Systems and methods for securely training and using models

Abstract

A system and method for securely training a model, the method comprising: incorporating the model's algorithm into a training module; and transferring the training module from a vendor computer system to a client computer system. The training module operates on the client computer system using data from the client, and after operating the training module on the client computer system, a trained training module is implemented. The trained training module is used to initialize at least one additional module, the at least one additional module being used to score observations, and the at least one additional module being transferred from the client computer system to the vendor computer system.

Description

Technical Field

[0001] The present disclosure relates to systems and methods for machine learning. Background Art

[0002] Machine learning and related technologies are currently pervasive across many technology sectors due to their usefulness in the design, manufacture, management, and improvement of modern industrial assets. Machine learning algorithms can help provide improved product designs and smarter asset management. In typical applications, assets with adaptive or learning algorithms continuously learn or adjust their operations based on real and / or virtual data streams to improve operational stability and efficiency in dynamic environments. Adaptive algorithm-enabled operating and learning systems are typically responsible for processing large amounts of structured and unstructured sensor data from various platforms in test or production environments and deriving scoring assessments and predictions relative to those assets that can be used to improve operational efficiency, thereby eliminating unplanned downtime and reducing maintenance costs.

[0003] By default, due to the nature of these systems, obtaining optimal model designs (including learning infrastructure and algorithms) and sufficiently complete and diverse data streams to fully train effective models is often challenging. This is because the model designs are typically proprietary to the asset vendor, while the underlying data is almost always generated on-site at the asset and is considered confidential and proprietary to each user of the asset. At the same time, the vendor of the machine learning system is neither located in the premises where the asset is operating nor working exclusively with a specific customer at any given time. This requires transferring data from the customer to the vendor, or transferring the model design from the vendor to the customer, which is not an attractive proposition for both the vendor and / or the customer. The compromise is to transfer the data and perform training on a remote environment that hosts the learning infrastructure and algorithms (e.g., a cloud service), which is separate and isolated from the customer or vendor's internal systems. This approach also has the disadvantage that the vendor will not trust the cloud to keep its model designs private, and the customer will generally not trust the cloud to keep its data private and secure. Neither party will trust the cloud network to operate reliably and continuously. Vendors also don’t want their training infrastructure, algorithms, and intellectual property to exist and run publicly at customer sites for access to local data streams, and for the same reason, customers don’t want their own unprotected data-trained models to exist and be accessible in remote locations. Summary of the Invention

[0004] In one aspect, the present disclosure describes a method for enabling a vendor to facilitate training a model using data from a customer. The model comprises an algorithm, and the training process involving the data from the customer resides in a training infrastructure. The method includes: incorporating the model's algorithm into a training module; and transferring the training module from a vendor computer system to a customer computer system. The training module operates on the customer computer system using the data from the customer, and after operating the training module on the customer computer system, a trained training module is implemented. The trained training module is used to initialize at least one additional module used to score observations, and the at least one additional module is transferred from the customer computer system to the vendor computer system.

[0005] In another aspect, the present disclosure describes a non-transitory computer-readable storage medium comprising instructions that, when executed by a processing device in a vendor computer system, cause the processing device to: construct a model as a training module using the processing device; communicate with a customer computer system containing customer data stored in a computer-readable form; transfer the training module to the customer computer system for training based on the customer data; and convert the training module into a trained training module in the customer computer system and initialize an action module and an insight module.

[0006] In yet another aspect, the present disclosure describes a system comprising: a memory for storing instructions; a processing device operably coupled to the memory to execute the instructions stored in the memory, the processing device being configured and operable to: construct a model as a training module; communicate with a customer computer system containing customer data stored in a computer-readable form; transfer the training module to the customer computer system for training based on the customer data; extract training results from the trained training module into an operation module and an insight module in the customer computer system; and receive an insight module from the customer computer system to complete the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a block diagram illustrating a topological structure of a collaborative fleet intelligence system (CFIS) according to the present disclosure.

[0008] Figure 2 The overall system architecture of an embodiment of the CFIS according to the present disclosure is shown.

