Systems and methods for optimal driver configuration using machine learning

Abstract

The present invention relates to a system for optimal driver configuration using machine learning; the system comprises: - a data collector configured to collect data and to establish interrelationships among the collected data; - a training dataset generator configured to compute a set of configurations based on the collected data and on the established interrelationships, further configured to compute a measured success value for the set of configurations, further configured to generate a training dataset comprising the set of configurations together with the corresponding measured success value; - a machine learning module configured to predict a predicted success value for the computed set of configurations using a machine learning algorithm using the training dataset provided by the training dataset generator; and - an optimization module configured to rank the computed set of configurations, the optimization module comprising a simulation module configured to simulate the computed set of configurations.

Description

TECHNICAL FIELD

[0001] The present invention relates to a system and method for optimal drive configuration using machine learning. BACKGROUND

[0002] Industrial drives are part of a powertrain, which is constituted by other devices such as motors and transformers. For successful configuration of the powertrain, it is necessary to know the details of the application (e.g. HVAC (heating, ventilation and air conditioning), water pumping, machine and motion control). Based on the application details, it is possible to generate configuration parameters for the controller before the system is assembled.

[0003] Since the drive is the central element in the powertrain, proper configuration of the drive is critical for the customer. Wrong configuration parameters have a negative impact on the productivity of the system.

[0004] Successful drive configuration requires knowledge of system design, simulation engineering and operation. It also requires expert knowledge honed with practical experience. Currently, there is no general tool that integrates these various sources of knowledge for drive configuration.

[0005] As Figure 1 depicted in the background art, separate tools are used for initial powertrain selection based on available product catalogues and available initial information about the application. After selection of the drive / powertrain devices, separate tools are used for simulation and configuration of the drive.

[0006] As a result, the initial configuration parameters are typically based on partial knowledge and simulations are employed to iteratively adjust in order to achieve a proper drive configuration, which has to be finally verified and often adjusted in the real system. It is desirable to take a more holistic approach in order to reduce engineering time and cost. Furthermore, wrong configurations discovered during the operation phase can have even greater impact on cost and productivity. SUMMARY

[0007] It would therefore be advantageous to have improved technology in order to improve industrial drives that are part of a powertrain.

[0008] The objects of the present invention are solved with the subject matter of the independent claims, wherein further embodiments are incorporated in the dependent claims.

[0009] The present invention fills this gap with a system concept and corresponding methods for the configuration of drive parameters. Machine learning is used to generate optimal configuration parameters for drives. The advantage of this approach is to reduce engineering time, engineering effort and maintenance of engineering tools, customer losses and risks when designing and operating industrial drive systems.

[0010] In a first aspect, there is provided a system for optimal drive configuration using machine learning, the system comprising: - a data collector configured to collect data and establish correlations among the collected data; - a training dataset generator configured to calculate a configuration set based on the collected data and based on the established correlations, the training dataset generator further configured to calculate a measured success value for the configuration set, the training dataset generator further configured to generate a training dataset comprising the configuration set together with the corresponding measured success value; - a machine learning module configured to predict a predicted success value for the calculated configuration set using a machine learning algorithm using the training dataset provided by the training dataset generator; and - an optimization module configured to order the calculated configuration set, the optimization module comprising a simulation module configured to simulate the calculated configuration set.

[0011] According to an exemplary embodiment of the present invention, the system further comprises a user interface module configured to provide a configuration of the system and configured to provide a visualization of the machine learning process performed by the machine learning module.

[0012] According to an exemplary embodiment of the present invention, the system further comprises a user interface module configured to initiate a feedback mechanism for continuously improving the machine learning algorithm.

[0013] According to an exemplary embodiment of the present invention, the system further comprises a user interface module configured to provide alternatives of the machine learning algorithm to a user.

[0014] According to an exemplary embodiment of the present invention, the data collector is configured to collect data by adopting a text processing approach or a text mining approach, or to collect data extracted from customer requirements.

[0015] According to an exemplary embodiment of the present invention, the optimization module is configured to collect real-time series data used by the simulation module to simulate the calculated configuration set.

[0016] According to a second aspect of the application, there is provided a method comprising the steps of:

[0017] - collecting data by a data collector and establishing correlations among the collected data;

[0018] - computing a set of configurations by a training data set generator based on the collected data and on the established correlations, computing a measured success value for the set of configurations, generating a training data set comprising the set of configurations together with the corresponding measured success value;

[0019] - predicting success values for the computed set of configurations by a machine learning module using a machine learning algorithm using the training data set provided by the training data set generator; and

[0020] - ranking the computed set of configurations by an optimization module and simulating the computed set of configurations by a simulation module.

