual tag and the estimated value of the individual tag satisfies a tag threshold; and transmitting an alert if the difference satisfies the tag threshold. (Item 10) 2. The method of claim 1, wherein the autoencoder is a deep neural network or a support vector machine (SVM). (Item 11) 2. The method of claim 1, wherein the historical data includes a data series collected during a period when no faults occurred in the subsystem. (Item 12) 2. The method of claim 1, wherein the system comprises a plurality of subsystems. (Item 13) 2. The method of claim 1, wherein the measurements of the first plurality of tags include a data series collected during a predetermined lookback horizon. (Item 14) Item 14. The method of item 13, further comprising dynamically adjusting the lookback horizon. (Item 15) 2. The method of claim 1, further comprising generating a risk score for the subsystem based on a magnitude of the difference between the measurement values ​​of the first plurality of tags and the estimated values ​​of the first plurality of tags. (Item 16) 2. The method of claim 1, wherein the alert comprises an email or a text message. (Item 17) 2. The method of claim 1, wherein the alert comprises a notification within a mobile or web application. (Item 18) 20. The method of claim 17, wherein the mobile or web application is configured to enable a user to implement one or more corrective actions regarding the subsystem. (Item 19) 2. The method of claim 1, further comprising determining actions to correct the anomaly in the subsystem using a trained machine learning algorithm, the trained machine learning algorithm being trained based on historical actions taken by a person and corresponding responses by the subsystem. (Item 20) 2. The method of claim 1, further comprising determining actions to correct the anomaly in the subsystem using rules-based logic based on engineering principles for correlating observed patterns in tag measurements to known failure mechanisms for processes and equipment in the subsystem. (Item 21) 2. The method of claim 1, wherein the alert includes a numerical risk score indicating the likelihood that the anomaly will occur, the numerical risk score being based, at least in part, on a magnitude of a difference between the measurements of the first plurality of tags and the estimated values ​​of the first plurality of tags. (Item 22) 2. The method of claim 1, wherein the alert includes a numerical risk score indicating the likelihood that the anomaly will occur, the numerical risk score being based, at least in part, on the amount that the difference exceeds the threshold. (Item 23) 2. The method of claim 1, wherein the alert includes a list of tags from the first plurality of tags that contributed most to the alert. (Item 24) 1. A system comprising one or more computers and one or more storage devices, the one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to: (a) determining a first plurality of tags indicative of operational performance of a subsystem of a system, each tag of the first plurality of tags including a data sequence, the first plurality of tags being obtained from (i) a plurality of sensors in the subsystem, and (ii) a plurality of sensors in the system that are not in the subsystem; (b) processing measurements of the first plurality of tags with an autoencoder by (1) generating compressed representations of the measurements and (2) decoding the compressed representations of the measurements into estimates of the first plurality of tags, the autoencoder being trained based on historical values ​​of the first plurality of tags; and (c) determining whether a difference between the measured value and the estimated value satisfies a threshold; and (d) transmitting an alert indicating that the subsystem is predicted to experience an anomaly if the difference satisfies the threshold. The system is operable to perform operations including: (Item 25) One or more non-transitory computer storage media having instructions stored thereon that, when executed by one or more computers, cause the one or more computers to: (a) determining a first plurality of tags indicative of operational performance of a subsystem of a system, each tag of the first plurality of tags including a data sequence, the first plurality of tags being obtained from (i) a plurality of sensors in the subsystem and (ii) a plurality of sensors in the system that are not in the subsystem; (b) processing measurements of the first plurality of tags with an autoencoder by (1) generating compressed representations of the measurements and (2) decoding the compressed representations of the measurements into estimates of the first plurality of tags, the autoencoder being trained based on historical values ​​of the first plurality of tags; and (c) determining whether a difference between the measured value and the estimated value satisfies a threshold; and (d) transmitting an alert indicating that the subsystem is predicted to experience an anomaly if the difference satisfies the threshold.

[0023] 1. One or more non-transitory computer storage media operable to cause a computer to perform operations including: (Incorporated by reference)

[0027] All publications, patents, and patent applications mentioned herein are incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference. To the extent that the publications and patents or patent applications incorporated by reference conflict with the disclosure contained herein, it is intended that the present specification takes precedence over and / or supersedes any such conflicting material. [Brief description of the drawings]

[0028] The novel features of the invention are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present invention will be obtained by reference to the following detailed description and the accompanying drawings (also herein "Figure and FIG.") that set forth illustrative embodiments, in which the principles of the invention are utilized.

[0029] [Figure 1] FIG. 1 diagrammatically illustrates a system for predicting and detecting anomalies.

[0030] [Diagram 2] FIG. 2 is a flowchart of an exemplary process for predicting anomalies in subsystems, generating alerts about the anomalies, generating cases to follow up on investigations of the anomalies, and imposing work orders to resolve any faults identified during the investigations.

[0031] [Diagram 3] FIG. 3 is a flow chart of an exemplary process for predicting anomalies in a subsystem of a system.

[0032] [Figure 4] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Diagram 5]4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 6] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 7] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 8] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 9] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 10] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 11] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 12] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 13]4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 14] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 15] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 16] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 17] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts. [Figure 18] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts.

[0033] [Figure 19] FIG. 19 diagrammatically illustrates a model that may be used to represent a processing pipeline within the application depicted in FIGS. 4-18.

[0034] [Figure 20] FIG. 20 illustrates a computer system that is programmed or otherwise configured to implement the methods provided herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0035] Detailed Description While various embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions may occur to those skilled in the art without departing from the invention. It is understood that various alternatives to the embodiments of the invention described herein may be employed.

[0036] The term "asset" as used herein generally refers to a single piece of equipment (eg, a valve) in a process pipeline.

[0037] The term "system" as used herein generally refers to a collection of subsystems that perform chemical reactions or physical operations in a manufacturing process. The relationships between the systems can be linear.

[0038] The term "subsystem" as used herein generally refers to a collection of assets or smaller subsystems that perform the chemical reactions or physical operations that occur on the materials in the process. The relationships between the subsystems can be linear or hierarchical.

[0039] The term "platform" as used herein generally refers to large offshore structures for exploring for, extracting, and processing oil and natural gas from beneath the ocean floor.

