Automated construction and deployment of predictive models for plant assets

Abstract

Systems and methods provide users with a way to deploy predictive diagnostics for industrial assets utilizing existing data-driven models, components, and functions. The systems and methods allow users to retrieve existing models, components, and functions and assemble them into separate packages or containers called "analytics avatars" that can be saved and stored as independent predictive diagnostic units or entities. The analytics avatars can then be deployed on almost any analytics platform, providing predictive diagnostics for industrial assets. The ability to utilize existing predictive diagnostic models and components across different platforms between plants for the same type of equipment provides consistency and reduces the time and effort required to develop and deploy predictive diagnostics.

Description

[0001] Cross-reference to related applications

[0002] This application claims the benefit and priority of U.S. Provisional Application No. 62 / 927,998, filed October 30, 2019, which is incorporated herein by reference in its entirety. Technical Field

[0003] The embodiments of this disclosure generally relate to using predictive diagnostics to monitor industrial assets, and more specifically, to systems and methods for the efficient construction and deployment of predictive diagnostics for such industrial assets. Background Technology

[0004] Predictive diagnostics in factories typically refers to pattern recognition tools and procedures that monitor industrial assets (e.g., equipment, instruments, sensors, etc.) to provide early warnings of problems. This predictive diagnostics, sometimes also called predictive analytics or prediagnosis, usually uses industrial asset models derived from historical data to infer changes in asset operation. Machine learning algorithms are used to process historical data, identifying or “learning” relationships between the asset’s inputs, outputs, and operating conditions. These relationships are then used to build models that can simulate the operational profile or behavior of a particular industrial asset. Real-time or real-time data from the industrial asset can then be applied to the model in real-time (or near real-time) to infer asset changes that may lead to failure.

[0005] Predictive diagnostics has proven highly beneficial for industrial asset management, particularly for plants with complex equipment such as turbines. However, providers of predictive diagnostics for industrial assets tend to develop each product / solution from scratch for a given asset, rather than reusing previously developed products / solutions or parts thereof. This approach often results in multiple products / solutions being developed separately for the cloud, edge, and / or for embedding in the equipment, providing essentially the same predictive diagnostics for essentially the same type of industrial asset.

[0006] Therefore, despite the many advances made in the field of predictive diagnostics of industrial assets, it is easy to understand that continuous improvement is still needed. Summary of the Invention

[0007] This disclosure generally relates to systems and methods for effectively building and deploying predictive diagnostics to detect faults, etc., in plant assets. The system and methods provide users with a way to efficiently assemble existing data-driven models, components, and functionalities into packages or containers called “analysis avatars,” which can be deployed across multiple different platforms. Each analysis avatar can be represented graphically using icons or images that can be easily manipulated, or as a file referenced using a filename and appropriate file extension, or both. The ability to leverage existing predictive diagnostic models and components for the same type of equipment across different platforms in a plant provides consistency and reduces the time and effort required to develop and deploy predictive diagnostics.

[0008] In one aspect, this disclosure relates to a system for providing predictive diagnostics for industrial assets. The system includes at least one processor and a storage system communicatively coupled to the at least one processor. The storage system stores instructions thereon that, when executed by the at least one processor, cause the system to specifically perform a process of selecting an analytical avatar from a plurality of analytical avatars in a repository, each analytical avatar being configured, independently of a computing system, to provide a corresponding predictive diagnostic function for a corresponding asset of a corresponding type. When executed by the at least one processor, the instructions also cause the system to perform a process of deploying the analytical avatar to a first computing system, whereby the analytical avatar operates to provide the given predictive diagnostic function. When executed by the at least one processor, the instructions further cause the system to perform a process of deploying the analytical avatar to a second computing system different from the first computing system, whereby the analytical avatar operates to provide the given predictive diagnostic function.

[0009] In general, on another aspect, this disclosure relates to a method for providing predictive diagnostics for industrial assets. The method specifically includes selecting a first analytical avatar from a plurality of analytical avatars in a repository, each of the plurality of analytical avatars being configured, independently of a computing system, to provide a corresponding predictive diagnostic function for a corresponding asset of a corresponding type. The method also includes deploying the first analytical avatar to a first computing system, the first analytical avatar running on the first computing system to provide a given predictive diagnostic function, and deploying the first analytical avatar to a second computing system different from the first computing system, the first analytical avatar running on the second computing system to provide the given predictive diagnostic function.

[0010] In another aspect, this disclosure relates to a system for providing predictive diagnostics for industrial assets. The system specifically includes a repository on which multiple analytical avatars are stored, each analytical avatar configured to provide corresponding predictive diagnostic functionality for a corresponding type of asset independently of a computing system. The system also includes a studio platform connected to the repository, the studio platform having multiple avatar tools. The multiple avatar tools are configured to select a first analytical avatar from the multiple analytical avatars in the repository and deploy the first analytical avatar to a first computing system, where the first analytical avatar runs to provide a given predictive diagnostic functionality. The multiple avatar tools are also configured to deploy the first analytical avatar to a second computing system, different from the first computing system, where the first analytical avatar runs to provide a given predictive diagnostic functionality.

[0011] According to any one or more of the foregoing embodiments, multiple workers are selected from the storage library and these workers are interconnected to build a data processing pipeline, with each worker performing a different task in the data processing pipeline.

[0012] According to any one or more of the foregoing embodiments, multiple workers and data processing pipelines are containerized to form a second analytics avatar, which is configured to provide diagnostic functionality different from that of the first analytics avatar. According to any one or more of the foregoing embodiments, the second analytics avatar is stored in a repository for subsequent use and reuse.

[0013] According to any one or more of the foregoing embodiments, the first analytics avatar provides a given predictive diagnostic function for a given asset of a given type and is loosely coupled to a data stream from the given asset of a given type. According to any one or more of the foregoing embodiments, the first analytics avatar provides a given predictive diagnostic function for a given asset of a given type and is coupled to a digital twin of the given asset of a given type.

