Motor intelligent bench method and system based on multi-modal data fusion

Abstract

The invention discloses a motor intelligent bench method and system based on multi-modal data fusion, relates to the technical field of motor control, and comprises a control layer, an execution object, a data acquisition layer and a multi-modal data fusion layer. And the control layer is used for executing a motor control algorithm, outputting an operation result containing a control instruction and a control period, driving an execution object to dynamically adjust an operation mode and generating operation data. The data acquisition layer determines a sampling frequency or an acquisition channel based on an execution object operation mode, and triggers a sensor to synchronously acquire operation data by taking an operation result as a synchronous reference signal; the multi-modal data fusion layer is used for predicting a future operation state, reasoning potential faults, dynamically calibrating a virtual model and generating and verifying a motor design scheme or a control parameter optimization scheme; a closed-loop structure of acquisition-prediction-diagnosis-simulation-optimization is constructed, continuous iteration and intelligent optimization of motor control and design are realized, and the reliability of motor operation and the design efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of motor control technology, specifically to a method and system for an intelligent motor test bench based on multimodal data fusion. Background Technology

[0002] As a core power unit in modern industrial and transportation systems, the control and testing technology of electric motors has always been an important topic in the field of electric motor research and application. Traditional electric motor control and testing systems usually rely on a single controller architecture and limited parameter acquisition methods, which have certain limitations in terms of R&D efficiency, monitoring accuracy, and fault diagnosis capabilities.

[0003] Firstly, in the development of motor control algorithms, existing technologies mostly employ manual programming and hardware adaptation, requiring a long development cycle from algorithm modeling to actual operation. Researchers often need to repeatedly port and debug code between different software and hardware platforms, which is not only time-consuming and labor-intensive but also prone to introducing compatibility issues, resulting in low algorithm verification efficiency.

[0004] Secondly, in terms of monitoring motor operating status, existing test bench systems mainly rely on the acquisition of single variables or a few physical quantities, such as current and speed. Due to the lack of comprehensive analysis of multi-dimensional operating data, the system struggles to reveal the correlation between different parameters and cannot achieve a comprehensive description of the motor's state under complex operating conditions. For example, relying solely on current monitoring cannot promptly reflect temperature anomalies caused by insufficient heat dissipation, which may lead to untimely fault warnings.

[0005] Furthermore, in the fault diagnosis stage, existing technologies generally rely on manual experience or rule-based judgment methods. When a motor malfunctions, the diagnostic results often depend on the operator's accumulated experience and interpretation of limited detection signals, lacking systematic and intelligent support. This approach is not only inaccurate but also carries the risk of misdiagnosis and missed diagnosis, increasing motor maintenance costs and downtime.

[0006] Furthermore, with the diversification and increasing complexity of motor application scenarios, the shortcomings of existing test bench systems in terms of algorithm verification, condition monitoring, and fault diagnosis are no longer sufficient to meet the needs of efficient R&D, intelligent operation, and reliable maintenance.

[0007] In view of this, the present invention provides a method and system for intelligent motor test bench based on multimodal data fusion, which solves the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for intelligent motor testing based on multimodal data fusion, which aims to improve the development efficiency of the motor body and control algorithm, realize comprehensive and accurate monitoring and optimization of motor operating status, and support intelligent diagnosis of motor faults, thereby improving the reliability and efficiency of motor operation, reducing energy consumption and fault risk, and shortening the motor R&D cycle.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] In a first aspect, the present invention provides a smart motor test bench system based on multimodal data fusion, comprising:

[0011] The control layer is used to execute object control algorithms and output running results, which include control instructions and control cycles.

[0012] An execution object is used to dynamically adjust its operating mode according to the control instructions.

[0013] The data acquisition layer determines the sampling frequency or acquisition channel based on the operating mode of the execution object, and uses the operation result as a synchronization reference signal to trigger the sensor to synchronously acquire operation data within the control cycle;

[0014] The multimodal data fusion layer includes:

[0015] The Transformer time series large model is used to perform multi-step autoregressive prediction on the running data to obtain a sequence of future running states.