[0009] Figure 3 FIG. 1 is a process diagram of an embodiment of a CFIS according to the present disclosure.

[0010] Figure 4 is a flow chart of a method of operating an embodiment of a CFIS according to the present disclosure.

[0011] Figure 5 is a flow chart of a method for facilitating training a model according to the present disclosure.

[0012] Figure 6 is a block diagram of a storage medium storing machine-readable instructions according to the present disclosure.

[0013] Figure 7 is a flow diagram of a system process contained in a memory as instructions for execution by a processing device coupled to the memory in accordance with the present disclosure. DETAILED DESCRIPTION

[0014] The present disclosure is applicable to creating complete machine learning applications or models using data generated by working assets located at a customer's location without passing the data to the supplier's system, thereby protecting the supplier's machine learning tool while it resides on the customer's system, and creating and using different parts of modules of the machine learning tool inherent in one or more customer's and supplier's respective systems. The modules, their contents and use are protected using security certificates and the like. For example, a module produced by a supplier to be shared with a customer is created using the customer's public key and can only be accessed using the customer's private key. In addition, a module produced by a customer to be shared with a supplier is created using the supplier's public key and can only be accessed using the supplier's private key. If there are no proprietary issues related to the module and / or the underlying data, either party may choose to produce an unlocked module.

[0015] Once value has been established and mutually agreed upon, the client and vendor can choose to share their intellectual property. This allows the vendor to assemble a complete solution without requiring access to the client's underlying data. Alternatively, the client can assemble a complete solution without publicly accessing the vendor's model design. Alternatively, the client and vendor can collaborate to leverage the value of advanced analytics applications without sharing proprietary information. The system includes a collection of submodules to measure suitability for asset types and operating environments, and also allows for extensive quality control of the data used to train the models.

[0016] In one general aspect, the customer always owns and controls the data generated by the customer. Such data can include a collection of operational data acquired by sensors associated with assets operating at the customer's facilities, which can relate to specific components or systems as well as overall processes, for example, for manufacturing or distributing products or other commodities such as electrical power or mechanical power. The volume, velocity, and variety of data are evaluated and assessed by the system, which can also classify the data with respect to any desired dimension or attribute (for example, the geographic location where the asset operates). This processing helps to produce more reliable data sets for training machine learning tools, and can also be used to create specialized modules and training mechanisms for models that are more suitable for use with different customers' assets operating under similar operating and / or environmental conditions.

[0017] A key area of focus for CFIS is collaboration, where any customer's data is processed and integrated into training mechanisms that include contextual information specific to the asset and its usage. The vendor doesn't use the data itself; instead, it's aggregated and analyzed at the customer's facility. This aggregation and analysis provides experience and context for models, parts of which are compiled on the customer's system. Representations of those parts of the model and the underlying data are used to create modules, which are then transferred outside the customer's system in a secure, encapsulated manner. The training modules are used to generate operational and insight modules, which are tailored specifically for specific assets operating under certain conditions and locations.

[0018] To illustrate, in one embodiment, the training module includes a model design that can be implemented with various modeling techniques such as neural network layers, regression parameters, etc. Data observations are provided to the training module to derive a trained model. The training module can also operate to aggregate data as well as training results. Advantageously, the training module exists in an encrypted form, making reverse engineering of the proprietary model architecture contained therein more difficult. The information generated by the training module is related to customer data, but does not contain actual customer data. The trained training module then provides information to the operation module, which is only available to the customer, and prepares the information required for the application solution. In addition, the trained training module then provides information to the insight module, which is only available to the supplier. In a sense, the operation module provides input to the insight module, which processes the intermediate data it ingests to provide output.

[0019] Multiple training modules can be used to create multiple operational modules, and in some instances, common trends and conclusions can be used to create insight modules, all without requiring direct access to the underlying customer data from the supplier's perspective. These and other aspects of the present disclosure will now be described in greater detail. Of course, in some embodiments, the training modules, operational modules, and insight modules can all be included in a single "combined module" that is provided as a complete solution to a customer operating an asset at a customer facility.

