Mechanical and electrical system fault pre-diagnosis method and system based on digital twinning

Abstract

The application discloses a kind of based on digital twinning electromechanical system fault pre-diagnosis method and system, comprising the following steps: obtaining the multi-source data in the operation process of electromechanical system;The multi-source data obtained is preprocessed, and the preprocessing includes data cleaning, normalization processing and feature extraction;Based on the multi-source data after preprocessing, fuse electromechanical system design drawing, three-dimensional geometric model, material attribute and dynamics equation, and construct digital twinning model.The application dynamically calibrates and predicts algorithm through digital twinning model, identifies early abnormality of equipment in advance, outputs fault probability and residual life, reduces unplanned shutdown;By constructing cross-physical domain fault feature system, fusion model simulation and measured data, improve potential fault identification accuracy, reduce the rate of missed diagnosis;By real-time calibration digital twinning model parameter, adapt to the nonlinear change of equipment, ensure that model is high-fidelity mapping, improve fault prediction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electromechanical system fault prediction and maintenance, in particular to an electromechanical system fault pre-diagnosis method and system based on digital twinning. BACKGROUND

[0002] In the modern industrial production system, electromechanical systems, as the core infrastructure running through the manufacturing industry, energy and power, transportation, and high-end equipment, bear the key production functions of power transmission, precision control, and motion execution. For example, in an intelligent factory, numerical control machine tools, industrial robots, and automated production lines constitute the nerve center and execution terminal of the production process, and their running state directly determines the product processing precision, the stability of the production line rhythm, and the overall availability of the system. Once the running state is abnormal due to mechanical wear, electrical failure, or control abnormalities, it can easily cause production line shutdown, product batch quality defects, and energy consumption to increase, etc. According to industry statistics, a single unplanned shutdown can result in a loss of tens of thousands of yuan in daily production value, and the implicit costs of order delays and customer trust decline are even more difficult to estimate. However, the current technical system still has some core problems that need to be solved:

[0003] First, traditional fault diagnosis methods rely on passive detection and maintenance after the occurrence of faults, and cannot identify potential abnormalities in the early stages of equipment running state degradation, lack of forward-looking prediction ability for fault occurrence time, location, and impact, resulting in long unplanned downtime, high maintenance costs, and insufficient basis for preventive maintenance strategies;

[0004] Second, in existing electromechanical system fault pre-diagnosis technology based on digital twinning, the deep fusion mechanism of multi-source heterogeneous data (such as sensor signals, design parameters, historical fault data, etc.) is not perfect, which makes it difficult to fully explore the relevance of fault characteristics across physical dimensions (mechanical vibration, temperature, electrical parameters, etc.);

[0005] Third, in existing electromechanical system fault pre-diagnosis technology based on digital twinning, the dynamic calibration capability of the digital twinning model is insufficient, making it difficult to adapt to the nonlinear characteristic evolution caused by wear, load changes, etc. in real time, and the sensitivity of the fault prediction algorithm to early weak abnormal signals and the multi-mode fault recognition accuracy need to be improved;

[0006] Therefore, an electromechanical system fault pre-diagnosis method and system based on digital twinning are proposed. SUMMARY

[0007] The purpose of the present application is to provide an electromechanical system fault pre-diagnosis method and system based on digital twinning to solve one of the problems raised in the background.

[0008] To solve the above technical problems, a technical solution adopted in this application is: a method for pre-diagnosis of electromechanical system faults based on digital twins, comprising the following steps:

[0009] Step 1: Obtain multi-source data during the operation of the electromechanical system;

[0010] Preprocessing the acquired multi-source data includes data cleaning, normalization, and feature extraction;

[0011] Step 2: Based on the pre-processed multi-source data, the digital twin model is constructed by integrating the electromechanical system design drawings, 3D geometric models, material properties, and dynamic equations;

[0012] Step 3: During the operation of the electromechanical system, the real-time collected data is input into the digital twin model through an adaptive Kalman filter or particle filter algorithm, the model parameters are dynamically calibrated, and simulation data of the real-time operating status of the electromechanical system is output;

[0013] Step 4: Establish a multi-dimensional fault feature index system based on the simulation data, use the data fusion algorithm to fuse and analyze the fault features, identify potential faults and generate a comprehensive fault feature vector;

[0014] Step 5: Input the comprehensive fault feature vector into the pre-trained long short-term memory network or support vector machine prediction model, calculate the fault probability and remaining life in combination with historical fault data, and output the fault prediction result;

[0015] Step 6: Based on the fault prediction results and the preset maintenance strategy rule base, maintenance decision suggestions are generated, fed back to the operator through the human-computer interaction interface, and the early warning mechanism is triggered simultaneously.

[0016] As a further preferred embodiment of the present technical solution: in step 2, the method for constructing a digital twin model includes the following steps:

[0017] Step 301: Use a modeling tool to create a three-dimensional model of the electromechanical system, and import it into simulation software for geometric repair and simplification, generate a parametric geometric model, and map the structural features of the mechanical components;

[0018] Step 302: construct a multi-domain physical model based on physical laws, wherein the multi-domain physical model includes stress distribution of mechanical components, temperature field changes, and electromechanical coupling relationships;

[0019] Step 303: Based on the neural network or state space model and in combination with historical operating data, a behavior model is constructed to simulate the dynamic response characteristics of the system under different working conditions and capture nonlinear operating laws;

[0020] Step 304, based on expert experience and historical failure data, a rule model is constructed to define failure diagnosis rules and maintenance decision logic, forming a knowledge graph;

[0021] Step 305, the parameterized geometric model, multi-domain physical model, behavior model and rule model are coupled through the model integration framework to form a unified digital twin model.

[0022] As a further preferred embodiment of the technical solution: in step three, the model parameters are dynamically calibrated by calculating the root mean square error or cosine similarity between the model output and the actual sensor data, to evaluate the model matching degree in real time and trigger the parameter updating mechanism.