[0014] According to any one or more of the foregoing embodiments, the result of a given predictive diagnostic function of the first analytical avatar is output to a monitoring and control system, wherein the monitoring and control system takes a specified corrective action based on the result of the given predictive diagnostic function. Attached Figure Description

[0015] A more detailed description of the present disclosure, which has been briefly outlined above, can be obtained by referring to various embodiments, some of which are illustrated in the accompanying drawings. While the drawings show preferred embodiments of the present disclosure, they should not be considered as limiting its scope, as the present disclosure may acknowledge other equally effective embodiments.

[0016] Figure 1 This is a block diagram illustrating an exemplary industrial scene according to embodiments of the present disclosure;

[0017] Figure 2 This is a block diagram illustrating an exemplary analytical embodiment according to embodiments of the present disclosure;

[0018] Figure 3 This is a block diagram illustrating exemplary data ingested by the analysis avatar according to embodiments of the present disclosure;

[0019] Figure 4 This is a block diagram illustrating an exemplary worker of an analytical embodiment according to an embodiment of the present disclosure;

[0020] Figure 5 This is a block diagram illustrating an exemplary motor fall detection avatar according to an embodiment of the present disclosure;

[0021] Figure 6 This is a block diagram illustrating an exemplary data source of an analytical embodiment according to an embodiment of the present disclosure;

[0022] Figure 7 This is a block diagram illustrating an exemplary data ingestion of an analytical avatar according to embodiments of the present disclosure;

[0023] Figure 8 This is a schematic diagram illustrating an exemplary IoT analytics deployment according to embodiments of the present disclosure;

[0024] Figure 9 This is a block diagram illustrating exemplary stages of an analytical embodiment according to embodiments of the present disclosure;

[0025] Figure 10 This is a block diagram illustrating an exemplary avatar studio according to embodiments of the present disclosure;

[0026] Figure 11 An exemplary workflow for constructing an analytics avatar according to embodiments of this disclosure is illustrated;

[0027] Figure 12 This is a functional block diagram of a general-purpose computer system that can be used to implement various embodiments of the present disclosure; and

[0028] Figure 13 This is a functional block diagram of a general storage system that can be used to implement various embodiments of the present disclosure.

[0029] Where possible, the same reference numerals are used to denote the same common elements in the drawings. However, elements disclosed in one embodiment may be advantageously used in other embodiments without specific description. Detailed Implementation

[0030] This specification and accompanying drawings illustrate exemplary embodiments of this disclosure and should not be construed as limiting. The claims define the scope of this disclosure, including equivalents. Various mechanical, compositional, structural, electrical, and operational changes may be made without departing from the scope of this specification and the claims (including equivalents). In some cases, well-known structures and techniques have not been shown or described in detail so as not to obscure this disclosure. Furthermore, elements and related aspects thereof described in detail with reference to one embodiment may be included in other embodiments where they are not specifically shown or described, as long as practicable. For example, if an element is described in detail with reference to one embodiment but not with reference to a second embodiment, that element may still be claimed to be included in the second embodiment.

[0031] It should be noted that, as used in this specification and the appended claims, the singular forms “a,” “an,” and “the,” and any singular use of any word, include plural references unless expressly and unambiguously limited to a single reference. As used herein, the term “comprising” and its grammatical variations are intended to be non-restrictive, such that the listing of items does not exclude other similar items that may substitute for or be added to the listed items.

[0032] As previously described, embodiments of this disclosure provide systems and methods for containerizing or packaging core analytics components and distributing containers or packages as “analysis avatars.” Each “analysis avatar,” or simply “avatar,” includes a selection or bundle of models, modules, algorithms, and other software components that work together to provide specific predictive diagnostic capabilities for a given type of industrial asset. These models, modules, algorithms, and other software components can be selected from existing registries or repositories of models, modules, algorithms, and components to provide the desired predictive diagnostic capabilities. Analytics avatars can then be deployed across multiple different computing platforms, including cloud, edge devices, and embedded devices such as embedded programmable logic controllers (PLCs). This arrangement provides a platform-independent way to package predictive diagnostic software components and their runtime dependencies for efficient and convenient use and reuse.

[0033] An avatar operates in a manner very similar to a predictive diagnostics solution / product built from scratch to provide predictive diagnostic capabilities. Therefore, an avatar can process the input data stream of a given industrial asset, applying machine learning (ML), artificial intelligence (AI), statistical algorithms, and other data analytics techniques to the asset and providing useful insights about it. These insights can include predicting operational anomalies, failures, time to failure, remaining useful life, and more. Depending on the deployment, the avatar's data processing or data predictor portion (i.e., where the computation occurs) can be functionally set up or fixed, while the avatar's data access and data storage portions can be variable and open. This variability provides the avatar with the flexibility to run on assets across multiple different platforms (e.g., edge, cloud, embedded devices, etc.).

[0034] In some embodiments, the avatar platform or avatar studio provides a toolset that can be used to consistently assist in the building, development, training, deployment, management, and maintenance of avatars. Once an avatar for a specific device has been built, trained, tested, and validated, it can be uploaded to an avatar registry or repository. Users can search and download desired avatars from the repository, then retrain and reattach the avatar to another device of the same type, or a data stream representing that device (i.e., an instance of that device), to provide insights about that device. This provides users with a simple and effective way to obtain predictive diagnostic capabilities without having to build or construct such capabilities from scratch.

[0035] In some embodiments, the core analytics components described above are provided in the form of “workers” that perform specific tasks required to achieve predictive diagnostic functionality. Examples of workers may include data transformation workers, data cleaning workers, decision tree predictors, etc. Multiple task-specific workers can then be combined with each other to build an avatar’s data processing pipeline. These workers can be written in any suitable programming language, such as Python, R, MATLAB, etc., and then stored in a repository for later use and reuse. The saved and stored workers can then be selected and dragged and dropped using a toolset from the avatar workshop described above to form a data processing pipeline. In some embodiments, each worker may have a configuration file or manifest associated with it, and workers can be configured by editing their respective configuration files or manifests using an appropriate editor (e.g., a text editor, an XML editor, etc.). Workers, pipelines, and manifests, along with their respective runtime execution environments, can then be containerized using appropriate container technologies to create or package the avatar.