[0020] Figure 1 , a block diagram of the topology of a CFIS 100 according to the present disclosure is shown. As can be seen from this diagram, CFIS 100 involves one or more systems that are communicatively and optionally connected via encrypted communications via a communication environment 102. Communication environment 102 can be any electronic mode of electronic communication that provides secure communications between physically remote servers. In the exemplary topology shown, three different environments are involved and described, including a provider 104, a training facility 106, and a user facility 108. As used herein, the term "environment" is intended to have a broad scope and can include any physical or virtual environment. For example, provider 104 can include an environment that exists in one or more locations, each of which includes computing infrastructure operated by professionals, such as servers, processors, databases, memory, communication equipment, etc. Training facility 106 can include a facility in which assets such as industrial equipment are operated, with associated sensors, monitoring and control systems, etc., which provide data streams to computer systems operating training facility 106 or associated therewith. In one embodiment, supplier 104 may be the original equipment manufacturer of the various assets and / or sensors operated at training facility 106, which the customer operating training facility 106 has purchased from supplier 104 to perform an industrial process. User facility 108 may be similar to training facility 106, i.e., owned and operated by a different customer of the supplier that has purchased and uses equipment similar to that of the customer owning training facility 106.

[0021] One embodiment of a process for creating the CFIS 100 includes, at 110, designing a suitable training module by the vendor 104. The training module itself may be implemented as a computer-executable algorithm that includes machine learning functionality and is stored in a computer-readable medium in the form of computer-executable instructions. Creating the training module at 110 may involve or require various assessments by the vendor, including problem definition and computational detail determination at 112. For example, the vendor's engineers may analyze sample data using manual computational methods or more typical computational methods to identify patterns in data generated for the intended asset using test data for that particular asset. A model feasibility study may also be conducted to determine the desired model type and configuration or model design to be incorporated into the training module. Technical considerations may be used to adjust the intended model as needed before the training module is actually created at 110.

[0022] It should be appreciated that in certain embodiments, the training module, as well as other modules discussed herein, may be embodied as a computer application or program. The computer program may exist in a variety of active and inactive forms. For example, the computer program may exist as software program(s) comprising: program instructions in source code, object code, executable code, or other formats; firmware program(s); or hardware description language (HDL) files. Any of the above may be embodied on a computer-readable medium, including computer-readable storage devices and media, in compressed or uncompressed form, and signals. Exemplary computer-readable storage devices and media include RAM (random access memory), ROM (read-only memory), EPROM (erasable programmable ROM), EEPROM (electrically erasable programmable ROM), and magnetic or optical disks or tapes of conventional computer systems. Exemplary computer-readable signals, whether modulated using a carrier wave or not, are signals that a computer system hosting or executing the present teachings can be configured to access, including signals downloaded via the Internet or other networks. Specific examples of the foregoing include the distribution of executable software program(s) of the computer program on a CD-ROM or downloaded via the Internet. In a sense, the Internet itself, as an abstract entity, is a computer-readable medium, as are computer networks in general.

[0023] Once the training module has been created at 110, it is signed, encrypted using the customer's public key, and transmitted to the training facility 106 via the communication environment 102 or equivalent or a direct connection. At 114, assets of the type used to evaluate the feasibility of the envisioned model may already be operating and in use within the environment of the training facility 106. At 116, the received training module is provided with the data or data stream generated or present at the training facility 106 at 114 by the operational assets to train the training module. Training the training module at 116 using the customer's private key may include many different operations and processes, such as using the training module to ingest data, analyze the data for trends and quality (e.g., by examining the statistical spread or distribution of data values), normalizing the data, cleaning up existing semantic models included in the training module, performing tests to determine training accuracy, and other data operations.