[0016] A cognitive intelligent agent is used to compare and reason with the sequence of future operating states and a knowledge graph, and to generate diagnostic information when potential faults or anomalies are identified.

[0017] A digital twin model is used to dynamically calibrate a virtual motor model based on the diagnostic information and to simulate and predict operating behavior under different working conditions.

[0018] The generative design module is used to automatically generate motor design schemes or control parameter optimization schemes based on the prediction results of the virtual motor model, and to perform iterative verification in a virtual test bench environment.

[0019] As a preferred embodiment of the first aspect of the present invention, the control layer includes a hardware-in-the-loop (HIL) controller based on RCP and HIL, and an MCU core control board based on MCU, wherein:

[0020] The hardware-in-the-loop simulation controller is used to automatically generate executable code adapted to the hardware from the modeling software.

[0021] The MCU core control board is used to execute the code and output the running results, including the control cycle, which serve as the synchronization reference signal for the data acquisition layer.

[0022] As a preferred embodiment of the first aspect of the present invention, when the execution object detects a sudden change in load, an environmental change, or an abnormal torque fluctuation...

[0023] Automatically adjusts the speed and torque parameters in the operating mode.

[0024] And the information on the change in operating mode is fed back to the data acquisition layer.

[0025] This enables dynamic matching between the data collection strategy and the state of the execution object.

[0026] As a preferred embodiment of the first aspect of the present invention, the data acquisition layer includes multiple types of heterogeneous sensors.

[0027] During the sampling process, the control cycle output by the control layer is used as the trigger condition.

[0028] The sensor is driven to achieve synchronous acquisition within the same control cycle, wherein: when the operation mode of the execution object is detected to be switched, the sampling frequency of the sensor is automatically adjusted or the acquisition channel is switched.

[0029] As a preferred embodiment of the first aspect of the present invention, the Transformer time-series large model includes an anomaly detection module.

[0030] The anomaly detection module is used to generate a sequence of future operating states simultaneously;

[0031] The real-time collected operational data is compared with the expected values.

[0032] When the comparison error exceeds the threshold, an abnormal alarm is output and the prediction step size is dynamically shortened.

[0033] As a preferred embodiment of the first aspect of the present invention, the cognitive intelligent agent includes a knowledge graph construction module and a reasoning engine module:

[0034] The knowledge graph construction module is used to structure and store motor design manuals, fault cases and expert experience to form a knowledge graph in the field of motors, which organizes motor components, fault types, fault causes and maintenance solutions into entities and relationships.

[0035] The inference engine module is used to receive future operating state sequences and perform matching inference in the knowledge graph to generate diagnostic information containing fault location, cause and development trend.

[0036] As a preferred embodiment of the first aspect of the present invention, the knowledge graph construction module includes an online learning unit:

[0037] This is used to dynamically update the entity relationships in the knowledge graph when new operational data or new failure cases are received.

[0038] And adjust the parameters of the inference engine in real time.

[0039] This enables diagnostic capabilities to continuously evolve as the motor operates.

[0040] As a preferred embodiment of the first aspect of the present invention, the digital twin model includes a dynamic calibration unit.

[0041] This is used to correct the parameters of the virtual motor model after receiving the diagnostic information.

[0042] Simulations were run and predictions were made under different load and environmental conditions.

[0043] This allows for the early identification of potential failure risks and maintenance time windows.

[0044] As a preferred embodiment of the first aspect of the present invention, the generative design module includes a virtual benchtop verification unit.

[0045] Used after generating motor design schemes or control parameter optimization schemes.

[0046] The digital twin model is used to simulate extreme working conditions and boundary conditions in a virtual environment.

[0047] The proposed solution is then iteratively optimized based on simulation feedback.

[0048] Until the target performance requirements are met.

[0049] Secondly, the present invention provides a smart test bench method for motors based on multimodal data fusion, which, based on the implementation of the first aspect, includes the following steps:

[0050] S101. Execute the motor control algorithm in the control layer, output the running results including control instructions and control cycle, and dynamically adjust the running mode in the execution object according to the control instructions to generate running data;

[0051] S102. The data acquisition layer determines the sampling frequency or acquisition channel based on the operating mode, and uses the operating result as a synchronization reference signal to trigger the sensor to synchronously acquire the operating data within the control cycle.