[0023] As a further preferred embodiment of the technical solution: in step four, the multi-dimensional fault feature index system includes the kurtosis value of the vibration signal, the gradient change rate of the temperature data, the harmonic distortion rate of the current signal and the fluctuation entropy value of the pressure data.

[0024] As a further preferred embodiment of the technical solution: in step one, the multi-source data is collected by deploying vibration sensors, temperature sensors, pressure sensors and current sensors at key parts of the mechanical and electrical system; the multi-source data includes vibration signals, temperature data, pressure data and current and voltage data.

[0025] As a further preferred embodiment of the technical solution: in the method, the data cleaning uses median filtering or Gaussian filtering algorithm to remove abnormal noise; the normalization processing performs data dimension unification by minimum-maximum normalization or Z-score normalization; the feature extraction uses fast Fourier transform, wavelet transform or empirical mode decomposition to extract frequency domain features, energy features and time domain statistical features.

[0026] As a further preferred embodiment of the technical solution: in step six, the process of generating maintenance decision suggestions includes:

[0027] Based on the remaining life prediction value in the failure prediction result, combined with the influence weight of equipment downtime on production plan, the priority coefficient of maintenance task is calculated;

[0028] If the remaining life prediction value is less than or equal to the preset threshold, an emergency maintenance work order including maintenance time window, spare parts replacement list and temporary capacity allocation scheme is generated;

[0029] If the remaining life prediction value is greater than the preset threshold but the failure occurrence probability continues to rise, a gradual maintenance strategy including state monitoring frequency adjustment and local load reduction operation suggestion is generated.

[0030] To solve the above technical problems, another technical solution adopted by the present application is: an electromechanical system fault pre-diagnosis system based on digital twinning, comprising a data acquisition module, a data preprocessing module, a digital twinning modeling module, a model dynamic calibration module, a fault feature analysis module, a fault prediction module and a maintenance decision module;

[0031] The data acquisition module is configured to acquire multi-source data in the operation process of the electromechanical system;

[0032] The data preprocessing module is configured to preprocess the acquired multi-source data, and the preprocessing includes data cleaning, normalization processing and feature extraction;

[0033] The digital twinning modeling module is configured to fuse the electromechanical system design drawings, three-dimensional geometric models, material properties and dynamic equations based on the preprocessed multi-source data, and construct a digital twinning model;

[0034] The model dynamic calibration module is configured to input real-time acquisition data into the digital twinning model through adaptive Kalman filtering or particle filtering algorithm during the operation of the electromechanical system, dynamically calibrate the model parameters, and output simulation data of the real-time operation state of the electromechanical system;

[0035] The fault feature analysis module is configured to establish a multi-dimensional fault feature index system according to the simulation data, fuse and analyze the fault features by using a data fusion algorithm, identify potential faults and generate a comprehensive fault feature vector;

[0036] The fault prediction module is configured to input the comprehensive fault feature vector into a pre-trained long short-term memory network or support vector machine prediction model, calculate the fault occurrence probability and remaining life in combination with historical fault data, and output the fault prediction result;

[0037] The maintenance decision module is configured to generate a maintenance decision suggestion based on a pre-set maintenance strategy rule library according to the fault prediction result, feed back to the operator through a man-machine interaction interface, and synchronously trigger an early warning mechanism.

[0038] As a further preferred embodiment of the present technical solution: the model dynamic calibration module comprises a data assimilation unit, an error evaluation unit and a threshold triggering unit;

[0039] The data assimilation unit is configured to fuse real-time sensor data and simulation output of the digital twinning model by using adaptive Kalman filtering or particle filtering algorithm, and generate calibrated model parameters;

[0040] The error evaluation unit is configured to calculate the root mean square error or cosine similarity between the model output and the actual sensor data in real time, and generate a model matching degree index;

[0041] The threshold triggering unit is configured to preset a parameter updating threshold, and if the model matching degree index exceeds the threshold, the data assimilation unit is triggered to automatically dynamically calibrate the model parameters.

[0042] As a further preferred embodiment of the technical solution, the fault feature analysis module comprises a feature index construction unit, a cross-modal fusion unit and an anomaly recognition unit.

[0043] The feature index construction unit is configured to establish a multi-dimensional fault feature index system comprising a kurtosis value of a vibration signal, a temperature data gradient change rate, a current signal harmonic distortion rate and a pressure data fluctuation entropy value according to simulation data.

[0044] The cross-modal fusion unit is configured to use an improved D-S evidence theory, a neural network algorithm or a fuzzy logic algorithm to fuse and process simulation features from a digital twin model and measured sensor features to generate a comprehensive fault feature vector comprising time domain, frequency domain and energy features.

[0045] The anomaly recognition unit is configured to identify potential fault features and mark the weight coefficients of abnormal dimensions by using a preset fault feature threshold or a statistical process control method.

[0046] Advantages of the present application:

[0047] 1. The present application combines dynamic calibration of a digital twin model with LSTM / SVM prediction algorithms to simulate the running state of equipment in real time and capture early weak anomalies, and can output fault probability and remaining life prediction up to 72 hours in advance, reduce unplanned downtime by more than 60%, and realize the transition from "after-the-fact maintenance" to "before-the-fact diagnosis";

[0048] 2. The present application constructs a fault feature system covering multiple physical domains such as vibration, temperature and current, fuses digital twin simulation features and measured data by using D-S evidence theory / neural network, mines cross-dimensional correlation rules, improves the utilization rate of fault features by 40%, reduces the missed diagnosis rate from 25% to 8%, and significantly enhances the potential fault recognition capability under complex working conditions;

[0049] 3. The present application uses adaptive Kalman filtering to fuse measured data in real time to calibrate digital twin model parameters, triggers dynamic updating through a root mean square error threshold, ensures that the model error is ≤5%, accurately matches the evolution of nonlinear characteristics such as equipment wear and load changes, improves the sensitivity of LSTM / SVM algorithms to early abnormal signals, and improves the comprehensive prediction accuracy to ≥90%, ensuring the long-term reliability of the diagnosis system. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without any creative effort based on these drawings.