[0040] The term "plant" as used herein generally refers to a facility where industrial manufacturing processes occur.

[0041] The term "tag" as used herein generally refers to an individual data series that is associated with a measurement (eg, pressure, temperature, or flow rate) on an asset.

[0042] The term "cluster" as used herein generally refers to a correlated group of tags that are associated with a single subsystem.

[0043] Each time the term "at least," "greater than," or "greater than or equal to" precedes a first number in a series of two or more numerical values, the term "at least," "greater than," or "greater than or equal to" applies to each and every numerical value in the series. For example, "greater than or equal to 1, 2, or 3" is equivalent to "greater than or equal to 1," "greater than or equal to 2," or "greater than or equal to 3."

[0044] Each time the term "less than", "less than", or "less than or equal to" precedes a first number in a series of two or more numerical values, the term "less than", "less than", or "less than or equal to" applies to each and every numerical value in the series. For example, "less than or equal to 3, 2, or 1" is equivalent to "less than or equal to 3", "less than or equal to 2", or "less than or equal to 1".

[0045] The present disclosure provides systems, methods, and computer program products for predicting and detecting anomalies in a system or subsystem of a processing pipeline and improving reliability of the processing pipeline. An exemplary system as described herein can determine a first plurality of tags indicative of an operational performance of the subsystem. The first plurality of tags can be time series data from sensors in the subsystem and sensors that are not in the subsystem but are nevertheless indicative of the operational performance of the subsystem. The system can determine that the first plurality of tags is indicative of the operational performance of the subsystem, for example, by using a clustering algorithm. Determining the tags can be considered a feature selection operation.

[0046] The system can process measurements of the first plurality of tags using an autoencoder. The autoencoder can have an encoder configured to generate a reduced dimensional or compressed output from the measurements, and a decoder configured to generate estimates of the first plurality of tags. The autoencoder can be trained based on historical values ​​of the first plurality of tags.

[0047] The system can determine whether a difference exists between measurements of the first plurality of tags provided to the autoencoder and estimates of the first plurality of tags generated by the autoencoder. If the measurements and estimates are identical or similar, the system may determine that the subsystem is not suffering from an anomaly. On the other hand, if the system determines that the difference between the measurements and estimates satisfies a threshold, the system may determine that the subsystem is suffering from an anomaly. The system can transmit an alert to one or more computing devices having an application (e.g., a web application, a mobile application, or a dedicated software program) indicating that the system is suffering from such an anomaly. The application can enable a user, e.g., an engineer or operations manager, to monitor and investigate the alert, generate a case to track action items related to the alert, and generate work orders to address the malfunction underlying the alert.

[0048] FIG. 1 diagrammatically illustrates a system for predicting and detecting anomalies. The system may have a process pipeline 100. The process pipeline 100 may be located in a manufacturing facility, e.g., a plant. Alternatively, the process pipeline 100 may be distributed among assets in different locations, connected by conduits. The process pipeline 100 may be configured to produce one or more products from one or more raw materials. The process pipeline 100 may implement a batch, semi-batch, or continuous process. A batch process may be a process that produces a definite quantity of a product. A continuous process may be a process that consumes raw materials continuously while producing an indefinite quantity of a product.

[0049] The processing pipeline 100 may be an oil and gas processing pipeline. Alternatively, the processing pipeline 100 may be a chemical or material synthesis pipeline (e.g., a pipeline for the synthesis of paints, fertilizers, textiles, rubber, steel, or aluminum), a drug synthesis pipeline, or a food or beverage production pipeline.

[0050] Processing pipeline 100 may have system 111. System 111 may have subsystem 112 and subsystem 116. Each subsystem may have individual components or portions of assets, such as equipment (not depicted) and sensors. One example of a system in an oil and gas processing pipeline is a crude distillation unit. A crude distillation unit may include a coker, a hydrocracker, a catalytic converter, a hydrotreater, a reformer, and a desulfurizer subsystem, each of which may have assets. Although processing pipeline 100 is depicted as having only a single system, processing pipeline 100 may have multiple systems, each with multiple subsystems.

[0051] The sensors in the processing pipeline 100 may be sensors that generate time series data. The sensors may be thermometers, pressure gauges, flow meters, accelerometers, magnetometers, and the like. In addition to such time series data, the sensors may generate metadata, such as data about the location of the sensor in the system 111 or subsystems 112 and 166. The sensors may collect data at a rate of at least about 0.1 Hz, 0.2 Hz, 0.3 Hz, 0.4 Hz, 0.5 Hz, 1 Hz, 5 Hz, 10 Hz, 100 Hz, 1,000 Hz, 10,000 Hz, 100,000 Hz, or more. The sensors in the processing pipeline 100 may transmit the time series data, referred to as "tags" in this disclosure, to one or more computing devices that implement anomaly prediction and detection. Although the processing pipeline 100 is depicted as having only six sensors, it may have tens of thousands of sensors.

[0052] 1 may include a training subsystem 120. The training subsystem 120 may be configured to train machine learning models to predict and detect anomalies.

[0053] Training subsystem 120 may include a cluster module 122. Cluster module 122 may be configured to train and execute a clustering algorithm to determine groups of tags from sensors 113-115 and 117-119 that are indicative of the operational performance of a subsystem in processing pipeline 100. Each group of tags can be correlated with the operational performance of a particular subsystem.

[0054] The clustering algorithm may be a hierarchical clustering algorithm. A hierarchical clustering algorithm is a clustering algorithm that clusters objects based on their proximity to other objects. For example, a hierarchical clustering algorithm may cluster tags from the processing pipeline 100 based on the physical proximity of their sensors to other sensors and related subsystems. The clustering algorithm may alternatively be a centroid-based clustering algorithm, such as a k-means clustering algorithm. A k-means clustering algorithm may partition n observations into k clusters (each observation belongs to the cluster with the closest mean value). The mean value may serve as a prototype for the clusters. In the context of tags from the processing pipeline 100, a k-means clustering algorithm may generate distinct groups of tags that are correlated with each other. Each group of tags may then be associated with a particular subsystem based on prior knowledge of that subsystem, such as knowledge of sensors within the subsystem. The clustering algorithm may alternatively be a variance-based clustering algorithm, such as a Gaussian mixture model or an expectation-maximization algorithm. Examples of other clustering algorithms that the clustering module 122 may implement are a cosine similarity algorithm, a topological data analysis algorithm, and Hierarchical Density-Based Spatial Clustering for Applications with Noise (HDB-SCAN).