[0052] S103. In the multimodal data fusion layer, the Transformer time series large model is used to predict the running data, and the predicted sequence is compared and reasoned with the knowledge graph in the cognitive agent to generate diagnostic information of potential faults or anomalies.

[0053] S104. In the multimodal data fusion layer, the digital twin model is dynamically calibrated based on the diagnostic information, and the generative design module generates a motor design scheme or control parameter optimization scheme based on the prediction results of the virtual model. After iterative verification in the virtual bench environment, the scheme is fed back to the control layer.

[0054] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0055] This invention significantly shortens the development cycle of motor control algorithms by seamlessly connecting the hardware-in-the-loop (HIL) controller in the control layer with Matlab and enabling one-click code generation. This allows researchers to quickly verify new algorithms, thereby significantly improving algorithm development efficiency. Based on the Transformer time-series large-scale model in the multimodal data fusion layer, comprehensive analysis and prediction of multivariate operating data are performed, achieving precise monitoring and optimization of motor operating status, effectively improving operating efficiency and reducing energy consumption and failure risks. Furthermore, a cognitive agent combined with knowledge graph reasoning enables intelligent diagnosis of motor faults, reducing reliance on human experience and thus improving diagnostic accuracy and efficiency, while lowering maintenance costs and downtime. Finally, the combination of generative AI design and digital twin simulation verification enables rapid generation and optimization of motor control algorithm schemes, achieving closed-loop iteration in the R&D process, thereby shortening the overall R&D cycle and improving design quality. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0057] Figure 1 This is a schematic diagram of the intelligent test bench system of the present invention;

[0058] Figure 2 This is a schematic diagram illustrating the working mechanism of the multimodal data fusion layer of the present invention. Detailed Implementation

[0059] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art. The drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0060] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure may be practiced with one or more specific details omitted, or methods, components, steps, etc. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0061] Example 1

[0062] like Figure 1 As shown, this embodiment provides a motor intelligent test bench system based on multimodal data fusion, including a control layer, an execution object, a data acquisition layer, and a multimodal data fusion layer.

[0063] The control layer is used to execute object control algorithms and output running results, which include control instructions and control cycles.

[0064] Specifically, the control layer is equipped with a hardware-in-the-loop (HIL) simulation controller based on RCP and HLL. This HIL controller is specifically designed for researchers to quickly develop and verify motor control algorithms. It has the ability to connect to Matlab software, and executable code can be generated with a single click, greatly simplifying the algorithm development process and accelerating the transformation from theoretical algorithms to practical applications. For example, after researchers complete the algorithm model in Matlab, they can directly use this controller to quickly generate code that can run on actual hardware, without the need for complex manual programming conversion.

[0065] Meanwhile, the control layer is also equipped with a DSP and STM32 core control board based on an MCU chip, and the control algorithm is programmed using the CCS software development platform. This approach provides another flexible way to implement control algorithms, suitable for algorithm development and application in different needs and scenarios.

[0066] The execution objects, including motors, drivers, mechanical test benches, and power supplies, are used to generate operational data according to the control commands. As the actual execution module, it receives and executes the control algorithm from the control layer. During execution, the various devices work together to generate rich experimental data, including but not limited to motor current, voltage, temperature, torque, speed, and vibration data. This data reflects the motor's operating status, providing a foundation for subsequent monitoring and analysis.

[0067] Specifically, the motor, driver, mechanical test bench, power supply, and connecting components are assembled and connected according to design requirements. The motor serves as the power output unit, and the driver adjusts its operating parameters, such as speed and torque, based on control signals from the control layer. The mechanical test bench provides support and fixation for the motor and other components, ensuring operational stability. The power supply provides electrical support for the entire system. The connecting components are responsible for connecting the various devices into an organic whole, ensuring effective transmission of power and signals. During system operation, the motor operates under the control of the driver, working collaboratively with other equipment to generate various test data, such as motor current, voltage, and temperature.