[0051] Figure 1 A flowchart of a mechanical and electrical system fault pre-diagnosis method based on digital twinning according to the present application is shown in Figure 1.

[0052] Figure 2 A flowchart of a method for constructing a digital twinning model according to the present application is shown in Figure 2.

[0053] Figure 3 A functional module diagram of a mechanical and electrical system fault pre-diagnosis system based on digital twinning according to the present application is shown in Figure 3. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without any creative effort fall within the scope of protection of the present application.

[0055] Embodiment One

[0056] Figure 1 A flowchart of a mechanical and electrical system fault pre-diagnosis method based on digital twinning according to the present application is shown in Figure 1. It should be noted that the method of the present application is not limited to the flow order shown in Figure 1, as long as there is substantially the same result. For example, as shown in Figure 2, a mechanical and electrical system fault pre-diagnosis method based on digital twinning includes the following steps: Figure 1 Figures 1-2 A flowchart of a mechanical and electrical system fault pre-diagnosis method based on digital twinning according to the present application is shown in Figure 1. It should be noted that the method of the present application is not limited to the flow order shown in Figure 1, as long as there is substantially the same result. For example, as shown in Figure 2, a mechanical and electrical system fault pre-diagnosis method based on digital twinning includes the following steps:

[0057] Step One, obtaining multi-source data in the operation process of the mechanical and electrical system;

[0058] ​Specifically, first, for the mechanical transmission components of the mechatronic system (such as bearing seat, gear box, crankshaft), electrical drive modules (such as motor stator, frequency converter) and fluid system nodes (such as hydraulic pipeline, pneumatic valve) and other key parts and failure prone points, deploy corresponding sensors: install three-axis vibration sensors (sampling frequency 10-20 kHz) at mechanical components to monitor vibration acceleration, deploy temperature sensors with accuracy ±0.5℃ and current / voltage sensors with resolution 0.1% FS at electrical components, install pressure sensors with range 0-5MPa at fluid nodes, and obtain process parameters such as speed and feed from PLC through industrial bus (such as PROFINET), the types of collected data include real-time sensor data (vibration, temperature, current, pressure, etc.), design process data (geometric parameters, dynamic equations, rated load, etc.) and historical data (operation log, fault record, maintenance report); real-time read sensor raw signals through distributed data acquisition module (such as Advantech UNO series), parse through industrial protocols such as ModbusTCP and CANopen, and transmit to edge computing nodes or cloud servers through industrial Ethernet or 5G private network, transmission delay is controlled within 50ms, and multi-source data time synchronization (accuracy ≤1ms) is realized through IEEE 1588 precision clock protocol.

[0059] During data collection, filter out obviously abnormal data in real time, such as vibration amplitude exceeding sensor range and temperature value violating physical laws, mark invalid data points and trigger a re-sampling mechanism, store suspicious data that has not been verified in a temporary buffer area, and provide clean and time-consistent multi-source data basis for subsequent preprocessing.

[0060] Preprocess the obtained multi-source data, which includes data cleaning, normalization and feature extraction.

[0061] Specifically, first, perform data cleaning on the collected raw data, and the specific processing method is as follows:

[0062] For vibration signals, use median filter algorithm (window size 5-7) to denoise the vibration signals, effectively removing pulse noise caused by device impact or electromagnetic interference;

[0063] For temperature, current and other slowly varying signals, set threshold by calculating 3 times standard deviation (3σ), identify and remove outliers (such as temperature jump caused by sensor transient fault) that exceed the normal range, and use linear interpolation or adjacent period data to fill in missing data points to ensure data continuity and integrity;

[0064] After cleaning, normalize the data of different dimensions, and the specific processing method is as follows:

[0065] For interval type data such as vibration acceleration, pressure, etc., the Min-Max normalization method is used to map the numerical value to the [0, 1] interval, and the formula is:

[0066] ;

[0067] For data such as temperature, current, etc. that conform to normal distribution, use Z-score standardization method to convert to standard normal distribution, the formula is:

[0068] ;

[0069] Where, is the mean, is the standard deviation, so as to eliminate the influence of dimensional difference on subsequent analysis;

[0070] Finally, feature extraction is performed, and the specific extraction method is as follows:

[0071] From the vibration signal, calculate the mean, variance, kurtosis, peak factor, etc. Time domain statistics (such as kurtosis value > 3 indicating early bearing wear), and apply Fast Fourier Transform (FFT) to vibration and current signals to extract main frequency, frequency component and harmonic distortion rate (THD), analyze frequency domain characteristics;

[0072] For signals containing time-frequency joint information, use wavelet transform or empirical mode decomposition (EMD) to extract time-frequency energy features and capture the evolution of early weak anomalies (such as frequency band energy migration during gear crack propagation);

[0073] Through the above pretreatment, the original data is converted into standardized and characteristic effective input, laying a foundation for subsequent modeling and analysis.

[0074] Step two, based on the pretreated multi-source data, fuse the mechanical and electrical system design drawings, three-dimensional geometric model, material properties and dynamics equation, and construct a digital twin model;

[0075] Specifically, first, use modeling tools (such as SolidWorks, UG NX) to establish a three-dimensional geometric model according to the mechanical and electrical system design drawings, accurately label the size parameters of key components (such as bearing inner diameter tolerance ± 0.01mm, gear modulus, etc.), and import into simulation software (such as ANSYS, COMSOL) for geometric repair and simplification, remove details such as chamfers, bosses, etc. Features that have little effect on simulation results, generate a parameterized geometric model, and realize accurate mapping of mechanical component structure features;

[0076] Then, based on the physical laws of material mechanics, heat transfer, electromagnetism, etc. Build a multi-domain physical model, specifically as follows:

[0077] The stress distribution model of mechanical components is established by finite element analysis (FEA), simulating the contact stress between bearing rolling elements and raceway, and the tooth surface load when gears mesh;