[0055] The training subsystem 120 may additionally include an autoencoder training module 124. The autoencoder training module 124 may be used to train the autoencoder 132. The autoencoder training module 124 may train the autoencoder 132 based on historical values ​​of tags indicative of the operational performance of a selected subsystem. The historical tags may be tags collected during normal operation of the subsystem, e.g., during a period during which no faults occurred in the subsystem. Training the autoencoder 132 based on normal tags may result in the autoencoder 132 having parameters that represent "normal" characteristics. Although FIG. 1 depicts only a single autoencoder, each subsystem in the system may have its own autoencoder that may be trained based on historical values ​​of tags indicative of the operational performance of that particular subsystem.

[0056] The autoencoder 132 may have an encoder 133 that encodes measurements of a first plurality of tags determined to be indicative of the operational performance of a particular subsystem in the processing pipeline 100. The autoencoder 132 may also have a decoder 134 that attempts to reconstruct the measurements. Given measurements representing training data, the autoencoder 132 may generate estimates that are similar or equivalent to the measurements. However, given anomalous measurements, the autoencoder 132 may produce poorly reconstructed values.

[0057] The autoencoder 132 may be a neural network. In general, a neural network may employ multiple layers of operations to predict one or more outputs from one or more inputs. A neural network may include one or more hidden layers that are mounted between an input layer and an output layer. The output of each layer may be used as an input to another layer, for example, the next hidden layer or the output layer. Each layer of a neural network may define one or more transformation operations to be performed on the inputs to the layer. Such transformation operations may be referred to as neurons. The output of a particular neuron may be a weighted sum of the inputs to the neuron, adjusted with a bias and multiplied by an activation function, for example, a rectified linear unit (ReLU) or a sigmoid function. Training a neural network may involve providing inputs to an untrained neural network to generate a predicted output, comparing the predicted output to an expected output, and updating the weights and biases of the algorithm to account for the difference between the predicted and expected outputs. Specifically, a cost function may be used to calculate the difference between the predicted and expected outputs. By calculating the derivative of the cost function with respect to the weights and biases of the network, the weights and biases can be iteratively adjusted over multiple cycles to minimize the cost function. Training can be completed once the predicted outputs meet a convergence condition, such as obtaining a small calculated cost.

[0058] Neural networks can be trained using supervised or unsupervised learning processes. In supervised learning, the training inputs to the neural network are labeled; that is, the training inputs have known outputs. In contrast, in unsupervised learning, the training inputs are not labeled. Although the autoencoder training module 124 can implement an unsupervised learning process to train the autoencoder 132, the inputs to the autoencoder 132 do not need to be labeled, since the desired output of the autoencoder 132 is simply the input.

[0059] The autoencoder 132 can be a feed-forward neural network, i.e., a neural network with only forward connections between neurons. Alternatively, the autoencoder 132 can be a recurrent neural network ("RNN"). The RNN can have recurrent connections that can encode dependencies in the time-series data. The RNN can include an input layer configured to receive a sequence of time-series inputs, e.g., time-series tags from a sensor in the processing pipeline 100. The RNN can also include one or more hidden recurrent layers that maintain state. For each time step, each hidden recurrent layer can calculate an output and a next state for the layer. The next state can depend on the previous state and the current input. The state can be maintained across time steps and can capture dependencies in the input sequence. Such an RNN can be used to encode time-series features of the processing pipeline 100.

[0060] One example of an RNN is a long short-term memory (LSTM) network, which may be made up of LSTM units. LSTM units may be made up of cells, input gates, output gates, and forget gates. The cells may be responsible for tracking dependencies between elements in the input sequence. The input gates may control the extent to which new values ​​flow into the cells, the forget gates may control the extent to which values ​​remain in the cells, and the output gates may control the extent to which values ​​in the cells are used to calculate the output activation of the LSTM unit. The activation function of the LSTM gates may be a logistic function.

[0061] After being trained, the inference subsystem 130 can implement an autoencoder 132. If the difference between the measurements of the first plurality of tags provided to the autoencoder 132 and the estimates of the first plurality of tags generated by the autoencoder 132 exceeds a threshold, an anomaly may exist in the subsystem associated with the first plurality of tags. In such a case, the system can transmit an alert to the application 140, which will be described in more detail below.

[0062] The inference subsystem 130 can implement other types of machine learning models. For example, the inference subsystem 130 can implement a machine learning model configured to (i) optimize a threshold at which an anomaly is predicted or detected, (ii) predict a failure mode associated with an anomaly, or (iii) determine a corrective action. The machine learning model can be trained using a supervised, semi-supervised, unsupervised, or reinforcement learning process. Examples of such machine learning models are regression algorithms, decision trees, support vector machines, Bayesian networks, neural networks (e.g., feed-forward neural networks, CNNs, RNNs, and the like), clustering algorithms, reinforcement learning algorithms, and the like.

[0063] As mentioned above, the system of FIG. 1 may additionally include application 140. Application 140 may be a mobile application or a web application. Application 140 may be used on any type of computing device, e.g., a desktop or laptop computer, an electronic tablet, a mobile device, or the like. Users, e.g., engineers and operations managers, may use application 140 to review alerts and predicted failure modes generated by inference subsystem 130, generate cases to follow up investigations of anomalies, and generate work orders to address any malfunctions underlying the alerts. Application 140 will be described in more detail with reference to subsequent figures.

[0064] The system of FIG. 1 and its components can be implemented on one or more computing devices. The computing devices can be servers, desktop or laptop computers, electronic tablets, mobile devices, or the like. The computing devices can be located in one or more locations. The computing devices can have a general purpose processor, a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or the like. The computing devices can additionally have memory, e.g., dynamic or static random access memory, read only memory, flash memory, hard drives, or the like. The memory can be configured to store instructions that, upon execution, cause the computing devices to implement the functionality of the training subsystem 120, the inference subsystem 130, or the application 140. The computing devices can additionally have a network communication device. The network communication device can enable the computing devices to communicate over a network with each other and with any number of user devices, e.g., a user device that launches the application 140. For example, the inference engine 130 can transmit alerts, predicted failure modes, and the like to a user device invoking the application 140. The network can be a wired or wireless network. For example, the network can be an optical fiber network, an Ethernet network, a satellite network, a cellular network, a Wi-Fi network, a Bluetooth network, or the like. In other implementations, the computing device can be a number of distributed computing devices accessible through the Internet. Such computing devices can be considered cloud computing devices.