[0036] Avatar Studios can help bundle the runtime execution environment of avatars for deployment. Containerized avatars can be configured for their data access and storage aspects and can be deployed on multiple analytics platforms (e.g., cloud, virtual machines (VMs), edge devices, etc.). The same avatars can also be deployed on embedded platforms such as PLCs. Avatars can be trained offline and subsequently deployed to computing platforms, or they can learn on their own while processing streaming data. Developed avatars can be trained and tested for device instances, and, depending on accuracy satisfaction or QoS, the trained and tested avatars, along with their runtime dependencies, can be packaged in a single container or multiple containers as needed. The avatar then begins to provide predictive insights about the device it is bound to. When another similar type of device requires similar predictive diagnostics, the trained avatar from the repository can be retrained as needed with data from the new device. The retrained avatar can then be bound, bundled, or otherwise deployed to instances of the new equipment.

[0037] Now for reference Figure 1 This illustration shows an exemplary industrial site 100 or a portion thereof according to embodiments of the present disclosure. The industrial site 100 has multiple industrial assets installed on-site and associated with one or more specific industrial processes. Industrial processes may include chemical processes, manufacturing processes, assembly processes, extraction processes, and various other types of industrial processes. Figure 1 In the example, only compressor 102, pump 104, and motor 106 are shown at industrial site 100. Within the scope of this disclosure, other industrial assets may, of course, be present at industrial site 100. Compressor 102 is a specific type of compressor, such as a centrifugal compressor, scroll compressor, reciprocating compressor, etc., referred to as type X for economic reasons. Similarly, pump 104 is a specific type of pump, such as a hydraulic pump, piston pump, electric semi-submersible pump, etc., referred to as type Y. Motor 106 is also a specific type of motor, such as a series motor, split motor, repulsion motor, etc., referred to as type Z.

[0038] Compressor 102, pump 104, motor 106, and other industrial assets may each include or be coupled to one or more sensors and instruments that monitor and acquire various data about the assets and their operation. The sensors and instruments for compressor 102, pump 104, and motor 106 are typically described herein as data acquisition units 108, 110, and 112, respectively. These data acquisition units 108, 110, and 112 may be configured to sample, sense, or monitor one or more parameters (e.g., industrial parameters) associated with the industrial assets. For example, one or more temperature sensors may be coupled to compressor 102 to obtain temperature data at the compressor's inlet and / or outlet. Other types of data may include pressure data, flow data, vibration data, and any other data associated with the industrial assets and / or one or more industrial processes.

[0039] Edge device 114 provides data acquisition units 108, 110, and 112 with access to the enterprise network 116 to transmit collected data to downstream computing nodes, including local computing node 118 and cloud computing node 120. Communication can be conducted via any suitable wired or wireless communication link (e.g., Ethernet, Wi-Fi, Bluetooth, GPRS, CDMA, satellite, etc.) or a combination thereof. Edge device 114, which is also a computing node, can be any type of edge device or apparatus, provided that it has sufficient processing power for the purposes discussed herein. Examples of suitable edge devices include gateways, routers, routing switches, integrated access devices (IADs), and various MAN and WAN access devices.

[0040] According to embodiments of this disclosure, edge device 114 has one or more analytics avatars 122 deployed thereon. The analytics avatars 122 are specifically designed to interact with… Figure 1 The compressor 102, pump 104, and / or motor 106 shown operate together. In an alternative embodiment, instead of edge device 114, one or more analytics avatars 122 may be deployed on one of the local computing nodes 118 or one of the cloud computing nodes 120. A portion of an analytics avatar 122 may also be distributed across two or more of the edge device 114, local computing nodes 118, and cloud computing nodes 120. Furthermore, although not in Figure 1 As explicitly shown, but in some embodiments, one or more analytics avatars 122 may also be deployed in embedded devices, such as PLCs, within compressor 102, pump 104, and / or motor 106.

[0041] Once deployed, one or more analytics avatars 122 operate on compressor 102, pump 104, and / or motor 106 or instances thereof to provide predictive insights. As previously described, such insights may include predictive operational anomalies, failures, time to failure, remaining useful life, and other predictive diagnostic capabilities. The predictive insights and forecasts generated by the one or more analytics avatars 122 can then be transmitted to asset monitoring and control system 124 for taking any actions that may be necessary based on the provided predictive insights, such as corrective actions (e.g., cutting off power, reducing speed, issuing alarms, etc.). Asset monitoring and control system 124 can be any system capable of ingesting the predictive insights and forecasts generated by the one or more analytics avatars 122, including internal and third-party systems.

[0042] Figure 2 More details are shown from Figure 1 A simplified view of an exemplary analytics avatar 122. As can be seen in this example, analytics avatar 122 includes a compressor fault detection avatar 200 for an X-type compressor, a pump fault detection avatar 202 for a Y-type pump, and a motor fault detection avatar 202 for a Z-type motor. Each analytics avatar 200, 202, 204 contains core analytics components (e.g., ML, AI, neural network algorithms, etc.) designed to operate with its respective industrial asset, namely compressor 102, pump 104, and motor 106 (or instances thereof). These analytics avatars 200, 202, 204 are atomic, autonomous packages or bundles of core analytics components that can run on any platform, including in the cloud, on edge devices, or in embedded devices.

[0043] Data ingestion process 206 loosely couples various analytical avatars 200, 202, 204 to their respective industrial assets. Similarly, data emission process 208 loosely couples the various analytical avatars 200, 202, 204 to asset monitoring and control system 124. Data ingestion process 206 effectively decouples each analytical avatar 200, 202, 204 from its corresponding industrial asset to allow the avatar to operate independently of the specific means used to stream or otherwise feed data from the industrial asset to the avatar. Likewise, data emission process 208 effectively decouples analytical avatars 200, 202, 204 from asset monitoring and control system 124 to allow the avatar to operate independently of the specific means used to stream or otherwise send the analytics generated by the avatar to monitoring and control applications. In some embodiments, loose coupling is achieved via containerized software that containerizes the analytical avatars, allowing the avatars to communicate with any computing platform to which they can be deployed using appropriate messaging protocols. Suitable messaging protocols may include AMQP (Advanced Message Queuing Protocol), MQTT (Message Queuing Telemetry Transport), JMS (Java Message Service), XMPP (Extensible Communication and Presentation Protocol), and other messaging protocols.