[0024] After sufficient training has been performed, the trained training module is formalized at 118. The trained training module 118 uses the raw data generated by operating the asset at 114, including those raw data records themselves, and achieves its training by computing information about the data required for the correct operation of the model. The trained training module 118 serves as the basis for extracting the operational module at 122 and the insight module at 124. The extracted operational module 122 is constrained for use only by the customer, and the insight module 124 is constrained for use only by the supplier. These constraints can be removed at a later stage after agreement between the supplier and the customer. Extraction of the operational module 122 and insight module 124 occurs at the training site under the direction and control of the customer operating the training facility 106. Alternatively, a customer who does not wish to expose their data can send the trained training module 118 to the supplier 104, where the operational module 122 and insight module 124 are extracted under the direction and control of the supplier 104.

[0025] In one embodiment, the extraction of the operational module at 122 is initialized using only a portion (e.g., half) of the split configuration of the training information and begins scoring observations. The operational module is protected by the client's public key and can only be operated with knowledge of the client's private key. This allows the client to partially control the value generated from the entire solution. At an intermediate stage, at 124, the operational module results are extracted and used as input to the insight module. The insight module is protected by the provider's public key and can only be operated with knowledge of the provider's private key. This allows the provider to partially control the value generated from the entire solution. Splitting and staging the information in this way ensures that the module is processing client data and that value is transferred without sharing actual data. Thus, the insight module is initialized at 124 using the second half of the split configuration and completes scoring. The asset provider 104 obtains the final results, which can be fed back to the client interface 125. This ensures that the full model is never exposed to the client at 106 or 108, but the results are based on the fully trained module.

[0026] The operation module extracted at 122 and the insight module extracted at 124 are modules that can be used or adapted for assets operated at the training facility 106 or at any other different customer operating similar assets at 126 (e.g., a customer operating the user facility 108). For example, another customer operating the user facility 108 can use the operation module at 128 and the insight module at 130 to more efficiently operate its assets at 126. For example, these modules can be provided from the vendor's customer interface 125. Because of the similarity of the assets operated at facilities 106 and 108, the quality checks performed on the data during training, and other factors, the operation module and insight module created based on the assets in the training facility 106 are applicable to assets operated at the customer operating the user facility 108. Of course, the operation module 122 and insight module 124 can also be used in the training facility 106.

[0027] It will be appreciated that by engaging one or more training facilities 106, training one or more training modules, and then using these modules to generate one or more operational modules and insight modules tailored to a specific customer operating a user facility 108, many different dimensions of the parameters that most impact the operational and insight modules can be taken into account. Additional customers operating user facilities 108 can also volunteer to act as training facilities 106 themselves. Importantly, the training facility's data never leaves its native environment, and the modules are extracted by the customer at the customer operating the training facility 106, so that with the training modules protected by the customer and only the operational and insight modules deployed, no single deployment environment contains the data and resulting models, which eliminates any concerns about the security and integrity of the data and models, as with the previously proposed solutions. The communication environment 102 only transmits encrypted information. In one embodiment, the training modules themselves remain encrypted within the training facility environment throughout their lifecycle.

[0028] Figure 2 An embodiment of the overall system architecture for an embodiment of the CFIS 200 according to the present disclosure is shown in FIG to illustrate the collaborative aspects of the CFIS. Figure 1 The corresponding components and systems discussed are the same or similar Figure 2 Elements and features shown and discussed in the accompanying drawings are denoted by the same reference numerals used previously for simplicity. Figure 2 CFIS 200 includes a vendor 104 that owns and develops a fleet of models 202 that are deployed within a vendor asset cloud 204 as empty training modules 203 configured for a specific customer operating a training facility 106. The vendor asset cloud 204 is a secure cloud or other secure electronic communication modality owned, operated, and controlled by the vendor 104. The group of training modules 203 is deployed in the vendor asset cloud 204 and can be retrieved by the customer at the training facility 106.