[0065] 2 is a flow chart of an exemplary process for predicting anomalies in a subsystem, generating alerts about the anomalies, generating cases to follow up on the investigation of the anomalies, and imposing work orders to resolve any faults identified during the investigation. The process can be performed by one or more users using one or more computer systems and applications at one or more locations, such as application 140 of FIG.

[0066] Initially, the system may predict or detect an anomaly in a subsystem (210). The prediction may be a qualitative prediction, a quantitative prediction, or both. A qualitative prediction may be a prediction that an anomaly will occur within a defined time period. A quantitative prediction may be a numerical risk score indicating the likelihood that an anomaly will occur within a defined time period. The prediction may additionally include one or more predicted failure modes associated with the anomaly. Operation 210 will be described in more detail with reference to FIG. 3.

[0067] The system can then transmit an alert to applications on any number of user devices indicating that the subsystem is or will be experiencing an anomaly (220). The alert identifies the affected system and subsystem and can include a qualitative or quantitative prediction, a predicted failure mode, and tags that contributed to the predicted or detected anomaly. The alert can serve as the basis for further investigation by a user using the application.

[0068] Finally, using the application, users can analyze alerts, investigate anomalies, and generate and manage cases and work orders to resolve them (230).

[0069] In one embodiment, the user can assign the alert to an authorized reviewer. The reviewer may be an engineer or an operations manager. The reviewer can provide feedback regarding the predicted failure mode. The feedback can be a text comment. Alternatively, or in addition, the reviewer can reject or accept the predicted failure mode, or select one or more new failure modes from a drop-down menu. If the user does not provide feedback regarding the predicted failure mode, the application may assume that the predicted failure mode is correct. Once the reviewer provides feedback, the system can then automatically generate a case that tracks all actions taken regarding the alert.

[0070] Cases may be distinct from alerts. The purpose of a case may be to track activity, coordinate across users, and enable collaboration. Cases may have a one-to-one or one-to-many relationship with alerts; that is, multiple related alerts may be addressed with one case in some cases. Cases may enable collaboration through tagging of other users and through comments and uploaded files that are available for editing or viewing by all users associated with the case. The application may enable users to take the following actions on a case: (1) assign the case and underlying alerts to a reviewer, (2) add comments to the case, (3) tag another user, (4) provide feedback regarding predicted failure modes, (5) generate associated work orders, and (5) close or reopen the case.

[0071] After providing feedback regarding the predicted failure mode, the reviewer can investigate the alert using alert analysis tools within the application, which will be described in more detail with reference to subsequent figures. Briefly, the reviewer can review specific tags to identify the root cause of the alert, e.g., a defective component. The reviewer can record comments about their findings that can become part of the case history. Based on the results of their investigation, the reviewer can generate a work order. Once the work order is completed, the application can automatically close the alert and the case.

[0072] The data collected during operation 230 can be used to train a machine learning algorithm to predict appropriate corrective actions for the detected anomalies. For example, the specific predicted anomalies, the tags that cause the predicted anomalies, and the corresponding failure modes selected by the user can serve as tagged data for training a supervised machine learning algorithm. As more training data is collected over time and incorporated into the algorithm, the performance of the algorithm can be improved.

[0073] 3 is a flow chart of an exemplary process for predicting anomalies in a subsystem of a system. The process of FIG. 3 can be implemented by one or more computer systems in one or more locations. For example, the training subsystem 120 and the inference subsystem 130 of FIG. 1 can implement the process of FIG. 3.

[0074] In the first operation, the system can determine a first plurality of tags that indicate the operating performance of the subsystem (310). The tags within the first plurality of tags can be (i) sensors within the subsystem and (ii) time series data from sensors within the system that are not within the subsystem but nevertheless indicate the operating performance of the subsystem. For example, the tags can be from another subsystem that provides an input to the subsystem, receives an output from the subsystem, or is otherwise correlated with the subsystem.

[0075] The system can use one or more machine learning algorithms to determine tags that indicate the operating performance of the subsystem. For example, the system can use a clustering algorithm, such as the k-means clustering algorithm, to process all of the tags within the system and generate clearly distinct groups of tags, including the first plurality of tags. The tags within each clearly distinct group may be correlated with the operating performance of the mutual or selected subsystem. Operation 310 can be regarded as a feature selection operation. Operation 310 can be performed offline, that is, prior to implementing the system within the processing pipeline.

[0076] In a second operation, the system can process measurements of a first plurality of tags (i.e., tags determined to be indicative of the subsystem's operation) using an autoencoder (320). The measurements can be limited to a predefined lookback horizon. The predefined lookback horizon can be at least about 1 second, 2 seconds, 3 seconds, 4 seconds, 5 seconds, 10 seconds, 30 seconds, 1 minute, 5 minutes, 10 minutes, 30 minutes, 1 hour, 5 hours, 10 hours, 15 hours, 20 hours, 1 day, 5 days, 10 days, or more. The system can dynamically adjust the lookback horizon to optimize the accuracy of the anomaly prediction. If the lookback horizon is too long, the measurements may contain irrelevant data. If the lookback horizon is too short, the measurements may lack important contextual information. The system can optimize the lookback horizon using any of the machine learning methods described in this disclosure.

[0077] The autoencoder may have an encoder configured to generate a reduced-dimensional or compressed output of measurements of the first plurality of tags. The autoencoder may additionally have a decoder configured to generate estimates of the first plurality of tags from the reduced-dimensional or compressed output. The autoencoder may be a deep neural network, e.g., a feed-forward neural network, a convolutional neural network, a recurrent neural network, or the like. The autoencoder may be trained based on historical values ​​of the first plurality of tags. In some cases, the historical values ​​may be collected during normal operation of the subsystem, e.g., during a period of time during which the subsystem did not suffer from anomalies. Training the autoencoder based on normal historical values ​​may result in the autoencoder having parameters representative of "normal" characteristics. When provided with measurements representative of the historical values ​​used for training, the autoencoder may generate estimates that are similar to or equivalent to the measurements. However, when provided with anomalous measurements, the autoencoder may produce poor estimates.