[0068] The data acquisition layer dynamically adjusts the operating mode and sampling frequency or acquisition channel according to the control command, and uses the operating result as a synchronization reference signal to trigger the sensor to synchronously acquire operating data within the control cycle.

[0069] Specifically, the sensors include various types such as temperature sensors, vibration sensors, voltage sensors, current sensors, and speed sensors. These sensors collect various types of data generated during the execution of the object in real time and transmit the data synchronously to the multimodal data fusion layer to ensure the timeliness and integrity of the data, providing an accurate data source for multimodal data fusion analysis.

[0070] Temperature sensors are installed on critical components such as the motor housing to monitor the motor's operating temperature in real time; vibration sensors are installed on the motor's base or key components to detect vibrations during operation; voltage and current sensors are connected to the motor's power supply circuit to collect voltage and current data in real time; speed and torque sensors are installed near the motor's shaft to obtain the motor's speed and torque information. These sensors collect data in real time during motor operation and transmit the collected data synchronously to the relevant processing modules in the multimodal data fusion layer via data transmission lines, such as cables or wireless transmission modules.

[0071] The multimodal data fusion layer includes a Transformer temporal large-scale model, a cognitive agent, a digital twin model, and a generative design module, such as... Figure 2 As shown.

[0072] Transformer Temporal Model: The Transformer architecture boasts powerful sequence modeling capabilities, fully capturing long-term dependencies and temporal characteristics in sensor signals. This model can process massive amounts of signals from tens of thousands of sensors in real time, such as current, temperature, vibration, and speed signals. Through learning and analysis of vast amounts of historical and real-time data, it accurately predicts the operating status of motors. For example, it can provide early warnings of abnormal power consumption and sudden load changes, and generate corresponding optimization strategies, such as adjusting motor control parameters and optimizing operating modes, to improve motor efficiency and reliability, and reduce energy consumption and failure risks. When predicting motor load changes, it can comprehensively consider the changing trends of multiple variables such as current and speed. Furthermore, the model is pre-trained using large-scale historical motor operating data, covering motor operation under different conditions, loads, and environments, to improve the model's generalization ability across various scenarios. During pre-training, self-supervised learning is employed to allow the model to learn patterns and features from the temporal data. Furthermore, it integrates multiple tasks such as time series prediction, anomaly detection, and data imputation into a unified model training framework. For example, while training for time series prediction, it uses some data containing missing and outlier values ​​to train the model's anomaly detection and imputation capabilities.

[0073] Specifically, data preprocessing: The various sensor data received from the data acquisition layer are first preprocessed, including data cleaning, normalization and other operations, to ensure the quality and consistency of the data and facilitate subsequent model processing.

[0074] Model Training: A large-scale historical operating data set of motors, including operating data of different models under various working conditions, loads, and environmental conditions, is used to pre-train the Transformer time-series model. During pre-training, self-supervised learning methods are employed, such as predicting missing values ​​in the data sequence or the value at the next time step, allowing the model to learn patterns and features in the time-series data. After pre-training, the model is fine-tuned according to the actual application scenario and requirements to better adapt it to the specific characteristics of the motor's operating data.

[0075] Real-time Operation: During real-time motor operation, the model receives multivariate time-series data from the data acquisition layer. Through its internal attention mechanism and multi-layer neural network structure, it performs in-depth analysis of the data, capturing long-term dependencies and complex features. For example, when analyzing motor load changes, it considers the changes in multiple parameters such as current, speed, and torque simultaneously. Using a multi-step autoregressive approach, it makes multi-step predictions of the motor's future operating state, such as predicting the current change trend over the next 10 minutes. Simultaneously, a predictive anomaly detection method is employed, comparing the model-generated expected normal value sequence with the actual acquired values. When the error exceeds a set threshold, the model determines that the motor is operating abnormally and issues an alarm.

[0076] Specifically, in terms of time series forecasting, a multi-step autoregressive approach is used to output a sequence segment with each inference step, enabling multi-step predictions of the motor's future operating state, such as predicting current changes, temperature trends, and load fluctuations over a future period. During the forecasting process, the model can generate accurate prediction results based on long-term dependencies and trend information in historical data, helping to formulate maintenance plans and optimization strategies in advance.