[0029] As previously described, the training facility may include: a plurality of assets 206; and training modules 203, which are developed for training at other customer facilities (e.g., at the training facility 106). Each module may correspond to or be tailored for a specific machine 208 or machine system. After training is complete, an operational module and an insight module are extracted from the trained training modules 209 and the deployed applications as previously described, where the operational module is executed at the customer site operating the training facility 106 and the insight module is executed at the supplier asset cloud 204. If both modules are transferred to the supplier asset cloud 204, this comprises a complete trained model 212. For collaboration, the portion of the operational module 205 of the trained model 212 that has been revised with constraints is ready for deployment to other customers (such as customers operating the user facility 108), and the portion of the insight module 207 of the trained model 212 is ready for deployment to one or more partner integration clouds 214, with only restrictions placed on their use. It should be appreciated that if the customer chooses to lift the restriction and share the trained training module 212 with the supplier asset cloud 204 , the trained training module can be processed into action modules and insight modules on the supplier's systems either online or within the supplier asset cloud 204 .

[0030] By design, it is contemplated that multiple training modules may be used to create multiple operational module configurations that generate intermediate results that are ingested into multiple insight module configurations. These multiple parallel model creation processes will result in a functional training model fleet 212 maintained by the vendor 104, for example, within the vendor asset cloud 204. Individual modules may have local or global applicability and may be tailored to operate with various asset models and configurations. In one embodiment, a customer operating a user facility 108 (particularly a larger customer with multiple physical facilities) may act as a strategic partner to the vendor and maintain its own partner integration cloud 214 from which modules specific to the assets of the customer operating the user facility 108 may be available as needed. Portions of the insight module fleet 207 of trained models 212 may be provided and maintained on an ongoing basis by the vendor 104 on the partner integration cloud 214 and as a service, for example, a subscription service.

[0031] Figure 3, a schematic diagram of an exemplary process for initializing the operational and insight modules from portions of the model configuration contained in a trained training module is shown. In the illustrated embodiment, a neural network model configuration is used to illustrate the initialization process, but it should be appreciated that other machine learning techniques may also be encompassed under the framework as model synergies develop. Referring to the illustrated process, the training module may first analyze the input data stream to determine quality metrics such as consistency. In the illustrated embodiment, for example, the data point distribution 302 of various data streams, which evaluates the probability distribution, mean, and standard deviation of the data points, may be used to determine data spread and consistency. These characterizations are captured as part of the extracted operational module. Characterizing the data in this manner not only ensures data quality, but also serves as a classification of the data characteristics used for the training module and provides a basis for later selecting the trained module for application to another client's system whose data characteristics closely match the characteristics of the data generated by assets that the client may use.

[0032] Data determined to have good repeatability (i.e., there is a high confidence that new values will fit within the initial distribution used to train the model) is provided as the input layer 304 of model 306. Using a neural network as an example, after training is complete, the input layer, combined with the hidden layers, essentially becomes part of the operational module, while after training is complete, the output layer essentially becomes part of the insight module. The input and output segments of model 306, embodied in this example as a neural network model, are combined in the training module for training purposes. Model 306 can be divided into multiple blocks for distributed scoring, including structural weight splits, layer splits, loss activation splits, hash encoder splits, and so on. Training feedback also spans partitions 308. Split scoring, or segmented scoring, between the operational and insight modules effectively isolates the customer input layer 304 from the supplier output layer 310. Transmitting intermediate values across partitions 308 isolates the input and output values from each other and ensures that data leaked from the input layer cannot be directly accessed on the output layer 310, and vice versa. Thus, if the segmentation 308 includes a communication environment 102 (e.g., between a training facility (e.g., training facility 106) and a supplier (e.g., supplier 104), Figure 1 ), the data and the resulting split or segmented models remain solely the property of the customer or supplier, with no possibility of proprietary information being accidentally passed from one to the other.

[0033] Additional advantages can be achieved by separating the input layer 304 from the output layer 310. For example, each intermediate pass 308 can include a timestamp during the pass to prevent replay attacks and ensure security and traceability.