[0078] In a third operation, the system may determine whether a difference between the measurements of the first plurality of tags and the estimated values ​​of the first plurality of tags meets a threshold (330). If the system determines that the difference does not meet the threshold, this may indicate that the operational performance of the subsystem is normal, e.g., the subsystem is not experiencing an anomaly, the subsystem is operating within normal parameters, or the subsystem is producing a quality product. On the other hand, if the system determines that the difference does meet the threshold, this may indicate that the subsystem is experiencing an anomaly.

[0079] In some cases, the system may apply weighting to the tag measurements and estimates prior to determining whether the difference satisfies a threshold. The system may do so because some tags may be more highly correlated with the subsystem's operational performance than others. The system may apply greater weighting to such tags than tags that are less highly correlated with the subsystem's operational performance.

[0080] The system can dynamically adjust the threshold during operation of the system as new data becomes available and new anomalies arise. For example, the system can lower the threshold if the system predicts or detects one or more anomalies that are classified by the user as false positives. On the other hand, the system can raise the threshold if the system fails to predict or detect an anomaly. The system can optimize the threshold using any of the machine learning methods described herein.

[0081] In a fourth operation, the system may transmit an alert indicating that the system is experiencing an anomaly if the difference satisfies a threshold (340). The alert may additionally include a numeric risk score. The numeric risk score may indicate the severity of the predicted anomaly. The numeric risk score may be based, at least in part, on (i) the magnitude of the difference between the measurements of the first plurality of tags and the estimates of the first plurality of tags, or (ii) the amount by which the difference exceeds a threshold. The alert may additionally include a list of tags that contributed to the alert, the amount each tag contributed, and the values ​​of the tags over a lookback window. The alert may additionally include a predicted failure mode associated with the alert. The system may predict such failure modes, for example, using a machine learning classifier or any of the other machine learning methods described herein.

[0082] The alert can be an email or a text message. Alternatively, or in addition, the alert can be a notification within a mobile or web application, such as application 140 of FIG. 1. The alert can specify a quantitative risk score associated with the alert, a risk threshold, the systems and subsystems associated with the alert, and the date and time the alert was generated. As will be explained in more detail in subsequent figures, the application can be configured to allow the user to conduct an investigation of the alert.

[0083] In some cases, in addition to determining whether the differences between the measurements and the estimates of the first plurality of tags collectively satisfy a threshold, the system may determine whether the differences between individual measurements and estimates satisfy individual thresholds for the corresponding tags, and if so, transmit an alert indicating so. This may alert a user to irregular data points that may not yet be causing an anomaly. (User Interface)

[0084] 4-18 show application user interfaces that may enable a user to investigate alerts and generate and manage cases and word orders associated with such alerts.

[0085] 4 shows the home screen of the application. The home screen may display key performance indicators ("KPIs") related to the processing pipeline being monitored. The KPIs may include open (e.g., unresolved) alerts, unplanned subsystem events, maintenance costs against a maintenance budget, and total plant uptime, all over a selected period of time. The home screen may display in graphs the open alerts, unplanned subsystem events, and total plant uptime as a function of time.

[0086] The home screen may additionally display a grid showing the highest priority alerts in the processing pipeline. Each row in the grid may correspond to a different alert. The inference subsystem 130 of FIG. 1 may generate and transmit alerts to applications by implementing the process of FIG. 3. The columns in the grid may show each alert's name, originating subsystem, qualitative risk score (e.g., low, medium, or high), numeric risk score, date and time of occurrence, and status (e.g., case generated, unprocessed, etc.). The numeric risk scores may range, for example, from 0 to 100, with 100 indicating the highest risk anomaly. The numeric risk scores may be color-coded to indicate the predicted severity of the corresponding anomaly. For example, the most severe anomalies may have a red risk score, while less severe anomalies may have a yellow risk score.

[0087] The home screen may additionally display a stacked bar graph showing the systems in the processing pipeline that are causing the most alerts. The alerts in each bar may be color coded to show the number of alerts in each bar within each qualitative risk category (e.g., low, medium, or high). The stacked bar graph may be configured to show the systems in the processing pipeline that are causing the most alerts within yesterday, last week, last month, last three months, last six months, last year, or more.

[0088] The home screen may additionally display a grid showing the highest priority cases in the processing pipeline. Each row in the grid may correspond to a different case. An application or a user may raise a case in response to an alert. Columns in the grid may show each case's name, originating subsystem, priority, status (e.g., under review, assigned, etc.), assignee (e.g., person assigned to resolve the case), creation date, and last activity date. Cases may be assigned priority by the assignee. Alternatively, the application may automatically assign priorities to cases based on a number of factors. For example, the application may consider the importance of the underlying subsystem to the processing pipeline, the severity of the anomaly, and the estimated cost and time to resolve the anomaly.

[0089] The grid described above can be sorted alphabetically or numerically by any column.

[0090] The home screen may additionally have a navigation bar that may allow the user to navigate to the alerts screen, the cases screen, the systems screen, the analysis screen, and the settings screen.

[0091] Figure 5 shows the Alerts screen. A user can open the Alerts screen by selecting the Alerts icon in the navigation bar. The Alerts screen can allow a user to search and filter the processing pipelines within the processing pipeline. A user can view their own alerts (e.g., alerts assigned to the user, alerts the user is collaborating on, or alerts in which the user is mentioned) or all alerts in the processing pipeline.

[0092] The alerts screen can have an overview window that can display the total number of alerts for the user or in the processing pipeline. The overview window can additionally display the number of high risk alerts, medium risk alerts, and low risk alerts.

[0093] The alerts screen can have a search bar. The search bar can allow a user to search for alerts by keyword. The alerts screen can also have a filtering window. The filtering window can allow a user to filter the alerts returned as a result of a keyword search. A user can filter the alerts by status (e.g., open, case generated, or closed), triggered date, originating system or subsystem, or qualitative risk category of the alert.

[0094] The alerts screen can display a grid of alerts that satisfy the search and filtering criteria. Each row in the grid can correspond to a different alert. Columns in the grid can indicate each alert's name, originating subsystem, qualitative risk score (e.g., low, medium, or high), numeric risk score, date and time of occurrence, and status (e.g., case generated, outstanding, etc.).