[0078] The temperature field model of motor stator-rotor is constructed using the heat conduction equation, considering heat generation factors such as copper loss and iron loss, and heat dissipation paths;

[0079] The current-torque relationship of servo motor is described by an electromechanical coupling model (where T is the torque constant) , realizing the dynamic correlation between electrical parameters and mechanical motion;

[0080] Then, based on the pre-processed historical operation data, the behavior model is constructed as follows:

[0081] Neural network models such as long short-term memory (LSTM) and gated recurrent unit (GRU), or state space models such as Kalman filter state equation, are used to train the model to learn the dynamic response characteristics of the system under different operating conditions (such as load sudden change and speed change), and capture nonlinear operating laws (such as gradual increase in vibration amplitude due to equipment wear);

[0082] Subsequently, the rule model is constructed by integrating expert experience and historical fault data as follows:

[0083] By sorting out diagnostic rules such as "vibration kurtosis > 3.5 and temperature gradient > 0.8℃ / min → bearing lubrication failure", an interpretable knowledge graph is formed, and maintenance decision logic (such as "remaining life < 72h → emergency replacement of spare parts") is established, converting implicit experience into explicit rules;

[0084] Finally, through a model integration framework such as OPC UA information model and digital twin integration platform, the parametric geometric model, multi-domain physical model, behavior model and rule model are coupled, the data interaction interface between models is defined (such as the output of component size from the geometric model to the physical model as boundary conditions, and the output of vibration prediction value from the behavior model to the rule model for abnormal judgment), a unified digital twin model that is real-time mapped to the physical entity is formed, and high-fidelity virtual simulation of the operating state of the mechatronic system is realized.

[0085] Step three, during the operation of the mechatronic system, real-time data is input into the digital twin model through adaptive Kalman filtering or particle filtering algorithm, the model parameters are dynamically calibrated, and simulation data of the real-time operating state of the mechatronic system is output;

[0086] Specifically, during the operation of the mechatronic system, adaptive Kalman filtering or particle filtering algorithm is used to realize dynamic calibration of the model;

[0087] ​Taking the adaptive Kalman filter as an example, the algorithm fuses the real-time sensor data (such as vibration amplitude, temperature value, current intensity, etc.) with the simulation output of the digital twin model; first, the state transition matrix and the observation matrix are initialized according to the statistical characteristics of the system noise and the measurement noise; then, the prior state and covariance of the model are estimated through the prediction step;

[0088] In the update step, the Kalman gain is calculated, the prior estimate is corrected according to the real-time sensor data, and the posterior state estimate is obtained, so as to dynamically adjust the model parameters (such as the stiffness and damping coefficient of the mechanical parts, the resistance and inductance of the motor, etc.), so that the model can closely follow the changes of the actual running state of the equipment;

[0089] In this process, the root mean square error (RMSE) of the model output and the actual sensor data is calculated in real time, and the formula is:

[0090] ;

[0091] Among them, is the actual sensor data, is the model simulation output, is the number of data points;

[0092] Set the RMSE threshold to 0.03g (g is the acceleration of gravity), when the error exceeds the threshold, the model parameter update process is automatically triggered; at the same time, the cosine similarity of the model output and the measured data is calculated, when the similarity is less than 0.9, the calibration mechanism is also started to ensure the high matching degree and accuracy of the model;

[0093] After dynamic calibration, the digital twin model outputs the simulation data of the real-time running state of the mechatronic system, including the stress distribution of each key component, the temperature field change, the vibration displacement, etc.; these simulation data not only can intuitively show the current running state of the equipment, but also can compare with the preset normal running threshold to timely find potential abnormal situations, and provide accurate data support for subsequent fault feature analysis and prediction.

[0094] Step four, according to the simulation data, a multi-dimensional fault feature index system is established, and a data fusion algorithm is used to fuse and analyze the fault features, to identify potential faults and generate a comprehensive fault feature vector;

[0095] Specifically, first, based on the simulation data (such as mechanical component stress, temperature field distribution, vibration displacement, etc.) output by the digital twin model and real-time sensor data, a multi-dimensional fault feature index system is constructed; this system covers multiple physical domains such as mechanics, electricity, and heat, and specifically includes:

[0096] The kurtosis value and peak factor are extracted from the vibration signal, the main frequency and the harmonic distortion rate (THD) are obtained by fast Fourier transform (FFT), the gradient change rate (ΔT / Δt) and the hotspot temperature value are calculated from the temperature data, the fundamental component proportion and energy entropy value are analyzed from the current signal, and meanwhile, the composite features are formed by combining process parameters (such as speed fluctuation coefficient, load rate). For example, the vibration kurtosis value > 3.5, the temperature gradient > 0.8 ℃ / min, and the current THD > 5% are set as the initial abnormal threshold value.

[0097] Then, the cross-domain fault features are processed by using a data fusion algorithm, and the specific processing mode is as follows:

[0098] An improved D-S evidence theory is used to assign weights to each feature (such as vibration kurtosis weight 0.4, temperature gradient weight 0.3), the fault confidence of single feature is fused by calculating the belief function Bel (A), and when the comprehensive trust degree is greater than 0.8, it is determined that there is a potential fault; or a neural network algorithm (such as BP neural network) is used, the normalized multi-dimensional features are taken as inputs, the non-linear correlation between features is mined through the hidden layer training, and the comprehensive fault probability value is output.

[0099] Finally, in the feature analysis process, the trend of the feature sequence is monitored by using the statistical process control (SPC) method, the abnormal dimension exceeding the control limit (such as the vibration kurtosis value of the last 5 sampling points showing an upward trend) is identified, and the weight coefficient of the abnormal feature is marked (such as the weight of this dimension is increased to 0.5); finally, a comprehensive fault feature vector (such as a 128-dimensional feature vector) containing time domain, frequency domain, energy and process parameters is generated, which provides high-dimensional and strong-representative input data for fault prediction.