[0034] Figure 4 A flowchart of the CFIS operational method is shown in Figure 2. This method provides a solution consisting of three modules working in tandem: a training module, an operational module, and an insight module. This module separation replaces the existing system, in which all data is sent by the data owner to an external entity for model training. Instead, the customer or data owner receives an appropriate model from the model fleet for the training module, based on the level of fit with their environment and design choices. The customer can then use the training module to locally train the model on their data. This involves a certain amount of preprocessing, which acts as a funnel to ensure an appropriate data-model match. The training module generates operational and insight modules without any raw customer data. The training module configuration is partitioned and encrypted to ensure the security of the model fleet. The operational module is initialized using one half of the partitioned configuration and begins scoring observations. At an intermediate stage, the results are extracted and sent to the insight module. This ensures that the model is processing customer data and that value can be conveyed without sharing the actual data. The insight module is initialized and completes scoring using the second half of the partitioned configuration. The asset provider receives the results, which can be fed back to the customer interface. This ensures that the full model is never exposed to clients, but the results are based on the fully trained model.

[0035] Therefore, more specifically, Figure 4 As shown, the method begins at 402 with the asset vendor creating a training module. As used herein, the term "module" may be understood to mean a container for a model, multiple models, or a portion of a model, as well as processing instructions for configuring, analyzing, and scoring the model. At 404, the training module is deployed to the customer training facility. Deploying the training module may include selecting a training module from a plurality of training modules that is adapted to the level and type of assets (e.g., motors, pumps, relays, turbines, etc.) of the training facility, as well as the facility's environment (e.g., ambient temperature, altitude, humidity, vibration, etc.) and design choices (the type of process or combination of processes in which the asset is used). When training the training module is initiated, pre-processing of the data is performed at 406 to ensure a good match. Although this step may be performed optionally, the pre-processing includes achieving a correct data-model match. It is worth noting that the trained model does not contain any original customer data, and its configuration is segmented and encrypted to ensure the security of the model fleet.

[0036] At step 408, the model in the training module is trained locally using the client's data at the training facility. Then, at 410, the operations module is launched, and at 414, the insights module is launched using the remainder of the training module, either of which can be performed on the vendor system or the client system using the separate configurations of the training module. Observations are also scored at 410, and the results are extracted and provided to the insights module at 412. Scoring is completed at 414 before providing the results to the vendor at 416 or, optionally, to the client at 418.

[0037] In one embodiment, the training module, the operation module, and the insight module are all parts of an overall model framework that is trained in stages and on different systems. Therefore, the creators and developers of the technology will be the vendors who design the model and deploy it on different systems. Although various parts of the system may operate on different environments, the proprietary nature of the cryptographic module retains ownership of the vendor and its development and utilization, making a single participant (the vendor) responsible for manufacturing and using the system described herein. Additionally, it should be appreciated that each module can be its own model that evolves as each component is trained in stages and on different systems, but always under the ownership and control of the vendor.

[0038] Figure 5 A flow chart of a method for facilitating training a model according to an exemplary embodiment of the present disclosure is shown. In this embodiment, the flow chart illustrates a method for enabling a vendor to use data from a customer to facilitate training a model. The model may include an algorithm. The training process may involve the data from the customer being present in a training infrastructure. Figure 5 As shown, at 502, the model's algorithm is incorporated into a training module. At 504, the training module containing the model's algorithm is transferred from the vendor's computer system to the client's computer system. At 506, the training module is operated on the client's computer system using data from the client. After operating the training module on the client's computer system at 506, the trained training module is implemented at 508. At 510, at least one additional module is initialized using the trained training module. At 512, the at least one additional module is transferred from the client's computer system to the vendor's computer system. Then, at 514, the observations are scored using the at least one additional module.

[0039] Figure 6A block diagram of a system process contained in a memory as instructions for execution by a processing device coupled to the memory is shown in accordance with an exemplary embodiment of the present disclosure. The instructions included on the non-transitory computer-readable storage medium 600, when executed, cause the processing device of the supplier computer system to perform various tasks. In the illustrated embodiment, the memory includes construction instructions 602 for constructing a model into a training module using the processing device. The memory also includes communication instructions 604 for communicating with a customer computer system containing customer data stored in a computer-readable form; and transfer instructions 606 for transferring the training module to the customer computer system for training based on the customer data. The memory 600 also includes: conversion instructions 608 for converting the training module into a trained training module in the customer computer system; and initialization instructions 610 for initializing the operation module and the insight module.