[0095] 6 shows an alert details screen. The alert details screen can display more detailed information about a particular alert. The alert details screen can have an overview window that displays the name of the alert, its subsystem of origin, its qualitative risk score, its current quantitative risk score and quantitative risk score over time, the date it was triggered, and its status (e.g., new). The overview window can additionally have a button that is configured to generate a case or work order when selected by the user.

[0096] The alert details screen may display a grid of tags contributing to the alert. Each row in the grid may correspond to a different tag. Columns in the grid may show the name of each tag (e.g., temperature, pressure, vibration, etc.), the amount each tag contributes to the alert, and the value of each tag when the alert was generated. A user can expand each tag to reveal data quality and threshold alerts associated with the tag. A data quality alert may indicate that data from a sensor associated with the tag has been lost or corrupted. A threshold alert may indicate that a data value from a sensor associated with the tag exceeds a threshold.

[0097] The alert details screen can additionally display the most common failure modes associated with the alert and assist the user in resolving the alert.

[0098] FIG. 7 illustrates a Cases screen. The Cases screen can allow a user to search and filter cases in the processing pipeline. Cases can be distinct from alerts in that cases can support collaboration and traceability of an investigation, while alerts can simply be a method of notification. Additionally, in some situations, there may not be a one-to-one correspondence between alerts and cases. For example, one case may be used to resolve many related alerts.

[0099] The case screen can be similar in form to the alert screen. The case screen can allow a user to search and filter cases in the processing pipeline. A user can view their own cases (e.g., cases assigned to the user, cases the user is collaborating on, or cases in which the user has been mentioned) or all cases in the processing pipeline.

[0100] The case screen may have a search bar. The search bar may allow a user to search for cases by keyword. The case screen may also have a filtering window. The filtering window may allow a user to filter the cases returned as a result of a keyword search. A user may filter cases by status (e.g., unassigned, assigned, under review, closed), date created, originating system or subsystem, or assignee.

[0101] The Cases screen can display a grid of cases that meet the search and filtering criteria. Each row in the grid can correspond to a different case. Columns in the grid can indicate each case's name, originating subsystem, status, assignee, and the date and content of the last action taken on the case (e.g., case assigned, comments added, attachments uploaded, priority changed, etc.).

[0102] Figure 8 illustrates a case details screen. The case details screen can provide a collaboration and research workspace for users to work on a particular case.

[0103] The case details screen can display an overview window. The overview window can show the name of the case, the systems and subsystems associated with the case, the date of the last activity on the case, the priority of the case (e.g., as defined manually by the user when the case is started), the status of the case (e.g., alerts open, case created, work order created, work order completed, case closed), and collaborators on the case (e.g., the case owner and other users who have mentioned or commented on the case). The overview window can also have action buttons configured to allow the user to create a work order for the case, close the case, or edit the priority of the case.

[0104] The case details screen can additionally display a grid of actions associated with the case. Each row in the grid can correspond to a different action. Each row can indicate the collaborator who took the action, a summary of the action, a list of attachments (e.g., marked-up schematics, photos of subsystems associated with the case, etc.), and the time the action was taken.

[0105] The case details screen may additionally display an alert summary window. The alert summary window may show the top three tags that contributed to the alert, the systems or subsystems that may be at risk as a result of the alert, the date the alert was generated, a quantitative risk score for the alert, and a list of suggested failure modes typically associated with that type of alert. The list of suggested failure modes may be configured to allow the user to accept or reject the suggested failure modes, provide feedback regarding the suggested failure modes, or add new failure modes from a drop-down list.

[0106] Figure 9 shows the Subsystem Analysis tab of the Subsystem Detail screen. The Subsystem Detail screen can provide the user with the functionality to focus on a single subsystem within a processing pipeline or facility. The Subsystem Detail screen can display the subsystem's quantitative risk score over time, historical alerts, cases, and work orders associated with the system.

[0107] The Subsystem Analysis tab can display a grid of tags that contribute to the quantitative risk score for the subsystem. Each row in the grid can correspond to a different tag. Columns in the grid can show each tag's minimum, maximum, and average contribution to the quantitative risk score over a selected time period, as well as the tag's most recent measured value.

[0108] The Subsystem Analysis tab can additionally display a chart showing the contribution of the top three tags to the quantitative risk score of the subsystem over time. A filtering window adjacent to the chart can allow the user to select the tags that are displayed in the chart. For example, the user can specify which tags should be displayed and the time period for displaying them.

[0109] 10 shows the Cases tab of the Subsystem Detail screen. Although not depicted, the Subsystem Detail screen may additionally have an Alerts tab and a Word Order tab. These tabs may allow a user to search, filter, and view cases, alerts, and work orders for a particular subsystem. These tabs may be similar to the previously described screens, except that the cases, alerts, and word orders are limited to the particular subsystem.

[0110] 11 shows the Cluster tab of the Subsystem Details screen. The Cluster tab can display tags that have been identified by the system as being associated with the subsystem quantitative risk score, for example, by implementing the unsupervised clustering method described with reference to FIG.

[0111] 12 shows the Subsystem Diagram tab of the Subsystem Detail screen. The Subsystem Diagram tab can show a schematic diagram of the subsystem. The schematic diagram can be static, e.g., an image, or it can be dynamic, e.g., an interactive model with traversable links.

[0112] Figure 13 shows a Systems screen of the application. The Systems screen can provide a user of the application with a consolidated view of all systems in the processing pipeline. The Systems screen can include a grid that displays each system in a separate row. Columns in the grid can show a quantitative risk score for each system, the number of active alerts per system, the number of open cases per system, and the location of the system.

[0113] The summary window in the systems view can additionally show the number of systems or subsystems that have alerts.

[0114] 14 shows the Subsystems tab of the System Details screen. The Subsystems tab can show more detailed information about the subsystems within the system, including each subsystem's (i) quantitative risk score, number of active alerts, number of open cases, and the last activity for that subsystem (e.g., alert created, case started, alert closed, case closed, work order created, etc.).

[0115] 15 shows the Subsystem Alerts tab of the System Details screen, which may show more detailed information about alerts within the various subsystems of the system.