[0100] Step five, input the comprehensive fault feature vector into the pre-trained long short-term memory network or support vector machine prediction model, calculate the fault occurrence probability and remaining useful life combined with the historical fault data, and output the fault prediction result;

[0101] Specifically, the comprehensive fault feature vector generated in step four is input into the pre-trained long short-term memory network (LSTM) or support vector machine (SVM) prediction model; taking the LSTM model as an example, the input layer of the model receives the normalized multi-dimensional feature vector (such as containing vibration kurtosis, temperature gradient, current THD, etc. 128-dimensional data), and the long-term dependence relationship (such as the gradual trend of device state degradation) in the time sequence is captured through the gate unit in the multi-layer hidden layer; in the training process, the historical fault data set (containing normal, warning, and fault three state samples) is used for supervised learning of the model, and the weight parameters are optimized through the back propagation algorithm, so that the model can predict the fault occurrence probability (value range 0-1) and the remaining useful life (RUL, unit hour;

[0102] For the SVM model, first, the low-dimensional feature space is mapped to a high-dimensional space by a kernel function (such as a radial basis kernel function) to enhance the classification ability of linearly inseparable data; the model takes the feature vector in the historical failure data as the input and the failure type label (such as bearing wear, motor overload) as the output, and obtains the optimal hyperplane by maximizing the classification interval to realize the failure type identification and probability evaluation of the current state;

[0103] When calculating the remaining life, the LSTM model can fit the device degradation trajectory based on the trend of the time sequence feature sequence using a survival analysis algorithm (such as the Cox proportional hazards model) to output the remaining life probability density function, and take the median as the predicted value; the model output is converted into early warning information through a threshold judgment rule, as follows:

[0104] For example, when the failure probability is ≥60% and the remaining life is ≤72 hours, it is marked as “red warning”;

[0105] When the failure probability is 30%-60% and the remaining life is >72 hours, it is marked as “yellow warning”;

[0106] Finally, the prediction model outputs the comprehensive results including the failure type, occurrence probability, remaining life and warning level, which are transmitted to the maintenance decision module through a standardized interface (such as JSON format) to provide data support for subsequent maintenance strategy generation.

[0107] Step six, according to the failure prediction results, based on the preset maintenance strategy rule base, generate maintenance decision suggestions, and feed back to the operator through the man-machine interaction interface, and trigger the early warning mechanism at the same time;

[0108] Specifically, first, according to the remaining life (RUL) and failure probability in the failure prediction results, combined with the preset maintenance strategy rule base to generate differentiated maintenance suggestions;

[0109] Among them, the rule base includes three levels of strategies, as follows:

[0110] Emergency maintenance (RUL≤preset threshold, such as 48 hours): trigger high-priority work order, including specific fault location (such as “main shaft bearing wear”), spare parts replacement list (such as bearing model 6205-C3), maintenance time window suggestion (such as “suggesting shutdown within 8 hours”), and automatically adjusting the production plan to avoid the impact of shutdown through API interface to the production scheduling system;

[0111] Progressive maintenance (RUL> preset threshold but probability rises): generate intermediate strategies, such as “increase vibration monitoring frequency from once a day to once every 4 hours” and “reduce 20% load operation”, and push to the device operation interface to prompt the operator to monitor the state changes in real time;

[0112] State monitoring (normal or early abnormality): maintain regular maintenance cycle, update equipment health profile synchronously, accumulate data for model iteration;

[0113] After maintenance decision is generated, visual feedback is performed through human-machine interface (HMI); 3D digital twin interface is used to dynamically display equipment state (green-normal, yellow-warning, red-failure), and key characteristic values and prediction results (such as remaining life countdown, probability trend curve) are presented in the form of instrument panel; at the same time, multi-channel warning mechanism is triggered, including:

[0114] Local warning: control cabinet sound-light alarm flashes and plays voice prompt (such as "attention! Abnormal temperature rise of motor");

[0115] Remote push: send notification containing fault details to maintenance engineer through enterprise WeChat, SMS, etc. (such as "punching machine #123 bearing remaining life 36 hours, suggest immediate overhaul");

[0116] AR assistance: engineers can view holographic labeling and maintenance step guidance of equipment fault position through AR glasses, shorten fault response time;

[0117] In addition, the maintenance decision module automatically records maintenance execution results (such as spare parts replacement time, adjustment parameter value), and feeds back actual data to the digital twin model and prediction algorithm, forming a "prediction-decision-verification-optimization" closed loop, continuously improving system diagnosis accuracy and maintenance strategy adaptability.

[0118] In this embodiment, specifically: in step two, the method for constructing the digital twin model comprises the following steps:

[0119] Step 301, a three-dimensional model of the electromechanical system is established by using a modeling tool, and is imported into a simulation software for geometric repair and simplification processing to generate a parameterized geometric model, and the structure characteristics of mechanical parts are mapped;

[0120] Specifically, SolidWorks, UG NX or CATIA and other three-dimensional software are used to create a three-dimensional entity model according to electromechanical system design drawings (such as assembly drawing, part drawing), and key mechanical parts (such as bearing, gear, crankshaft) are highlighted to mark geometric dimensions (accuracy to ±0.01mm), material properties (such as elastic modulus, Poisson's ratio) and assembly constraint relationship; for example, when modeling the main shaft system of a numerical control machine tool, the three-dimensional structure of the inner ring, outer ring and rolling body of the bearing needs to be accurately modeled, and the fitting tolerance (such as interference fit H7 / k6) is defined;

[0121] The 3D model is then imported into simulation software such as ANSYS Twin Builder and COMSOL Multiphysics to perform geometric repairs (such as gap closure and surface fill) and simplify detailed features that have less than 5% impact on the simulation results (such as undercuts and process holes) to reduce computational complexity. Key structures that affect stress distribution and vibration characteristics (such as gear tooth profiles and bearing raceway curvature) are retained, while minor features that do not affect the physical field distribution are removed.

[0122] Finally, the simplified model is parameterized, and geometric dimensions (such as gear module and bearing width) and material properties (such as steel density and magnetic permeability of motor silicon steel sheets) are defined as adjustable parameters to facilitate subsequent model calibration and working condition simulation.