[0021] The application advantageously provides a tool for creating operational parameters for industrial drive systems created based on knowledge from various sources.

[0022] The application advantageously provides that the tool is able to learn from experience embodied in existing expert systems, human expert training, and knowledge gained from simulations and from installed systems. The tool is continuously updated with new experience gained from these various sources.

[0023] A user describes his application (e.g. motor load(s), environmental conditions, electrical network parameters, etc.) as input. The tool then presents him with recommended drive operational parameters.

[0024] Experience is gained at least from the following sources:

[0025] i) existing expert systems embodied in tools used e.g. during engineering

[0026] ii) human expert knowledge

[0027] iii) simulation results for the drive system gained during engineering and testing, including electrical, thermal and mechanical simulation of the system,

[0028] iv) system data in deployed production, including measurements, observed anomalies during operation and service information

[0029] For successful configuration of industrial drive systems, it is necessary to know the details of the application (e.g. HVAC (heating, ventilation and air conditioning), pumping, machine and motion control). Based on the application details, it is possible to generate configuration parameters for the controller before the system is assembled.

[0030] Proper configuration is of utmost importance for the customer. Wrong configuration parameters can have a negative impact on the productivity of the system.

[0031] The present invention fills this gap for the configuration of drive parameters with a system concept and corresponding method. Machine learning is used to generate the best configuration parameters for the drive. The advantage of this method is to reduce engineering time, engineering effort, maintenance of engineering tools, and customer's loss and risk when designing and operating industrial drive systems.

[0032] In a second aspect, there is provided a method for best drive configuration using machine learning.

[0033] Step 1 comprises extracting the parameters of interest from the customer requirements by using text processing or text mining approaches such as NLP.

[0034] Step 2 comprises based on the system description identified in step 1, machine learning provides a set of drive configuration parameters (candidates) with the highest success value known so far (i.e. above a certain threshold).

[0035] Step 3 comprises all candidates are evaluated in a simulation (in parallel or iteratively can be as a cloud service).

[0036] Step 4 comprises generating a report providing the results of all candidates.

[0037] Step 5 comprises if the simulation results converge, the best set of drive configuration parameters is provided to the user. Otherwise, the above steps are repeated with better design requirements or constraints in step 1 or higher success threshold in step 2 until convergence is achieved.

[0038] Step 6 comprises learning by imitation. The expert engineer takes the automatically selected set of drive configuration parameters. The expert can change things during commissioning. These changes are fed back to the machine learning module in order to learn how to fine-tune / configure the real system.

[0039] Step 7 comprises simulation improvement: During commissioning, real-time time series data is collected in order to improve the simulation model (simulation parameter identification and optimization).

[0040] Step 8 comprises when in service, alarms and event logs are collected and fed back to the machine learning module for learning from event patterns.

[0041] Step 9 comprises time series data for optimizing the simulation and the learned model.

[0042] Step 10 comprises: the process can be stopped at any point in time, e.g. after the first parameterization of the drive or after a number of iterations of improvement.

[0043] However, the data collection part from the real drive can continue to be used for future systems.

[0044] When done iteratively, the method can generate the best set of drive configuration parameters in step 2.

[0045] According to a further aspect, there is provided a computer program element for operating a system, which, when being executed by the unit, is adapted for performing the method according to the first aspect.

[0046] According to a further aspect, there is provided a computer readable medium having stored the computer program element of the fifth aspect. The computer readable medium can be provided as a physical data carrier, such as a CD-ROM, USB stick or the like, or can be provided digitally via a communication network, such as the Internet.

[0047] For instance, the computer program element can also be transmitted or otherwise provided for download over the wireless

[0048] Advantageously, the advantages provided by any of the above aspects and examples apply equally to all of the other aspects and examples, and vice versa.

[0049] These and other aspects of the application will become apparent from the embodiments described hereinafter.

[0050] The above aspects and examples will become apparent from the embodiments described hereinafter, and will be explained with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0051] The exemplary embodiments will be described below with reference to the following drawings:

[0052] Figure 1 A schematic illustration of a schematic process diagram employing separate tools for initial transmission system selection based on available product catalogues and available initial information about the application is shown for explaining the present patent application;

[0053] Figure 2 An example of a method for optimal drive configuration using machine learning according to exemplary embodiments of the present patent application is shown;

[0054] Figure 3An example of a method for optimal drive configuration using machine learning according to an exemplary embodiment of the present patent application is shown. DETAILED DESCRIPTION

[0055] Figure 1 A schematic illustration of a schematic process diagram is shown employing a separate tool for initial drive train selection based on available product catalog and available initial information about the application for explaining the present patent application.