[0116] 16 shows the Subsystem Instances tab of the System Details screen. The Subsystem Instances tab may show more detailed information about the instances associated with the various subsystems of the system.

[0117] 17 shows the Work Orders tab of the System Details screen. The Work Orders tab can show more detailed information about the work orders associated with the various subsystems of the system.

[0118] 18 shows a data analysis screen for an application. The data analysis screen can provide a user with the functionality to plot any metric or time within an application case, such as tags, quantitative risk scores of subsystems, quantitative risk scores of systems, or the like.

[0119] FIG. 19 schematically illustrates a model that can be used to represent a processing pipeline within an application. The model can be an abstract representation of various materials, assets, processes, data, and the like. The model can be arranged in a hierarchical manner. For example, a plurality of process step models can be arranged in series or in parallel to represent a manufacturing process. As another example, asset models can be combined together to represent a subsystem, and subsystem models can be combined together to represent a system. Material models and data models can each represent the inputs and outputs of a process step or a manufacturing process model. The model can be defined by fields and functions. The values of the fields and functions can define the particular materials, assets, processes, and data that the model represents. (Computer system)

[0120] The present disclosure provides a computer system 2001 programmed to implement the methods of the present disclosure. FIG. 20 shows a computer system 2001 programmed to implement the system of FIG. 1 or otherwise configured, which can be a user's electronic device or a computer system located remotely from the electronic device. The electronic device can be a mobile electronic device.

[0121] The computer system 2001 includes a central processing unit (CPU, also referred to herein as "processor" and "computer processor") 2005, which may be a single-core or multi-core processor, or multiple processors for parallel processing. The computer system 2001 also includes memory or memory locations 2010 (e.g., random access memory, read-only memory, flash memory), an electronic storage unit 2015 (e.g., a hard disk), a communication interface 2020 (e.g., a network adapter) for communicating with one or more other systems, and peripheral devices 2025, such as cache, other memory, data storage devices, and / or electronic display adapters. The memory 2010, storage unit 2015, interface 2020, and peripheral devices 2025 communicate with the CPU 2005 through a communication bus (solid lines) such as a motherboard. The storage unit 2015 may be a data storage unit (or data repository) for storing data. The computer system 2001 may be operatively coupled to a computer network ("network") 2030 with the aid of the communication interface 2020. The network 2030 may be the Internet, an intranet and / or an extranet, or an intranet and / or an extranet in communication with the Internet. The network 2030 may, in some cases, be a telecommunications and / or data network. The network 2030 may include one or more computer servers, which may enable distributed computing such as cloud computing. The network 2030 may, in some cases, implement a peer-to-peer network, which may enable devices coupled to the computer system 2001 to behave as clients or servers with the aid of the computer system 2001.

[0122] The CPU 2005 can execute sequences of machine-readable instructions, which may be embodied in a program or software. The instructions may be stored in a memory location, such as the memory 2010. The instructions can be directed to the CPU 2005, which can then program or otherwise configure the CPU 2005 to implement the methods of the present disclosure. Examples of operations performed by the CPU 2005 can include fetch, decode, execute, and writeback.

[0123] The CPU 2005 may be part of a circuit, such as an integrated circuit. One or more other components of the system 2001 may be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0124] The storage unit 2015 can store files such as drivers, libraries, and saved programs. The storage unit 2015 can store user data, such as user preferences and user programs. The computer system 2001 can include one or more additional data storage units that are external to the computer system 2001, in some cases, such as located on a remote server that communicates with the computer system 2001 through an intranet or the Internet.

[0125] Computer system 2001 can communicate with one or more remote computer systems through network 2030. For example, computer system 2001 can communicate with a user's remote computer system (e.g., a user device running application 140 of FIG. 1). Examples of remote computer systems include a personal computer (e.g., a portable PC), a slate or tablet PC (e.g., Apple® iPad®, Samsung® Galaxy Tab), a phone, a smartphone (e.g., Apple® iPhone®, Android-enabled devices, Blackberry®), or a personal digital assistant. A user can access computer system 2001 via network 2030.

[0126] Methods as described herein can be implemented using machine (e.g., computer processor) executable code stored on electronic storage locations of the computer system 2001, such as, for example, on the memory 2010 or electronic storage unit 2015. The machine executable or machine readable code can be provided in the form of software. During use, the code can be executed by the processor 2005. In some cases, the code can be read from the storage unit 2015 and stored on the memory 2010 for quick access by the processor 2005. In some situations, the electronic storage unit 2015 can be omitted and the machine executable instructions are stored on the memory 2010.

[0127] The code can be pre-compiled and configured for use with a machine having a processor adapted to execute the code, or it can be compiled during run-time. The code can be provided in a programming language that can be selected to allow the code to be executed in a pre-compiled or immediately compiled manner.

[0128] Aspects of the systems and methods provided herein, such as computer system 2001, can be embodied in programming. Various aspects of the technology may be considered as a "product" or "article of manufacture" typically in the form of machine (or processor) executable code and / or associated data carried on or embodied in some type of machine-readable medium. The machine executable code may be stored on an electronic storage unit, such as a memory (e.g., read-only memory, random access memory, flash memory) or a hard disk. A "storage" type medium may include any or all of the tangible memory of a computer, processor, or equivalent, or its associated modules, such as various semiconductor memories, tape drives, disk drives, and the like, that may provide non-transitory storage at any time for software programming. All or portions of the software may be communicated from time to time over the Internet or various other telecommunications networks. Such communication may, for example, enable loading of the software from one computer or processor to another, for example, from a management server or host computer to a computer platform of an application server. Thus, other types of media that may bear software elements include optical, electrical, and electromagnetic waves, such as those used across physical interfaces between local devices, through wired and optical fixed networks, and via various air links. Physical elements that carry such waves, such as wired or wireless links, optical links, or the like, may also be considered media bearing the software. As used herein, unless limited to non-transitory tangible "storage" media, terms such as computer or machine "readable medium" refer to any medium that participates in providing instructions to a processor for execution.