[0077] In terms of anomaly detection, after fine-tuning the model on the normal sequence, it generates an expected normal value sequence based on the input real-time data, compares it with the actual collected values, and gives the confidence level of the abnormal interval based on the comparison error. When the error exceeds the set threshold, it is determined that the motor operation is abnormal, and an alarm is issued in a timely manner.

[0078] The cognitive intelligent agent, built upon the architecture of a cognitive engine, systematically organizes and stores a wealth of internal enterprise knowledge resources, including motor design manuals, motor control theory, automation theory, fault case libraries, and expert experience, constructing a knowledge graph. In actual operation, this module, combining data results from the Transformer time-series model, uses reasoning and matching from the knowledge graph to quickly and accurately diagnose motor faults, automatically generating detailed diagnostic reports including fault location, cause, and development trend, and providing specific operational guidance such as maintenance suggestions and parameter adjustment schemes. This effectively reduces reliance on human experience and improves the accuracy and efficiency of fault diagnosis.

[0079] This study employs various algorithms, including supervised learning, unsupervised learning, and reinforcement learning, to mine and analyze data from the knowledge graph. For example, clustering algorithms are used to classify motor fault cases to better understand and organize fault knowledge. The knowledge graph embedding model is first pre-trained on a large-scale general knowledge graph to learn common entity and relation representations, and then fine-tuned on a professional knowledge graph in the motor field to better reflect the characteristics of motor-related knowledge. A continuous learning mechanism is established to regularly update the data in the knowledge graph and retrain the model to adapt to the continuous development and changes in motor-related knowledge. Simultaneously, online learning technology is used to adjust model parameters in real time, improving the model's accuracy and timeliness.

[0080] Specifically, knowledge graph construction involves collecting internal enterprise knowledge resources such as motor design manuals, fault case libraries, and expert experience. This knowledge is then structured to extract entities and relationships, such as motor components, fault types, and fault causes. These entities and relationships are then linked and organized to construct a knowledge graph for the motor domain. For example, a specific fault type in a motor can be associated with possible fault causes and solutions.

[0081] Reasoning and Diagnosis: During motor operation, the cognitive agent receives data processed by the Transformer time-series model in real time and performs reasoning analysis in conjunction with the knowledge graph. For example, when an abnormal increase in motor temperature is detected, the reasoning engine searches the knowledge graph for information related to the temperature anomaly, analyzes possible causes of the fault, such as overload or cooling fan failure, and generates a detailed diagnostic report, including the fault location, cause, development trend, and corresponding maintenance suggestions and parameter adjustment schemes.

[0082] Continuous learning: Regularly collect new motor fault cases, technological updates, and other information to update the knowledge graph. Simultaneously, utilize online learning technology to adjust model parameters in real time when new motor operating data becomes available, enabling the cognitive agent to continuously adapt to the development and changes in motor-related knowledge, thereby improving the accuracy and timeliness of fault diagnosis.

[0083] Digital Twin Model: Based on a knowledge graph constructed using intelligent agents, and utilizing specialized modeling software and algorithms, a high-precision digital twin model corresponding to the physical motor is built. This model includes not only the motor's static attributes such as geometry and appearance, but also its dynamic behaviors, including physical characteristics, operating principles, and performance indicators, realistically reflecting the motor's operating status under different working conditions. By combining historical and real-time data with the digital twin model, simulation and predictive analysis of motor operation are performed. By simulating the motor's operation under different conditions, potential faults and problems can be predicted in advance, providing decision support for motor maintenance and optimization.

[0084] Specifically, model building involves using specialized modeling software to construct a geometric model of the motor based on its design drawings and physical parameters. Building upon this, and considering the motor's operating principles and physical characteristics, appropriate material properties and boundary conditions are added to create a digital twin model that incorporates both the motor's static and dynamic behaviors. For example, this simulates the motor's electromagnetic performance and temperature distribution under different speeds and loads.