[0040] Figure 7 A flow chart of a system process contained in a memory as instructions for execution by a processing device coupled to the memory is shown according to an exemplary embodiment of the present disclosure. In this embodiment, system 700 includes: a memory 702 for storing computer-executable instructions; and a processing device 704 operatively coupled to memory 702 for executing the instructions stored in the memory. Processing device 704 is configured and operable to: execute construction instructions 706, which construct a model into a training module; and execute communication instructions 708, which communicate with a client computer system containing client data stored in a computer-readable form. Further, transmission instructions 710 transmit the training module to the client computer system for training based on the client data, and extraction instructions 712 extract the training results from the trained training module into an operation module and an insight module in the client computer system. Receive instructions 714 receive the insight module from the client computer system to complete the model. All references, including publications, patent applications, and patents, that may be cited herein are hereby incorporated by reference to the same extent as if each reference were individually and specifically indicated to be incorporated by reference and were set forth in its entirety herein.

[0041] Unless otherwise indicated herein or clearly contradicted by context, the use of the terms "a" and "an", "the" and "at least one" and similar reference objects in the context of describing the present invention (especially in the context of the appended claims) should be interpreted as covering both the singular and the plural. Unless otherwise indicated herein or clearly contradicted by context, the use of the term "at least one" followed by a list of one or more items (e.g., "at least one of A and B") should be understood to mean one item (A or B) selected from the listed items or any combination of two or more of the listed items (A and B). Unless otherwise indicated, the terms "comprising", "having", "including" and "containing" should be interpreted as open-ended terms (i.e., meaning "including but not limited to") unless otherwise indicated herein. Unless otherwise indicated herein, the enumeration of value ranges herein is merely intended to serve as a shorthand method of referring separately to each individual value falling within the range, and each individual value is incorporated into the specification as if it were individually set forth herein. Unless otherwise indicated herein or clearly contradicted by context, all methods described herein can be performed in any suitable order. Unless otherwise claimed, the use of any and all examples or exemplary language (e.g., "such as") provided herein is intended merely to better illustrate the invention and does not limit the scope of the invention. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the invention.

[0042] Preferred embodiments of the present invention are described herein, including the best mode known to the inventor for carrying out the present invention. Variations of those preferred embodiments will become apparent to those of ordinary skill in the art upon reading the foregoing description. The inventors expect that those skilled in the art will employ such variations as appropriate, and the inventors intend to implement the present invention in a manner different from that specifically described herein. Thus, the present invention includes all modifications and equivalents of the subject matter set forth in the appended claims as permitted by applicable law. Furthermore, unless otherwise specified herein or clearly contradicted by the context, the present invention encompasses any combination of the above-described elements in all their possible variations.

Claims

1. A method for enabling a provider to use data from a customer to facilitate training a model, the model comprising an algorithm, the training process involving the data from the customer being resident in a training infrastructure, the method comprising: Including said algorithm of said model into a training module; transferring the training module from the supplier's computer system to the customer's computer system; operating said training module on said client's computer system using said data from said client; After operating the training module on the client's computer system, obtaining a trained training module; Initializing at least one operational module using a portion of the trained training module and initializing at least one insight module using a remaining portion of the trained training module; transferring the at least one insight module from the customer's computer system to the supplier's computer system, the at least one insight module being used only by the supplier's computer system; as well as Initiate scoring of observations using the at least one operational module; as well as Scoring the observations is accomplished using the at least one insight module.

2. The method of claim 1 , wherein the trained training module is divided into an input portion and an output portion, the input portion and the output portion being communicatively associated via a separation layer, wherein the observation result is input to the input portion, and wherein the output of the input portion is used as the input of the at least one operation module, and wherein the output of the output portion is used to score the observation result using at least one insight module.