[0129] Thus, a machine-readable medium such as a computer executable code may take many forms, including, but not limited to, a tangible storage medium, a carrier wave medium, or a physical transmission medium. Non-volatile storage media include optical or magnetic disks, such as any of the storage devices in any computer or equivalent, such as those that may be used to implement the databases, etc., shown in the figures. Volatile storage media include dynamic memory, such as the main memory of such a computer platform. Tangible transmission media include coaxial cables, copper wire and optical fibers, including the wires that comprise a bus in a computer system. Carrier wave transmission media may take the form of electric or electromagnetic signals, or acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer readable media thus include, for example, floppy disks, flexible disks, hard disks, magnetic tape, any other magnetic media, CD-ROMs, DVDs or DVD-ROMs, any other optical media, punch cards, paper tape, any other physical storage media with patterns of holes, RAM, ROM, PROMs and EPROMs, FLASH-EPROMs, any other memory chips or cartridges, carrier waves that transmit data or instructions, cables or links that transmit such carrier waves, or any other medium from which a computer may read programming code and / or data. Many of these forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0130] The computer system 2001 may include or communicate with an electronic display 2035 that has a user interface (UI) 2040 for providing applications, such as those described with reference to Figures 4-18. Examples of UIs include, but are not limited to, graphical user interfaces (GUIs) and web-based user interfaces.

[0131] The methods and systems of the present disclosure can be implemented using one or more algorithms. The algorithms can be implemented using software upon execution by the central processing unit 2005. The algorithms can be, for example, algorithms that implement the process of FIG. 3 for predicting anomalies in a subsystem of the system.

[0132] While preferred embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. It is not intended that the present invention be limited by the specific examples provided herein. Although the present invention has been described with reference to the above specification, the descriptions and illustrations of the embodiments herein are not meant to be construed in a limiting sense. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the present invention. Furthermore, it should be understood that all aspects of the present invention are not limited to the specific depictions, configurations, or relative proportions set forth herein, which depend upon a variety of conditions and variables. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention. It is therefore contemplated that the present invention also covers any such alternatives, modifications, variations, or equivalents. It is intended that the following claims define the scope of the present invention, and that methods and structures within the scope of these claims and their equivalents are covered thereby.

Claims

1. 1. A method comprising: determining a first plurality of tags indicative of operational performance of a subsystem of a system, each tag of the first plurality of tags including a data sequence associated with a time period, the first plurality of tags being obtained from at least (i) a plurality of sensors in the subsystem, and (ii) a plurality of sensors in the system that are not in the subsystem; processing measurements of the first plurality of tags with an autoencoder, the autoencoder generating compressed representations of the measurements and decoding the compressed representations of the measurements into estimates of the first plurality of tags, the autoencoder being trained based on historical values ​​of the first plurality of tags; determining a difference between the measured value and the estimated value, taking into account a threshold value; predicting an anomaly in the operation of the subsystem based at least in part on a magnitude of a difference between the measurements of the first plurality of tags and the estimated values; providing one or more corrective actions with respect to the subsystem based at least in part on the prediction; and A method comprising:

2. 2. The method of claim 1 , wherein determining the first plurality of tags includes using an unsupervised learning algorithm to process data from sensors in the system to generate a plurality of populations of tags, the plurality of populations of tags including the first plurality of tags indicative of the operational performance of the subsystem.

3. The method of claim 2 , wherein each tag of the first plurality of tags is correlated with the operational performance of at least one other tag of the first plurality of tags or the subsystem.

4. 3. The method of claim 2, wherein the unsupervised learning algorithm is a clustering algorithm, the clustering algorithm comprising a k-means clustering algorithm, a cosine similarity algorithm, a topological data analysis algorithm, or a Hierarchical Density-Based Spatial Clustering for Applications with Noise (HDB-SCAN).

5. The method of claim 1 , wherein determining the first plurality of tags is performed offline.

6. The method of claim 1 , further comprising applying a weighting process to each of the measured value and the estimated value prior to determining the difference between the measured value and the estimated value.

7. The method of claim 1 , further comprising dynamically adjusting the threshold during operation of the system.

8. 2. The method of claim 1, further comprising: determining whether a difference between the measurement value of an individual tag and the estimated value of the individual tag satisfies a tag threshold; and transmitting an alert if the difference satisfies the tag threshold.

9. The method of claim 1 , wherein the autoencoder is a deep neural network or a support vector machine (SVM).

10. The method of claim 1 , wherein the historical data comprises a data series collected during a period during which no faults occurred within the subsystem.

11. The method of claim 1 , wherein the system is one of a plurality of systems in a material processing or manufacturing pipeline.

12. The method of claim 1 , wherein the time period comprises a predetermined lookback horizon.

13. The method of claim 12 , further comprising dynamically adjusting the lookback horizon.

14. 10. The method of claim 1, further comprising generating a risk score for the subsystem based at least in part on a magnitude of the difference between the measured values ​​of the first plurality of tags and the estimated values ​​of the first plurality of tags.

15. The method of claim 1 , further comprising transmitting an alert indicating that the subsystem is predicted to experience the anomaly if the difference satisfies the threshold.

16. 2. The method of claim 1 , wherein providing the one or more corrective actions is performed using a trained machine learning algorithm, the trained machine learning algorithm being trained based on historical actions and corresponding responses by the subsystem.

17. 17. The method of claim 16, further comprising optimizing the threshold at which the anomaly is predicted or detected prior to providing the one or more corrective actions, and predicting a failure mode associated with the anomaly using the trained machine learning algorithm.

18. 10. The method of claim 1 , wherein providing the one or more corrective actions is performed using rules-based logic to correlate observed patterns in tag measurements to known failure mechanisms for processes and equipment in the subsystem.

19. further comprising, after predicting the abnormality, providing a numerical risk score indicative of the likelihood that the abnormality will occur, the numerical risk score being based at least in part on the magnitude of the difference between the measured values of the first plurality of tags and the estimated values of the first plurality of tags, or the amount by which the difference exceeds the threshold value, the method of claim 1.

20. The method of claim 15, wherein the alert includes a list of tags among the first plurality of tags that contributed most to the alert.

21. The autoencoder is trained based on a dataset comprising historical values representative of normal operation of the subsystem, the autoencoder being such that the estimated values are similar to or equivalent to the measured values generated by the decoder when the plurality of tags are captured during normal operation of the subsystem, and the estimated values are poor estimates of the measured values generated by the decoder when the plurality of tags are captured during abnormal operation of the subsystem, the method of claim 1.

Images