[0044] Figure 3 Exemplary types of real-time input signals from devices are illustrated. These signals can be streamed or otherwise fed to exemplary analysis avatars 200, 202, and 204 from their respective industrial assets (via data acquisition process 206). In this example, compressor data 300 from an X-type compressor may include pressure (e.g., suction pressure, discharge pressure), valve cover temperature, cylinder temperature, etc. Based on this compressor data, compressor fault detection avatar 200 can predict, for example, compressor faults, such as suction line blockage, and the time until compressor fault. On the other hand, pump data 302 from a Y-type pump may include vibration frequency, pump speed, rod load, etc. Based on this pump data, pump fault detection avatar 202 can predict, for example, pump faults, such as cavitation, and the time until pump fault. Motor data 304 from a Z-type motor may include stator current, accelerometer readings, motor load, etc. Based on this motor data, motor fault detection avatar 204 can predict, for example, motor faults, such as shaft imbalance, and the time until motor fault. In addition to the examples above, it can also provide avatars for other common problems and malfunctions in different types of rotating machinery, such as avatars for monitoring excessive vibration, overheating, etc.

[0045] Figure 4Several exemplary workers, as previously mentioned, are shown that can be used to build a data processing pipeline 400 for the analytics avatar. These workers provide the core analytics components required to build an avatar (e.g., the compressor fault detection avatar 200). Each of the specific workers shown here contains or performs a specific task; each worker performs different tasks that the avatar 200 requires to provide its predictive diagnostic capabilities specifically for the X-type compressor. In this example, the workers include a data injector worker 402, a data cleansing worker 404, a data transformation worker 106, a data classifier and / or predictor worker 408, and a result emitter worker 410. Users can select these workers (e.g., from a worker library) and drag and drop them into the compressor fault detection avatar 200 to build the data processing pipeline 400 for it. It should be understood that, depending on the specific predictive diagnostic capabilities required, other workers besides those shown here, or those replacing those shown here, may also be selected and dragged and dropped to form the avatar 200.

[0046] Figure 4 The operation of the workers shown is generally well-known and therefore will not be described in detail here. The data injector worker 402 serves as the data entry point for the avatar 200, receiving and accepting data streamed to the avatar using an appropriate messaging protocol. As its name suggests, the data cleaning worker 404 is used to detect and correct or filter any corrupted data or any excessive or irrelevant data (i.e., noise). The data transformation worker 406 converts the received data from its received format into a format usable by the data classifier and / or predictor worker 408. The data classifier and / or predictor worker 408 processes the data and uses, for example, ML, AI, various neural network algorithms, etc., to look for patterns in the data in order to infer the probability of faults or other problems, and in some cases, the time until a fault or problem occurs. The result transmitter worker 410 runs to output any predictive diagnostics generated by the avatar 200 in a specified form.

[0047] Now for reference Figure 5An exemplary avatar packaging toolset 500 is illustrated, which users can use to build analytics avatars, such as motor fault detection avatar 204. The exemplary toolset 500 may contain or include various software tools required by the user to assemble workers into the data processing pipeline of avatar 204. In the example shown, the avatar packaging toolset 500 includes a search tool 502 that can access a worker registry or repository (not explicitly shown) to allow the user to search for specific workers. Toolset 500 also includes an editing tool 504, which the user can use to modify or adjust workers according to the needs of a specific avatar. A connection tool 506 is also present, which the user can use to create or otherwise specify connections between various workers to assemble the data processing pipeline. A verification tool 508 allows the user to check and verify that the connections between workers function as expected. The user can then use one or more containerization tools 510 to package the selected workers into packages or containers that can be deployed on virtually any type of computing platform, such as cloud, edge, embedded devices, etc. Examples of suitable container software that can be used as a containerization tool include Docker and Nano containers.

[0048] exist Figure 5 In this example, the user has assembled a data processing pipeline 512, specifically designed for Z-type motors, made from worker components. In this example, the data processing pipeline 512 is used to analyze measured vibrations and stator currents to predict motor faults, such as bearing failures, rotor failures, etc. Therefore, the user has used search tool 502 to search the worker repository and found suitable data cleaning worker 514, data transformation worker 516, and classifier and / or predictor worker 518. Data cleaning worker 514 removes noise and other unwanted artifacts from the data and only passes "good" data samples to subsequent workers for processing. Data transformation worker 516 converts the "good" data into a format suitable for subsequent worker processing, for example, by converting time-domain vibration data 302 into frequency-domain data (e.g., using Fast Fourier Transform (FFT)) or by converting measured current data 304 into RMS data. Classifier and / or predictor worker 518 is trained to process the data and detect certain data features indicating the likelihood of a fault.

[0049] The results described above are analytical avatar 204 specifically developed for deployment in Z-type motors or instances thereof to detect faults. Those skilled in the art will understand that it can be used (or reused) with... Figure 5The different workers shown are used to develop other types of analytics avatars for the Z-type motor, such as anomaly detection avatars, aging detection avatars, etc. Of course, these workers can also be used (or reused) after appropriate training (or retraining) of the same or different workers to develop analytics avatars designed to be deployed with completely different assets and asset types, such as the compressors and pumps discussed earlier. Furthermore, although avatars are specifically developed for a particular asset and asset type, once built, each avatar can be deployed on multiple computing platforms by being containerized and having loose coupling to input and output data streams (via data ingestion process 206 and data emission process 208).