[0123] Step 302: construct a multi-domain physical model based on physical laws, where the multi-domain physical model includes stress distribution of mechanical components, temperature field changes, and electromechanical coupling relationships;

[0124] Specifically, first, a contact stress model of transmission components such as bearings and gears is established based on Hertz contact theory, and the contact pressure distribution between the rolling element and the raceway is calculated through finite element analysis (FEA). The formula is:

[0125] ;

[0126] in, is the load, is the comprehensive elastic modulus, is the contact width, is the integrated curvature radius;

[0127] Then, the lumped mass method or finite element method is used to establish a vibration model of the mechanical system, simulate the free vibration and forced vibration characteristics, and identify the natural frequency and mode shape;

[0128] For heat-generating components such as motors and inverters, a three-dimensional temperature field model is constructed based on Fourier's heat conduction law, taking into account copper loss ( )、Iron loss( ) and other heat-generating factors, as well as convection heat dissipation ( ), radiation heat dissipation path; use ANSYS Thermal module or COMSOL Heat Transfer module for thermal simulation, and encrypt the heat source area (such as motor winding) when dividing the mesh;

[0129] Finally, the coupling relationship between electrical parameters and mechanical motion is described by transfer functions or state-space equations, for example:

[0130] Servo motor torque-current relationship: ;in, is the torque constant;

[0131] Motor speed-voltage relationship: ; wherein, is the back EMF constant.

[0132] Step 303, based on neural network or state space model, combined with historical operation data, behavior model is constructed, dynamic response characteristics of system under different working conditions are simulated, and nonlinear operation law is captured;

[0133] Specifically, first, long short-term memory network (LSTM), gated recurrent unit (GRU) and other recurrent neural networks are used, historical operation data (such as vibration acceleration, temperature, speed, load) are input, and state prediction value (such as vibration displacement, motor temperature rise) at future time is output;

[0134] The specific training process is as follows:

[0135] Divide the data set: 70% for training, 20% for verification, and 10% for testing;

[0136] Feature engineering: time domain and frequency domain features (such as kurtosis, main frequency) and process parameters (such as feed speed, stamping frequency) are combined into input vector;

[0137] Optimization goal: minimize root mean square error (RMSE) or mean absolute error (MAE);

[0138] Then, for the system with known dynamics equation, state space model is used for description:

[0139] , ;

[0140] Among them, is the state vector, is the input vector, is the output vector; and the model parameters are fitted by system identification algorithm (such as least square method) 、 、 、 ; applicable to linear or weak nonlinear system (such as constant speed running gear box), capture the influence of working condition change (such as load step) on system response.

[0141] Step 304, based on expert experience and historical fault data, rule model is constructed, fault diagnosis rule and maintenance decision logic are defined, and knowledge graph is formed;

[0142] Specifically, first, collect expert experience, historical failure cases, industry standards (such as ISO 10816 vibration specifications), extract diagnostic rules, for example: "vibration kurtosis value > 3.5 and temperature gradient > 0.8℃ / min → bearing wear";

[0143] Then, use production rules (IF-THEN structure) or knowledge graph visualization rules, such as "motor temperature anomaly ↑ and current harmonic distortion rate ↑ → winding fault", and associate maintenance measures (such as "replace winding");

[0144] Finally, verify the accuracy of the rules through historical data, eliminate rules with high false positive rates, and adjust the confidence level (such as adjusting the fault confidence of "current THD > 5%" from 0.7 to 0.6).

[0145] Step 305, coupling the parameterized geometric model, multi-domain physical model, behavior model and rule model through the model integration framework to form a unified digital twin model;

[0146] Specifically, first, use OPC UA (Open Platform Communications Unified Architecture) or industrial internet platform (such as Predix, MindSphere) as the model integration framework to define the data interface and interaction protocol of each model;

[0147] Then, couple the parameterized geometric model, multi-domain physical model, behavior model and rule model, the specific method is as follows:

[0148] Geometric-physical coupling: pass the size parameters (such as bearing inner diameter) of the parameterized geometric model to the physical model as boundary conditions (such as contact area, heat dissipation surface area);

[0149] Physical-behavior coupling: stress, temperature and other data output by the physical model are used as input features of the behavior model to drive the neural network to predict system response (such as vibration amplitude growth trend);

[0150] Rule-physical / behavior coupling: input the simulation results of the physical model (such as stress exceeding yield strength) or the prediction value of the behavior model (such as remaining life < 72h) into the rule model to trigger diagnostic rules and maintenance logic.

[0151] Finally, through adaptive Kalman filtering or data assimilation technology, fuse real-time sensor data (such as vibration acceleration, temperature) with model output to update the wear parameters of the geometric model and the material properties (such as bearing clearance increase) of the physical model, ensuring that the state of the digital twin is consistent with the physical entity (error ≤ 5%).

[0152] In this embodiment, specifically: in step three, the model parameters are dynamically calibrated by calculating the root mean square error or cosine similarity between the model output and the actual sensor data, real-time evaluation of model matching degree and trigger parameter update mechanism;

[0153] Specifically, based on the equipment design parameters and historical operation data, set the initial parameters for the digital twin model, such as bearing stiffness , motor stator resistance ;

[0154] The sensor collects vibration, temperature and other data in real time (such as 1000 vibration samples per second), and inputs the digital twin model to generate simulation output (such as predicted vibration displacement time history curve);

[0155] RMSE and cosine similarity are calculated every minute, for example:

[0156] If the measured vibration amplitude is 2.3g and the simulation value is 2.1g, then:

[0157] ;

[0158] If the vector dot product of the measured and simulated temperature curves is 0.98, and the modulus product is 1.0, then the cosine similarity is 0.98;

[0159] When RMSE>0.05g, start the adaptive Kalman filter algorithm, the specific steps are as follows:

[0160] According to the previous time state and input , predict the current state:

[0161] ;

[0162] Calculate the Kalman gain:

[0163] ;

[0164] Combine the measured data Update the state:

[0165] ;

[0166] Where, is the covariance matrix, is the observation matrix;

[0167] Through iterative calculation, gradually adjust the bearing stiffness, motor resistance and other parameters, so that the RMSE is reduced to below the threshold value;

[0168] After calibration, the consistency of the trend of the comparative model output and the measured data (such as the vibration amplitude fluctuation period, the temperature rise rate) is ensured, so that the calibrated model can accurately reflect the current state of the equipment (such as the stiffness reduction caused by bearing wear).