[0085] Simulation and Prediction: During motor operation, the digital twin model receives real-time operating data from the processed Transformer time-series model, updating and calibrating the model in real time to ensure consistency between the model and the actual motor state. Combining historical and real-time data, simulation analysis is performed on the motor's operation under different conditions to predict potential faults and problems. For example, by simulating temperature changes during prolonged high-load operation, the model predicts the likelihood of overheating faults, providing decision support for motor maintenance and optimization.

[0086] Generative AI Design: Building upon knowledge graphs and digital twin models, the system automatically generates a suitable motor design and optimal control algorithm by simply inputting motor performance requirements such as torque and energy efficiency. The generated design is then validated through multi-dimensional simulations using the digital twin model, including simulations of the motor's electromagnetic, mechanical, and thermal performance to assess its feasibility and reliability. During validation, the design is iteratively optimized based on simulation results until the optimal design effect is achieved. The digital twin model performs real-time simulation analysis on the generated motor design, providing detailed performance feedback to the generative AI design optimization module. This module adjusts the design in a timely manner based on the feedback, creating a tight closed loop between design and simulation. A virtual test bench is constructed using digital twin technology to comprehensively test and validate the generated motor design. Various operating conditions and extreme scenarios of the motor are simulated in the virtual environment, allowing for early detection of potential problems and reducing actual testing costs and risks.

[0087] Through the analysis and design of the multimodal data fusion layer, an optimal control algorithm or motor design scheme is output to the control layer. The control algorithm generated by the execution object is then tested and verified. The feedback data is then input into the multimodal data fusion layer for analysis and comparison, forming an experimental closed loop to achieve continuous iteration of motor optimization.

[0088] Specifically, users input motor performance requirements into the system interface, such as a torque requirement of 50 N*m and an energy efficiency rating of Level 1. The generative AI design module, based on a diffusion model, automatically generates motor control algorithm schemes that meet these requirements, combining the motor's physical characteristics and design rules. These schemes include detailed information such as the algorithm's structure and parameter settings. Figure 2 The above-mentioned multimodal data fusion mechanism is shown.

[0089] Simulation Verification and Optimization: The generated design scheme automatically calls the digital twin model to perform multi-dimensional simulation verification. In the virtual environment constructed by the digital twin model, the simulated motor exhibits electromagnetic, mechanical, and thermal performance characteristics according to the design scheme. Based on the simulation results, the design scheme is evaluated. If problems such as insufficient torque at high speeds are found, the generative AI design optimization module adjusts and optimizes the scheme based on the evaluation results, performs simulation verification again, and iterates repeatedly until the optimal design effect is achieved. The design scheme or control algorithm is then input into the controller.

[0090] By performing specific operations on the Matlab interface, the one-click executable code generation function can be triggered. The code generation module inside the hardware-in-the-loop controller will automatically generate executable code suitable for actual hardware operation based on the algorithm model in Matlab and according to preset rules and algorithms, and download it to the core processing unit of the controller for subsequent execution.

[0091] Simultaneously, the CCS software development platform can be used to program and implement the control algorithm according to the design scheme or control algorithm. The control algorithm code is written by configuring registers and writing functions for MCU control chips such as DSPs or STM32s. After completion, the code is sent to the execution object for experimental verification.

[0092] Example 2

[0093] The parts not described in detail in this embodiment are as shown in Embodiment 1. This embodiment provides a method for a smart motor test bench based on multimodal data fusion, including the following steps:

[0094] S101. Execute the motor control algorithm in the control layer, output the running results including control instructions and control cycle, and dynamically adjust the running mode in the execution object according to the control instructions to generate running data;

[0095] S102. The data acquisition layer determines the sampling frequency or acquisition channel based on the operating mode, and uses the operating result as a synchronization reference signal to trigger the sensor to synchronously acquire the operating data within the control cycle.

[0096] S103. In the multimodal data fusion layer, the Transformer time series large model is used to predict the running data, and the predicted sequence is compared and reasoned with the knowledge graph in the cognitive agent to generate diagnostic information of potential faults or anomalies.