[0050] An example of loosely coupled input data streams can be found in Figure 6 As seen in the text, Figure 6 A general-purpose computing node 600 is shown for illustrative purposes. One or more general-purpose analytics avatars 602 are deployed on the computing node 600, each avatar connected to a data ingestion process 604 and a data transmission process 606. In the example shown, the data ingestion process 604 and the data transmission process 606 may take the form of message brokers or be implemented as message brokers configured to receive and send messages using specific messaging protocols. Message brokers 604 and 606 essentially decouple the analytics avatars 602 from their asset data source 608 and output applications (e.g., asset monitoring and control applications), respectively. This allows data from the asset data source 608 to be streamed or otherwise fed to the analytics avatars 602 via any suitable wireless and / or wired data communication link, including Wi-Fi, LPWAN (LoRA), ZigBee, Bluetooth, Ethernet, cellular, satellite, etc.

[0051] exist Figure 6In this configuration, data from asset data source 608 can be streamed or otherwise fed to analytics avatar 602 via one or more message servers 610 that execute suitable messaging protocols. Examples of suitable message servers 610 that can be used include MQTT servers, OPC UA (Open Platform Unified Architecture) servers, and data history servers. Other servers that can be used include Amazon S3 servers, NoSQL servers, ODBC / JDBC (Open Database Connectivity / Java Database Connectivity) servers, and IoT / IIoT (Industrial Internet of Things) servers. Moreover, analytics avatar 602 can be loosely coupled to a “digital twin” or digital representation (i.e., virtualization) of the asset residing on a cloud or local computing node. This arrangement provides the flexibility to develop or design data ingestion and output processes 604 and 606 separately to ingest and output data using any desired or required messaging protocol. In fact, in embodiments where data is provided by a digital twin, the analytics avatar can connect directly to the digital twin without requiring a data ingestion process.

[0052] Figure 7 It shows Figure 6 A more specific example of the general implementation is shown below. Several analytics avatars are deployed on the local compute nodes 700 of the X-type compressor 702 and the Y-type pump 704. The analytics avatars include an anomaly detection avatar 706 for the X-type compressor using a random forest ML algorithm, a device aging avatar 708 for the X-type compressor also using a random forest ML algorithm, and an anomaly detection avatar 710 for the Y-type pump using an NFR (Natural Forest Regeneration) ML algorithm. A data ingestion process 712 (e.g., a message broker) loosely couples these analytics avatars 706, 708, and 710 to the compressor 702 and the pump 704. Data from the compressor 702 is collected by an MQTT source 714 and forwarded to an MQTT server 716 for streaming to the data ingestion process 712. Similarly, data for the pump 704 is collected by a history source 718 and forwarded to a data history server 720 for streaming to the data ingestion process 708. At the output end, the data transmission process 722 (e.g., a message broker) loosely couples the analytics avatars 706, 708, and 710 to one or more end applications, such as asset monitoring and control applications.

[0053] Figure 8An example of an analytics avatar deployed in an IoT / IIoT environment is illustrated. In this example, data from multiple complex machines 800 (e.g., motors, compressors, pumps, etc.) and devices 802 (e.g., boilers, coolers, mixers, etc.) is wirelessly transmitted via one or more gateways or concentrators 804 to several analytics avatars residing on a cloud platform 806. This data may include, for example, device sensor readings, process instrument readings, distributed control system data, device control system data, computerized maintenance management system data, and any other type of data related to the operation of industrial facilities. Each type of data from each type of machine or device is processed by an analytics avatar developed specifically for that type of data from that type of machine. For example, there may be a machine avatar 808 for type X machines, a machine avatar 810 for type Y machines, a device avatar 812 for type A devices, a device avatar 814 for type B machines, and so on. Predictive diagnostics generated by analytics avatars 808, 810, 812, and 814 are then provided to end applications, such as asset monitoring and control applications.

[0054] Figure 9 An exemplary embodiment of this disclosure is shown, in which workers are used (or reused) to construct an analytical avatar 900 with multiple data processing pipelines. The analytical avatar 900 in this example employs a decision tree ML algorithm to provide anomaly detection for an X-type compressor. Two distinct data processing pipelines have been built to support the decision tree algorithm—a stage 1 training pipeline 902 and a stage 2 application pipeline 904—by selecting and dragging and dropping existing workers. Each worker is written in a supported programming language (e.g., Python, R, MATLAB, etc.) and contains one or more predictive diagnostic tasks configured to be executed or run within the pipeline configuration.

[0055] Phase 1 training pipeline 902 includes a worker 906 for injecting one or more training datasets, a worker 908 for filtering and cleaning the datasets (e.g., removing noise), a worker 910 for transforming the training datasets (e.g., transformation from the time domain to the frequency domain), a worker 912 for training one or more decision trees modeling the X-type compressor, and a worker 914 for outputting the results produced by one or more models. Each worker and each analytics avatar can be graphically represented using icons or images that the user can manipulate (e.g., drag and drop), or as a file with a filename and appropriate file extension, or both.

[0056] Phase 2 application pipeline 904 includes a data injector worker 916, a data cleaning worker 918, a data transformation worker 920, a model application worker 922, and a results emitter 924. The data injector worker 916 receives and accepts data for the avatar 900 via the previously mentioned data ingestion process, using an appropriate messaging protocol. The data cleaning worker 918 cleans and filters any corrupt or irrelevant data, while the data transformation worker 920 transforms the data into an appropriate format. The model application worker 922 operates similarly to the data model training worker 912 to detect anomalies, except that it uses actual data streamed to the avatar instead of historical data. The results emitter worker 924 outputs any predictive diagnostics generated by the model application worker 922 for external ingestion.

[0057] In some embodiments, each worker has a worker profile 926 associated with it. In some embodiments, the worker profile can be a text-based file, such as a JSON (JavaScript Object Notation) file, which a user can edit to change the worker's behavior. Similarly, each pipeline 902, 904 also has an associated text-based stage profile 928, which a user can edit to change the worker arrangement (and thus functionality) for the stage. Likewise, avatar 900 has an associated text-based avatar profile 930, which a user can edit to change the avatar's worker arrangement and pipeline (and thus predictive diagnostics). The workers, pipelines, and various profiles can then be packaged and containerized into discrete and independent predictive diagnostic units or avatars (e.g., anomaly detection.avt). Avatar 900 can then be stored in a registry and / or repository for subsequent retrieval, deployment, and reuse on virtually any analytics platform.