[0056] Figure 2 An example of a method for optimal drive configuration using machine learning according to an exemplary embodiment of the present patent application is shown.

[0057] Figure 3 An example of a method for optimal drive configuration using machine learning according to an exemplary embodiment of the present patent application is shown.

[0058] Figure 1 A DriveSize DS is shown connected with a virtual drive VD, a simulation engine and a drive composer DC.

[0059] For successful configuration of an industrial drive system, it is necessary to know the details of the application (e.g. HVAC (heating, ventilation and air conditioning), pumping, machine and motion control). Based on the application details, it is possible to generate configuration parameters for the controller before the system is assembled.

[0060] Proper configuration is critical for the customer. Wrong configuration parameters can have a negative impact on the productivity of the system.

[0061] Figure 2 An example of a method for optimal drive configuration using machine learning according to an exemplary embodiment of the present patent application is shown.

[0062] According to exemplary embodiments of the present patent application, a tool for making engineering decisions is created based on data collected from various sources. The engineering tool learns the task of making decisions based on data without explicit programming. The system improves as more data becomes available over time.

[0063] According to exemplary embodiments of the present patent application, data from various sources is used to create models for industrial engineering decisions. Sources include existing expert systems, human experts, simulation results, observations from production systems.

[0064] According to exemplary embodiments of the present patent application, the method includes and is based on training data samples readily available, for example, from existing expert systems, to initialize the training data set.

[0065] According to the exemplary embodiments of the present patent application, the training dataset is continuously updated with new samples from other sources.

[0066] According to the exemplary embodiments of the present patent application, data from different sources can require proper weighting in the creation and updating of the training dataset.

[0067] According to the exemplary embodiments of the present patent application, active learning techniques can be used to request more expensive training data samples, for example, to computationally expensive simulation experiments.

[0068] According to the exemplary embodiments of the present patent application, supervised learning and proper learners are used to create a model for creating operational drive parameters from the training dataset.

[0069] According to the exemplary embodiments of the present patent application, engineering data collected during the entire life cycle of the drive system installation is collected and used to create the training data.

[0070] According to the exemplary embodiments of the present patent application, machine learning is used to create operational drive parameters.

[0071] According to the exemplary embodiments of the present patent application, the learners use the following to build the model:

[0072] Simulation results during engineering

[0073] Configuration parameters before commissioning and after commissioning

[0074] Configuration parameters during life cycle (maintenance)

[0075] Alarm and event logs

[0076] Time series data

[0077] Changes done by experts

[0078] According to the exemplary embodiments of the present patent application, the learners will detect patterns of successful device parameterization during the life cycle of the system (from engineering, commissioning, and operation).

[0079] According to the exemplary embodiments of the present patent application, the learners will detect patterns of erroneous / bad configuration.

[0080] According to the exemplary embodiments of the present patent application, the above model will be continuously improved and refined during regular engineering tasks and operation in order to incorporate the latest findings and serve as a continuously reliable tool.

[0081] According to exemplary embodiments of the present patent application, the overall system comprises a drive engineering tool (e.g. DriveComposer Pro), a simulation module consisting of process / mechanical models (motor load), electrical models (motor, inverter, Trafo), thermal models (motor and inverter), virtual drive controller.

[0082] According to exemplary embodiments of the present patent application, the learning module comprises a data collection module that collects data from: feature extraction from customer files, such as environmental parameters; technical requirements from industrial applications.

[0083] Example: Pumping station describes technical requirements, expected input and output water pressure

[0084] According to exemplary embodiments of the present patent application, the simulation results are provided in a database of working / successful configurations.

[0085] According to exemplary embodiments of the present patent application, the simulation results are provided as a history of changes of configuration parameters from the tool (e.g. DriveComposer Pro, PLM or versioning system).

[0086] According to exemplary embodiments of the present patent application, alarms and / or event logs are recorded by the system.

[0087] According to exemplary embodiments of the present patent application, the training dataset generator is configured to derive data from the data collection, calculate configuration success values for different configuration sets based on the information collected from the data collection module.

[0088] According to exemplary embodiments of the present patent application, the training dataset generator is configured to generate a training dataset comprising the corresponding success values obtained in the above steps together with the possible configuration sets.

[0089] According to exemplary embodiments of the present patent application, the machine learning (data analysis) module is configured to predict success values for new configuration sets using the training dataset provided by the training dataset generator.

[0090] According to exemplary embodiments of the present patent application, the training dataset generator is configured to find the most successful configuration set for the user input data.