[0097] S104. In the multimodal data fusion layer, the digital twin model is dynamically calibrated based on the diagnostic information, and the generative design module generates a motor design scheme or control parameter optimization scheme based on the prediction results of the virtual model. After iterative verification in the virtual bench environment, the scheme is fed back to the control layer.

[0098] The control layer includes a hardware-in-the-loop (HIL) controller based on RCP and HIL, and an MCU core control board based on MCU, wherein:

[0099] The hardware-in-the-loop simulation controller is used to automatically generate executable code adapted to the hardware from the modeling software.

[0100] The MCU core control board is used to execute the code and output the running results, including the control cycle, which serve as the synchronization reference signal for the data acquisition layer.

[0101] When the execution object detects sudden load changes, environmental changes, or abnormal torque fluctuations...

[0102] Automatically adjusts the speed and torque parameters in the operating mode.

[0103] And the information on the change in operating mode is fed back to the data acquisition layer.

[0104] This enables dynamic matching between the data collection strategy and the state of the execution object.

[0105] The data acquisition layer includes various types of heterogeneous sensors.

[0106] During the sampling process, the control cycle output by the control layer is used as the trigger condition.

[0107] The sensor is driven to achieve synchronous acquisition within the same control cycle, wherein: when the operation mode of the execution object is detected to be switched, the sampling frequency of the sensor is automatically adjusted or the acquisition channel is switched.

[0108] The Transformer time series large model includes an anomaly detection module.

[0109] The anomaly detection module is used to generate a sequence of future operating states simultaneously;

[0110] The real-time collected operational data is compared with the expected values.

[0111] When the comparison error exceeds the threshold, an abnormal alarm is output and the prediction step size is dynamically shortened.

[0112] The cognitive agent includes a knowledge graph construction module and an inference engine module:

[0113] The knowledge graph construction module is used to structure and store motor design manuals, fault cases and expert experience to form a knowledge graph in the field of motors, which organizes motor components, fault types, fault causes and maintenance solutions into entities and relationships.

[0114] The inference engine module is used to receive future operating state sequences and perform matching inference in the knowledge graph to generate diagnostic information containing fault location, cause and development trend.

[0115] The knowledge graph construction module includes an online learning unit:

[0116] This is used to dynamically update the entity relationships in the knowledge graph when new operational data or new failure cases are received.

[0117] And adjust the parameters of the inference engine in real time.

[0118] This enables diagnostic capabilities to continuously evolve as the motor operates.

[0119] The digital twin model includes a dynamic calibration unit.

[0120] This is used to correct the parameters of the virtual motor model after receiving the diagnostic information.

[0121] Simulations were run and predictions were made under different load and environmental conditions.

[0122] This allows for the early identification of potential failure risks and maintenance time windows.

[0123] The generative design module includes a virtual bench verification unit.

[0124] Used after generating motor design schemes or control parameter optimization schemes.

[0125] The digital twin model is used to simulate extreme working conditions and boundary conditions in a virtual environment.

[0126] The proposed solution is then iteratively optimized based on simulation feedback.

[0127] Until the target performance requirements are met.

[0128] The method for implementing the intelligent motor test bench based on multimodal data fusion provided in the above embodiments of the present invention, and the specific methods and processes for realizing the corresponding functions of each structure in the intelligent motor test bench system based on multimodal data fusion are detailed in the above embodiments of the intelligent motor test bench system based on multimodal data fusion, and will not be repeated here.

[0129] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A smart motor test bench system based on multimodal data fusion, characterized in that, include: The control layer is used to execute object control algorithms and output running results, which include control instructions and control cycles. An execution object is used to dynamically adjust its operating mode according to the control instructions. The data acquisition layer determines the sampling frequency or acquisition channel based on the operating mode of the execution object, and uses the operation result as a synchronization reference signal to trigger the sensor to synchronously acquire operation data within the control cycle. The multimodal data fusion layer includes: The Transformer time series large model is used to perform multi-step autoregressive prediction on the running data to obtain a sequence of future running states. A cognitive intelligent agent is used to compare and reason with the sequence of future operating states and a knowledge graph, and to generate diagnostic information when potential faults or anomalies are identified. A digital twin model is used to dynamically calibrate a virtual motor model based on the diagnostic information and to simulate and predict operating behavior under different working conditions. The generative design module is used to automatically generate motor design schemes or control parameter optimization schemes based on the prediction results of the virtual motor model, and to perform iterative verification in a virtual test bench environment.

2. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, The control layer includes a hardware-in-the-loop (HIL) controller based on RCP and HIL, and an MCU core control board based on MCU, wherein: The hardware-in-the-loop simulation controller is used to automatically generate executable code adapted to the hardware from the modeling software. The MCU core control board is used to execute the code and output the running results, including the control cycle, which serve as the synchronization reference signal for the data acquisition layer.

3. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, When the execution object detects sudden load changes, environmental changes, or abnormal torque fluctuations... Automatically adjusts the speed and torque parameters in the operating mode. And the information on the change in operating mode is fed back to the data acquisition layer. This enables dynamic matching between the data collection strategy and the state of the execution object.

4. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, The data acquisition layer includes various types of heterogeneous sensors. During the sampling process, the control cycle output by the control layer is used as the trigger condition. The sensor is driven to achieve synchronous acquisition within the same control cycle, wherein: when the operation mode of the execution object is detected to be switched, the sampling frequency of the sensor is automatically adjusted or the acquisition channel is switched.

5. The intelligent motor test bench system based on multimodal data fusion according to claim 4, characterized in that, The Transformer time series large model includes an anomaly detection module. The anomaly detection module is used to generate a sequence of future operating states simultaneously; The real-time collected operational data is compared with the expected values. When the comparison error exceeds the threshold, an abnormal alarm is output and the prediction step size is dynamically shortened.

6. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, The cognitive agent includes a knowledge graph construction module and an inference engine module: The knowledge graph construction module is used to structure and store motor design manuals, fault cases and expert experience to form a knowledge graph in the field of motors, which organizes motor components, fault types, fault causes and maintenance solutions into entities and relationships. The inference engine module is used to receive future operating state sequences and perform matching inference in the knowledge graph to generate diagnostic information containing fault location, cause and development trend.

7. The intelligent motor test bench system based on multimodal data fusion according to claim 6, characterized in that, The knowledge graph construction module includes an online learning unit: This is used to dynamically update the entity relationships in the knowledge graph when new operational data or new failure cases are received. And adjust the parameters of the inference engine in real time. This enables diagnostic capabilities to continuously evolve as the motor operates.

8. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, The digital twin model includes a dynamic calibration unit. This is used to correct the parameters of the virtual motor model after receiving the diagnostic information. Simulations were run and predictions were made under different load and environmental conditions. This allows for the early identification of potential failure risks and maintenance time windows.

9. The intelligent motor test bench system based on multimodal data fusion according to claim 1, characterized in that, The generative design module includes a virtual bench verification unit. Used after generating motor design schemes or control parameter optimization schemes. The digital twin model is used to simulate extreme working conditions and boundary conditions in a virtual environment. The proposed solution is then iteratively optimized based on simulation feedback. Until the target performance requirements are met.

10. A method for a smart motor test bench based on multimodal data fusion, implemented based on the smart motor test bench system based on multimodal data fusion as described in any one of claims 1-9, characterized in that, Includes the following steps: S101. Execute the motor control algorithm in the control layer, output the running results including control instructions and control cycle, and dynamically adjust the running mode in the execution object according to the control instructions to generate running data; S102. The data acquisition layer determines the sampling frequency or acquisition channel based on the operating mode, and uses the operating result as a synchronization reference signal to trigger the sensor to synchronously acquire the operating data within the control cycle. S103. In the multimodal data fusion layer, the Transformer time series large model is used to predict the running data, and the predicted sequence is compared and reasoned with the knowledge graph in the cognitive agent to generate diagnostic information of potential faults or anomalies. S104. In the multimodal data fusion layer, the digital twin model is dynamically calibrated based on the diagnostic information, and the generative design module generates a motor design scheme or control parameter optimization scheme based on the prediction results of the virtual model. After iterative verification in the virtual bench environment, the scheme is fed back to the control layer.

Images