[0058] Figure 10An exemplary Avatar Studio 1000 is shown, which users (e.g., data scientists, analytics developers, etc.) can use to assemble workers into one or more data processing pipelines for analytics avatars. These users can also be (typically) the same users who initially developed workers for various industrial assets and asset types. Overall, Avatar Studio 1000 provides a unified engineering platform for data scientists and analytics developers to collaborate and share work, thereby minimizing duplication of effort. A worker 1012 designed and developed by a user and the resulting avatar 1018 can be uploaded to and made available on the Avatar Registry / Storage 1002 via Avatar Studio 1000. Other users can then retrieve and use (or reuse) the workers and avatars in the Registry / Storage 1002 to build their own predictive diagnostic capabilities, for example, by combining workers 1012 into one or more new avatars 1018, or simply by reusing one or more existing avatars.

[0059] As can be seen, the exemplary Avatar Studio 1000 provides users with several toolsets and capabilities for building and managing avatars, including an avatar packaging toolset 1004, an avatar management toolset 1006, avatar configuration and deployment capabilities 1008, and avatar system management capabilities 1010. Generally, the avatar packaging toolset 1004 includes tools that users can use to assemble workers 1012 into one or more data processing pipelines, edit workers and pipelines as needed via worker and pipeline configuration files 1014, and then package or containerize (1016) the workers and pipelines into an avatar 1018. This is similar to... Figure 5 The toolset 500 discussed in section 1006 includes tools that users can use on workers, pipelines, and avatars to perform various management tasks (e.g., organizing, uploading, downloading, etc.). The avatar configuration and deployment capabilities 1008 of the Avatar Studio 1000 are typically used to configure workers, pipelines, and avatars (e.g., via their configuration files) and deploy avatars on their desired analytics platforms and compute nodes. Similarly, avatar system management capabilities 1010 are typically used to monitor and manage (e.g., start, pause, stop, etc.) avatars that have been deployed.

[0060] The analytical avatar embodiments disclosed above provide a standard and consistent approach to developing predictive diagnostic products / solutions for a variety of industrial asset types. Once developed, analytical avatars can be reused and / or retrained and redeployed, significantly reducing development time for packaging new predictive analytics models for similar device types. Avatars are self-contained and loosely coupled to their respective data streams, allowing them to be bound to any device instance and operate with any DCS (Distributed Control System), SCADA (Supervisory Control and Data Acquisition) system, or any other type of monitoring and control system. This avoids the need for separate predictive diagnostic products / solutions for each computing platform (e.g., cloud, edge, embedded devices, etc.).

[0061] Turn now Figure 11 An exemplary flowchart representing method 1100 is shown, which can be used to provide predictive analytics for industrial assets using the analytics avatar embodiments disclosed herein. The method typically begins at 1102, where a user (e.g., a data scientist, analytics developer, etc.) accesses a registry or repository of existing analytics avatars and workers, for example, using the avatar packaging toolset discussed herein. As discussed, avatars and workers are specifically configured for specific industrial assets of a specific asset type. For example, some avatars and workers may be configured for type X compressors, while others may be configured for type Y pumps, and so on.

[0062] In step 1104, the user selects a specific worker from the available workers in the avatar registry. The selected worker constitutes (or contains) the core analytical components required to provide a given predictive analytics capability for a given asset of a given asset type. These workers may include, for example, data cleansing workers, data transformation workers, and data classifier and / or predictor workers. Depending on the desired deployment requirements, the selected workers may also include data injector workers and data emitter workers. In step 1106, the user assembles or combines the selected workers with each other to build one or more data processing pipelines. At this point, the user can also modify any worker by editing its configuration file as needed.

[0063] In 1108, a training dataset (e.g., historical data) for a given asset of a given type is applied to one or more data processing pipelines for training purposes. In 1110, one or more data processing pipelines are validated for a given asset of a given type, for example, using additional training datasets or real-world (e.g., live streaming) data. In 1112, workers and one or more data processing pipelines, along with their configuration files, are packaged or containerized into independent, self-contained predictive diagnostic units or entities, forming avatars. Users can perform containerization using any of the previously mentioned containerization tools (e.g., Docker, Nano, etc.). This containerization allows avatars to be deployed on virtually any computing platform to provide predictive diagnostics for a given asset of a given asset type.

[0064] In step 1114, the avatar and worker, along with their data processing pipeline, are stored in an avatar repository for later use and reuse by others. In step 1116, the user deploys the avatar to the desired computing platform to provide predictive diagnostics for a given asset of a given asset type. Deployment involves loosely coupling the avatar to the data flow of the given asset of a given asset type using an appropriate messaging protocol (e.g., AMQP, MQTT, JMS, XMPP, etc.), avoiding the need to provide a specific messaging protocol within the avatar. In step 1118, the same avatar can be redeployed on different computing platforms for the same given asset of the same asset type (e.g., in different plants) as needed. Optionally, the avatar can be retrained using training data from subsequent assets prior to subsequent deployments, as needed.

[0065] At 1120, the results of the predictive diagnostics performed by the avatar are provided to external systems, such as monitoring and control systems for industrial assets. Depending on the type of predictive diagnostics performed by the avatar, these results may include, for example, fault detection and the time until the fault, anomaly, wear, aging, etc., occur. At 1122, the monitoring and control system takes any corrective actions that may be necessary based on the results of the predictive diagnostics. For example, the monitoring and control system may cut off the compressor power, reduce the motor speed, issue an audible alarm, and other similar actions.

[0066] Figure 12 Exemplary computing systems (or nodes) that can be used to implement various embodiments of this disclosure are shown. In general, any general-purpose computer system used in the various embodiments of this disclosure, such as a general-purpose computer, such as those based on Intel... Motorola Advanced Micro Devices (AMD) HP A computer system consisting of one or more processors, or any other type of processor. Such a computer system can be physical or virtual.