[0091] According to exemplary embodiments of the present patent application, the optimization module is configured to use the most successful drive configuration sets provided by the machine learning module so far and test them in the simulation module.

[0092] According to exemplary embodiments of the present patent application, a user interface module is used to allow visualization and configuration of the machine learning process, to allow a feedback mechanism for continuous improvement of the machine learning algorithm, to allow providing alternatives to the user.

[0093] Figure 3 An example of a method for optimal drive configuration is shown. Simulation parameters or drive operating parameters can be used as parameters. The use of parameters can be iterated in parallel. Simulation results can be given in simulation data, anomalies or threshold violations. A database can include successful working configuration parameters or equivalent data sets. A DS project can have a load topology and selection constraints, a set of environments.

[0094] Figure 3 An example of a method for optimal drive configuration is shown. A machine learning module ML uses simulation values, installed base, configuration parameters, changes during life cycle, user knowledge / additional constraints, learning with different weights (bidding, engineering, operation). The machine learning module ML generates drive parameters and simulation parameters.

[0095] While the application has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The application is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a study of the drawings, the disclosure, and the appended claims.

Claims

1. A system for providing optimal drive configuration for industrial drives using machine learning, the system comprising: - a data collector configured to collect data and establish interrelationships among the collected data, wherein the data collector is configured to collect the data by employing text processing or text mining to extract parameters of interest from customer requirements; - a training dataset generator configured to compute a set of configurations based on the collected data and based on the established interrelationships, to compute measured success values for the set of configurations, and to generate a training dataset comprising the set of configurations together with corresponding measured success values; - a machine learning module configured to use a machine learning algorithm to predict success values for the computed set of configurations using the training dataset provided by the training dataset generator; - an optimization module configured to rank the computed set of configurations, the optimization module comprising a simulation module configured to simulate the computed set of configurations; and - a user interface module configured to provide configuration of the system and to provide visualization of the machine learning process performed by the machine learning module; wherein during commissioning, changes made by an expert to the automatically selected set of drive configuration parameters are fed back to the machine learning module in order to learn how to fine-tune or configure the real system; wherein during commissioning, real-time time series data is collected in order to improve the simulation model by simulation parameter identification and optimization; wherein during service, alarms and event logs are collected and fed back to the machine learning module for learning from event patterns; wherein time series data is used to optimize the simulation and the learned model; wherein the system is configured such that: - if the simulation results converge, the best set of drive configuration parameters is provided to the user; - otherwise the system employs better design requirements or constraints of the customer or a higher success threshold until convergence is achieved.

2. The system of claim 1, the system further comprising a user interface module configured to initiate a feedback mechanism for continuously improving the machine learning algorithm. wherein 3. The system of claim 1, the system further comprising a user interface module configured to provide the user with alternatives of the machine learning algorithm. wherein 4. The system of claim 1, the data collector is configured to collect data by employing text processing or text mining or to collect data extracted from customer requirements. wherein 5. The system of claim 1, the optimization module is configured to collect real-time time series data used by the simulation module to simulate the computed set of configurations. wherein, 6. A method for providing optimal drive configuration for industrial drives using machine learning, the method comprising the steps of: ​ - collecting data by a data collector and establishing correlations among the collected data, wherein the data are collected by employing text processing or text mining approaches to extract parameters of interest from customer requirements; - computing a configuration set based on the collected data and based on the established correlations by a training data set generator, computing a measured success value for the configuration set, generating a training data set comprising the configuration set together with the corresponding measured success value; - predicting success values for the computed configuration set using a machine learning algorithm by a machine learning module using the training data set provided by the training data set generator; - ranking the computed configuration set by an optimization module and simulating the computed configuration set by a simulation module, - providing a configuration of the system and providing a visualization of the machine learning process performed by the machine learning module, wherein during commissioning, changes made by an expert to the automatically selected set of driver configuration parameters are fed back to the machine learning module in order to learn how to fine-tune or configure a real system, wherein during commissioning, real-time series data are collected in order to improve the simulation model by simulation parameter identification and optimization, wherein during service, alarm and event logs are collected and fed back to the machine learning module for learning from event patterns; and - using time series data to optimize the simulation and learned model; wherein - if the simulation results converge, the best set of driver configuration parameters is provided to the user; - otherwise the above steps are repeated with better design requirements or constraints of the customer or with a higher success threshold until convergence is achieved.

7. The method of claim 6, wherein the method further comprising the step of initiating a feedback mechanism for continuously improving the machine learning algorithm.

8. A computer program product comprising a computer program configured to perform the method of any one of claims 6-7 when executed by a system of any one of claims 1 to 5.

Images