[0169] In this embodiment, specifically: in step four, the multi-dimensional fault feature index system includes the kurtosis value of the vibration signal, the gradient change rate of the temperature data, the harmonic distortion rate of the current signal, and the fluctuation entropy value of the pressure data.

[0170] The kurtosis value of the vibration signal is a fourth-order statistical quantity reflecting the amplitude distribution characteristics of the signal, and is used to measure the amount of impact components in the signal. When the mechanical and electrical system is running normally, the vibration signal is close to Gaussian distribution, and the kurtosis value is about 3. When early faults such as wear and tear and cracks occur in mechanical parts, the impact signal increases, and the kurtosis value will be significantly higher than 3. The abnormal impact characteristics of the parts can be captured in advance, for example, when the rolling body of the bearing is early peeling, the kurtosis value will gradually increase, which is a sensitive index for identifying mechanical impact faults.

[0171] The gradient change rate of the temperature data represents the change speed of the temperature per unit time, and reflects the abnormality of the energy loss or heat dissipation inside the equipment. Under normal conditions, the temperature change of the equipment is relatively slow, and the gradient change rate is usually low (such as <0.5℃ / min). If the equipment has faults such as lubrication failure and winding short circuit, the heat production will increase sharply or the heat dissipation will be blocked, causing the temperature gradient change rate to rise significantly (such as >0.8℃ / min). For example, when the motor bearing is short of oil, the temperature rises rapidly in a short time. Through this index, the abnormal temperature rise trend of the equipment can be found in time, and the potential risk of thermal fault can be warned.

[0172] The harmonic distortion rate (THD) of the current signal is an index for measuring the proportion of harmonic components in the fundamental wave, and is used to reflect the nonlinearity of the electrical system. When the mechanical and electrical system is running normally, the current waveform is close to a sine wave, and the THD is low (such as <5%). If the motor has faults such as rotor broken bar, frequency converter fault or mechanical load imbalance, a large amount of harmonic components will be generated in the current, and the THD will increase significantly (such as >8%). For example, when the rotor bar of an asynchronous motor is broken, the current harmonic distortion rate will increase significantly. This index can effectively monitor the abnormal change of the electrical system and assist in judging the faults of the motor or the drive module.

[0173] The fluctuation entropy of pressure data is calculated based on the complexity and irregularity of the signal and is used to assess the degree of disorder in the pressure signal. When the fluid system is operating normally, the pressure fluctuations have a certain regularity and the entropy value is low (e.g., <1.0). When abnormal conditions such as pipeline leakage, cavitation, or mechanical jamming occur, the pressure fluctuations will become disordered and the entropy value will increase significantly (e.g., >1.5). For example, when pneumatic valve wear causes leakage, the fluctuation entropy of the pressure signal will increase. This indicator can be used to promptly detect abnormal operation of the fluid system and identify potential faults in the pipeline or hydraulic system.

[0174] In this embodiment, specifically: in step 1, multi-source data is collected through vibration sensors, temperature sensors, pressure sensors, and current sensors deployed at key locations of the electromechanical system;

[0175] Specifically, vibration sensors (three-axis acceleration type, sampling frequency 10-20kHz) are installed on mechanical transmission components (such as bearing seats and gearboxes) to capture equipment vibration signals (such as vibration acceleration, velocity, and displacement) in real time for analyzing the operating status and wear of mechanical components.

[0176] Deploy temperature sensors (accuracy ±0.5°C) and current / voltage sensors (resolution 0.1% FS) in electrical drive modules (such as motor stators and inverters) to collect temperature data (such as motor temperature rise and drive module temperature) and current and voltage parameters (such as current waveform and voltage fluctuation) to monitor energy loss and operational stability of the electrical system.

[0177] Install pressure sensors (range 0-5MPa) at fluid system nodes (such as hydraulic pipelines and pneumatic valves) to obtain pressure data (such as hydraulic oil pressure and air pressure fluctuations) to evaluate the sealing and load status of the fluid transmission system.

[0178] Multi-source data includes vibration signals, temperature data, pressure data, and current and voltage data;

[0179] Vibration signals are collected by three-dimensional vibration sensors deployed on mechanical components (such as bearing seats, gearboxes, and crankshafts). These sensors include time-domain waveform data such as acceleration (m / s²), velocity (mm / s), and displacement (μm), as well as frequency-domain features (such as the main frequency and harmonic components) extracted through Fourier transform. This data directly reflects the operating status of mechanical components. For example, bearing wear can lead to increased vibration kurtosis, while abnormal gear meshing can cause vibration peaks at specific frequencies.

[0180] Temperature data is collected by temperature sensors (accuracy ±0.5℃) installed at heat generating parts such as motor stator, frequency converter, hydraulic oil pipeline, including real-time temperature value (℃) and temperature gradient change rate (℃ / min); temperature abnormalities are usually related to energy loss and heat dissipation failure, such as short circuit of motor winding leading to sudden temperature rise, and lubrication failure causing mechanical friction heat;

[0181] Pressure data is collected by pressure sensors (range 0-5MPa) arranged in hydraulic systems, pneumatic circuits or transmission mechanisms (such as guide rail sliders, air cylinders), covering pressure value (Pa), pressure fluctuation amplitude and fluctuation entropy value; pressure data is used to monitor the sealing performance and load balance of fluid systems, such as pipeline leakage leading to continuous pressure drop, and cavitation phenomenon causing severe pressure fluctuation;

[0182] Current and voltage data is collected by current / voltage sensors (resolution 0.1% FS) in electrical control cabinets, including current effective value (A), voltage effective value (V), harmonic distortion rate (THD) and waveform characteristics (such as sine wave distortion); this data reflects the running state of the electrical system, such as motor overload leading to increased current, and frequency converter failure causing increased voltage harmonic content.