[0067] For example, various embodiments of this disclosure can be implemented as dedicated software that executes in a general-purpose computer system 1200, such as... Figure 12 As shown. Computer system 1200 may include at least one processor 1220 connected to one or more memory devices 1230, such as disk drives, memory, or other devices for storing data. Memory 1230 is typically used to store programs and data during operation of computer system 1200. Computer system 1200 may also include a storage system 1250 providing additional storage capacity. Components of computer system 1200 may be coupled via interconnection mechanism 1240, which may include one or more buses (e.g., between components integrated within the same machine) and / or networks (e.g., between components residing on separate, discrete machines). Interconnection mechanism 1240 enables communication (e.g., data, instructions) to be exchanged between system components of system 1200.

[0068] The computer system 1200 also includes one or more input devices 1210, such as a keyboard, mouse, trackball, microphone, and touchscreen, and one or more output devices 1260, such as a printer, display screen, and speakers. Furthermore, the computer system 1200 may include one or more interfaces (not shown) for connecting the computer system 1200 to a communication network (as a supplement to or alternative to the interconnection mechanism 1240).

[0069] Figure 13The storage system 1250, shown in more detail below, typically includes a computer-readable and writable non-volatile recording medium 1310, in which signals are stored that define a program to be executed by at least one processor 1220, or information stored on or in the medium 1310 to be processed by the program, to perform one or more functions associated with the embodiments described herein. This medium may be, for example, a hard disk or flash memory. Typically, in operation, at least one processor 1220 causes data to be read from the non-volatile recording medium 1310 into a storage system memory 1320, which allows at least one processor to access information faster than the medium 1310. The storage system memory 1320 is typically a volatile random access memory, such as dynamic random access memory (DRAM) or static random access memory (SRAM). As shown, the storage system memory 1320 may be located within the storage system 1250 or within the system memory 1230. At least one processor 1220 typically manipulates data within memory systems 1230 and 1320, and then copies the data to medium 1310 after processing. Various mechanisms are known for managing data movement between medium 1310 and integrated circuit memory elements 1230 and 1320, and this disclosure is not limited thereto. This disclosure is not limited to any particular memory system 1230 or storage system 1250.

[0070] Computer systems may include specially programmed special-purpose hardware, such as application-specific integrated circuits (ASICs). The aspects of this disclosure can be implemented using software, hardware, or firmware, or any combination thereof. Furthermore, these methods, actions, systems, system elements, and components may be implemented as part of the aforementioned computer system or as independent components.

[0071] Although computer system 1200 is shown as an example of a type of computer system on which various aspects of this disclosure may be implemented, it should be understood that aspects of this disclosure are not limited to such... Figure 13 This disclosure can be implemented on the computer system shown. Various aspects of this disclosure can be implemented on systems with... Figure 13 The different architectures or components shown are implemented on one or more computers. Furthermore, where the functions or processes of embodiments of this disclosure are described herein (or in the claims) as being executed on a processor or controller, such description is intended to include systems that use more than one processor or controller to perform the functions.

[0072] Computer system 1200 can be a general-purpose computer system that can be programmed using a high-level computer programming language. Computer system 1200 can also be implemented using specialized hardware that is specifically programmed. In computer system 1200, at least one processor 1220 is typically a commercially available processor, such as the well-known Pentium processor available from Intel Corporation. Many other processors are also available. Such a processor typically runs an operating system, which can be, for example, Windows 125, Windows 128, Windows NT, Windows 2000, Windows ME, Windows XP, Vista, Windows 7, Windows 13, or their derivatives available from Microsoft Corporation; MAC OS System X or its derivatives available from Apple Computer; Solaris, UNIX, Linux (any distribution), or their derivatives available from various sources, available from Sun Microsystems. Many other operating systems can be used.

[0073] At least one processor and operating system together define a computer platform for which applications written in a high-level programming language are created. It should be understood that embodiments of this disclosure are not limited to a specific computer system platform, processor, operating system, or network. Furthermore, it will be apparent to those skilled in the art that this disclosure is not limited to a specific programming language or computer system. Moreover, it should be understood that other suitable programming languages ​​and other suitable computer systems may also be used.

[0074] One or more parts of a computer system may be distributed across one or more computer systems coupled to a communication network. These computer systems may also be general-purpose computer systems. For example, aspects of this disclosure may be distributed across one or more computer systems configured to provide services (e.g., servers) to one or more client computers or to perform overall tasks as part of a distributed system. As another example, aspects of this disclosure may be implemented on a client-server or multi-tiered system including components distributed across one or more server systems that perform various functions according to various embodiments of this disclosure. These components may be executable, intermediate (e.g., IL), or interpreted (e.g., Java) code that communicates over a communication network (e.g., the Internet) using a communication protocol (e.g., TCP / IP). For example, one or more database servers may be used to store device data, such as expected power consumption, for designing layouts associated with embodiments of this disclosure.

[0075] Various embodiments of this disclosure can be programmed using programming languages ​​such as Python, R, MATLAB, Smalltalk, Java, C++, Ada, or C# (C-Sharp). Other programming languages, such as BASIC, Fortran, Cobol, TCL, or Lua, can also be used. Various aspects of this disclosure can be implemented in non-programming environments (e.g., analysis platforms or documents created in HTML, XML, or other formats that, when viewed in a browser program window, present aspects of a graphical user interface (GUI) or perform other functions). Various aspects of this disclosure can be implemented as programmable or non-programmable elements, or any combination thereof.

[0076] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and / or operation of various implementations of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, code segment, or code portion, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may not appear in the order indicated in the drawings. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order, depending on the functionality involved. It will also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a system based on dedicated hardware that performs the specified function or action, or by a combination of dedicated hardware and computer instructions.

[0077] It should be understood that the above description is intended to be illustrative and not limiting. Many other embodiments will become apparent upon reading and understanding the above description. Although specific examples have been described in this disclosure, it should be recognized that the systems and methods of this disclosure are not limited to the examples described herein, but can be implemented with modifications within the scope of the appended claims. Therefore, the specification and drawings are to be considered illustrative and not limiting. Accordingly, the scope of this disclosure should be determined by reference to the appended claims and the full scope of their equivalents.