3. The method according to claim 1, further comprising: The data from the client is pre-processed before being provided to the training module for training. 4 . The method of claim 3 , wherein pre-processing the data from the customer comprises at least one of performing a trend analysis, performing a quality analysis, and normalizing the data from the customer.

5. The method of claim 1, wherein transmitting the trained training module is accomplished via encrypted electronic communication between the customer's computer system and the supplier's computer system.

6. The method according to claim 1, further comprising: A plurality of training modules are developed, each training module in the plurality of training modules corresponding to a target asset type.

7. The method of claim 6, wherein each of the plurality of training modules corresponds to a target environment in which an asset is operating, and wherein the data from the customer is generated based on the asset operating within each target environment.

8. A non-transitory computer-readable storage medium comprising instructions that, when executed by a processing device in a vendor computer system, cause the processing device to: constructing the model as a training module using the processing device; communicating with a customer computer system containing customer data stored in a computer-readable form; transferring the training module from the supplier computer system to the customer computer system for use in training based on the customer data; converting the training module into a trained training module in a computer system of the client; Initializing an operation module using a portion of the trained training module and initializing an insight module using a remaining portion of the trained training module; as well as The insight module is transferred from the customer computer system to the supplier computer system, the insight module being used only by the supplier computer system.

9. The non-transitory computer-readable storage medium of claim 8, wherein the training module comprises computer-executable instructions adapted for execution by a processing device included in the client computer system, the computer-executable instructions causing the processing device of the client computer system to: pre-processing the customer data; ingesting the customer data into the training module to train the model and generate the trained training module; and The training module and the trained training module are saved in encrypted form. 10 . The non-transitory computer-readable storage medium of claim 9 , wherein pre-processing the customer data comprises at least one of performing a trend analysis, performing a quality analysis, and normalizing the customer data.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the trained training module is divided into an input portion and an output portion, the input portion and the output portion being communicatively associated via a separation layer, wherein the output of the input portion is used to score the observation results of the operation module, and wherein the output of the output portion is used to score the observation results of the insight module.

12. The non-transitory computer-readable storage medium of claim 8, wherein receiving the operational module and the insight module is accomplished through encrypted electronic communications between the customer's computer system and the supplier's computer system. 13 . The non-transitory computer-readable storage medium of claim 8 , further comprising a plurality of models, each model of the plurality of models corresponding to a target asset type.

14. A system comprising: Memory for storing instructions; a processing device operatively coupled to the memory to execute the instructions stored in the memory, the processing device being configured and operative to: Construct the model as a training module; communicating with a customer computer system containing customer data stored in a computer-readable form; transferring the training module from the supplier computer system to the customer computer system for use in training based on the customer data; extracting training results from a portion of the trained training module into an operations module in the client computer system and from a remaining portion of the trained training module into an insights module in the client computer system; as well as The insight module is received at the supplier computer system from the customer computer system to complete the model, the insight module being used only by the supplier computer system.

15. The system of claim 14, wherein the training module comprises computer-executable instructions adapted for execution by the processing device included in the client computer system, the computer-executable instructions causing the processing device of the client computer system to: pre-processing the customer data; ingesting the customer data into the training module to train the training module and generate the trained training module; and The training module and the trained training module are saved in encrypted form.

16. The system of claim 15, wherein pre-processing the customer data comprises at least one of performing a trend analysis, performing a quality analysis, and normalizing the customer data.

17. The system of claim 16, wherein the trained training module is divided into an input portion and an output portion, the input portion and the output portion being communicatively associated via a separation layer, wherein the output of the input portion is used to score the observation results of the operation module, and wherein the output of the output portion is used to score the observation results of the insight module.

18. The system of claim 14, wherein receiving the trained training module is accomplished through encrypted electronic communication between the customer computer system and the provider computer system.

19. The system of claim 14, further comprising a plurality of models, each model of the plurality of models corresponding to a target asset type.

20. The system of claim 19, wherein each model of the plurality of models corresponds to a target environment in which an asset is operating, and wherein the customer data is generated based on the asset operating within each target environment.

Images