Claims

1. A system for providing predictive diagnostics for industrial assets, comprising: At least one processor; A storage system communicatively coupled to the at least one processor and storing instructions on the storage system that, when executed by the at least one processor, cause the system to: The process of selecting multiple workers from a repository and connecting the workers to each other to build a data processing pipeline, wherein each worker performs a different task in the data processing pipeline; The process of containerizing the plurality of workers and the data processing pipeline to form analytics avatars and storing the containerized analytics avatars in a repository, wherein the repository stores the plurality of analytics avatars for subsequent use and reuse; The process of selecting a first analytical avatar from the plurality of analytical avatars in the repository is performed, each analytical avatar being configured to provide corresponding predictive diagnostic functions for the corresponding asset of the corresponding type, independently of the computing system; The process of deploying the first analytical avatar to a first computing system is performed, wherein the first analytical avatar runs on the first computing system to provide given predictive diagnostic capabilities for a first asset of a given type; and The process involves deploying the first analytics avatar to a second computing system, different from the first computing system, for reuse, wherein the first analytics avatar runs on the second computing system to provide the given predictive diagnostic functionality for the given type of second asset.

2. The system according to claim 1, wherein, The instructions cause the system to perform a process of selecting multiple workers from the repository and connecting the workers to each other by dragging and dropping them to build a data processing pipeline for the second analytical avatar, wherein each worker performs a different task for the second analytical avatar in the data processing pipeline.

3. The system according to claim 2, wherein, The instructions cause the system to perform a process of containerizing the plurality of workers and the data processing pipeline to form the second analytics avatar, which is configured to provide diagnostic functions different from those of the first analytics avatar.

4. The system according to claim 3, wherein, The instructions cause the system to perform the process of storing the second analytical avatar into the repository for later use and reuse.

5. The system according to claim 1, wherein, The instructions cause the system to perform a process of loosely coupling the first analytical avatar to a data stream from the first asset of the given type.

6. The system according to claim 1, wherein, The instructions cause the system to perform a process of coupling the first analytical avatar to a digital twin of the first asset of the given type.

7. The system according to claim 1, wherein, The instructions cause the system to perform a process of outputting the results of the given predictive diagnostic function of the first analytical avatar to a monitoring and control system, wherein the monitoring and control system takes a specified corrective action based on the results of the given predictive diagnostic function.

8. A method for providing predictive diagnostics for industrial assets, comprising: Multiple workers are selected from the storage and connected to each other to build a data processing pipeline, in which each worker performs a different task; The plurality of workers and the data processing pipeline are containerized to form analytics avatars, and the containerized analytics avatars are stored in a repository, which stores a plurality of analytics avatars for subsequent use and reuse. Select the first analytical avatar from multiple analytical avatars in the repository. Each analytical avatar is configured to provide corresponding predictive diagnostic functions for the corresponding type of corresponding asset, independently of the computing system. The first analytical avatar is deployed to a first computing system, and the first analytical avatar runs on the first computing system to provide given predictive diagnostic capabilities for a first asset of a given type; and The first analytical avatar is deployed to a second computing system, different from the first computing system, for reuse. The first analytical avatar runs on the second computing system to provide the given predictive diagnostics for the given type of second asset.

9. The method of claim 8, further comprising selecting a plurality of workers from the repository and connecting the workers to each other by dragging and dropping the workers to construct a data processing pipeline for the second analytical avatar, each worker performing a different task for the second analytical avatar in the data processing pipeline.

10. The method of claim 9, further comprising containerizing the plurality of workers and the data processing pipeline to form a second analytics avatar, the second analytics avatar being configured to provide diagnostic functionality different from that of the first analytics avatar.

11. The method of claim 10, further comprising storing the second analytical avatar in the repository for subsequent use and reuse.

12. The method of claim 8, further comprising loosely coupling the first analytical avatar to a data stream from the first asset of the given type.

13. The method of claim 8, further comprising coupling the first analytical avatar to a digital twin of the first asset of the given type.

14. The method of claim 8, further comprising outputting the result of the given predictive diagnostic function of the first analytical avatar to a monitoring and control system, wherein, The monitoring and control system takes specified corrective actions based on the results of the given predictive diagnostic function.

15. A system for providing predictive diagnostics for industrial assets, comprising: A storage repository that stores multiple analytical avatars, each of which is configured to provide corresponding predictive diagnostic functions for a corresponding type of asset independently of the computing system; A studio platform connected to the repository, the studio platform having multiple avatar tools configured to: Multiple workers are selected from the storage and connected to each other to build a data processing pipeline, in which each worker performs a different task; The plurality of workers and the data processing pipeline are containerized to form analytics avatars, and the containerized analytics avatars are stored in a repository for subsequent use and reuse. Select a first analytical avatar from the plurality of analytical avatars in the repository; The first analytical avatar is deployed to a first computing system, and the first analytical avatar runs on the first computing system to provide given predictive diagnostic capabilities for a first asset of a given type; and The first analytical avatar is deployed to a second computing system, different from the first computing system, for reuse. The first analytical avatar runs on the second computing system to provide the given predictive diagnostics for the given type of second asset.

16. The system according to claim 15, wherein, The avatar tool is also configured to select multiple workers from the repository and connect the workers to each other by dragging and dropping them to build a data processing pipeline for the second analytical avatar, with each worker performing a different task for the second analytical avatar in the data processing pipeline.

17. The system according to claim 16, wherein, The avatar tool is also configured to containerize the plurality of workers and the data processing pipeline to form a second analytical avatar, the second analytical avatar being configured to provide diagnostic functions different from the first analytical avatar.

18. The system according to claim 17, wherein, The avatar tool is also configured to store the second analytical avatar in the repository for later use and reuse.

19. The system according to claim 15, wherein, The avatar tool is further configured to loosely couple the first analytical avatar to a data stream from the first asset of the given type.

20. The system according to claim 15, wherein, The avatar tool is further configured to couple the first analytical avatar to a digital twin of the first asset of the given type.

Images