[0183] In this embodiment, specifically: in the data cleaning, median filtering or Gaussian filtering algorithm is used to remove abnormal noise;

[0184] Specific implementation steps of median filtering are as follows:

[0185] For example, when collecting the vibration signal of the spindle bearing of a certain numerical control machine tool, pulse noise (such as abnormal points with amplitude increasing by 10g) is generated due to cutting impact;

[0186] A 5-point median filtering window is applied to the vibration acceleration signal (sampling frequency 10kHz), and each data point is traversed, and the median value is taken after sorting the data in the window to replace the current value;

[0187] The pulse noise is reduced from 10g to the normal range (<2g), and the true impact characteristics of the signal (such as normal vibration peak value when the tool cuts in) are preserved;

[0188] Specific implementation steps of Gaussian filtering are as follows:

[0189] For example, the temperature sensor of a certain motor stator is disturbed by environmental electromagnetic interference, and the data shows high-frequency small-amplitude fluctuation (fluctuation range ±3℃).

[0190] Gaussian filtering with standard deviation σ=1.5 is adopted, a 7-point Gaussian kernel (weight distribution [0.06, 0.24, 0.38, 0.24, 0.06]) is generated, and the temperature sequence is convolved and smoothed;

[0191] Noise fluctuation is reduced from ±3℃ to ±0.5℃, and the real temperature rise trend (e.g. temperature rises 0.2℃ per minute when load increases) is clearly presented.

[0192] Normalization is performed by minimum-maximum normalization or Z-score normalization to unify data dimension;

[0193] The minimum-maximum normalization specifically implements the following steps:

[0194] For example, hydraulic system pressure data (range 0-5MPa) and vibration displacement data (range 0-50μm) need to be fused and analyzed;

[0195] Pressure data:

[0196] ;

[0197] Vibration displacement:

[0198] ;

[0199] Two-dimensional data is unified to the [0, 1] interval, which is convenient for neural network input (such as LSTM model);

[0200] The Z-score standardization specifically implements the following steps:

[0201] For example, abnormal fluctuations occur in the motor current data of a production line (mean 5A, standard deviation 0.8A);

[0202] The current value of 6.2A at a certain time is standardized as:

[0203] ;

[0204] The current value of 4.1A at another time is standardized as:

[0205] ;

[0206] Data is transformed into a standard normal distribution with a mean of 0 and a standard deviation of 1, which is suitable for algorithms such as support vector machines (SVM) that require consistent feature variance.

[0207] Feature extraction uses fast Fourier transform, wavelet transform or empirical mode decomposition to extract frequency domain features, energy features and time domain statistical features;

[0208] The fast Fourier transform (FFT) specifically implements the following steps:

[0209] For example, in the analysis of a gearbox vibration signal, the normal meshing frequency is 50Hz, and an abnormal component of 100Hz (2 times frequency) appears when a fault occurs;

[0210] FFT the 1-second vibration data (1000 points) to calculate the amplitude spectrum;

[0211] Identify the 50Hz fundamental amplitude as 0.8g, and the 100Hz component amplitude from 0.1g to 0.5g (exceeding the normal threshold of 0.3g);

[0212] Determine that the gear has a local wear fault (2x frequency usually corresponds to gear eccentricity or wear);

[0213] The specific implementation steps of wavelet transform are as follows:

[0214] For example, the early peeling of a fan bearing contains weak impact (duration < 0.01 seconds) in the vibration signal;

[0215] Use db4 wavelet to decompose the vibration signal for 3 layers, and extract the energy of the detail coefficients at each scale;

[0216] The energy of the 2nd layer detail coefficient (corresponding to the 100-200Hz frequency band) is found to increase by 30% compared to the normal state, and the energy gradually increases over time;

[0217] The early energy feature evolution of bearing peeling is captured, which is 24 hours earlier than traditional FFT;

[0218] The specific implementation steps of time domain statistical features are as follows:

[0219] For example, the vibration signal of a compressor needs to be monitored to evaluate the running stability;

[0220] Calculate the kurtosis value of the vibration signal in 1 minute:

[0221] ;

[0222] The normal threshold is 3;

[0223] Calculate the peak factor:

[0224] ;

[0225] The normal threshold is 2.0;

[0226] Both kurtosis and peak factor exceed the standard, indicating abnormal impact, and combined with frequency spectrum analysis, it is located as a gas valve fault.

[0227] In this embodiment, specifically: in step six, the process of generating maintenance decision suggestions includes:

[0228] Based on the remaining life prediction value in the fault prediction result, combined with the influence weight of equipment downtime on production plan, calculate the number of maintenance tasks;

[0229] If the remaining life prediction value is less than or equal to the preset threshold, an emergency maintenance work order including a maintenance time window, a spare part replacement list, and a temporary capacity allocation scheme is generated;

[0230] If the remaining life prediction value is greater than the preset threshold but the probability of failure continues to rise, a progressive maintenance strategy including state monitoring frequency adjustment and local load reduction operation suggestion is generated;

[0231] Specifically, assuming that the remaining life of a punch press on an automobile production line is predicted to be 36 hours through a digital twin model, the shutdown of the equipment will cause the entire production line to stop production, affecting the production plan of 500 vehicles on the same day, so the production impact weight of the punch press is set to 0.8 (the weight range is 0-1, and the larger the value, the greater the impact); while the remaining life of another detection equipment on the production line is predicted to be 48 hours, its shutdown only affects the quality inspection link, and the production impact weight is 0.3; Through the formula priority coefficient = (1-remaining life / equipment design life) x production impact weight, it is calculated that the priority coefficient of the punch press is 0.667, and the priority coefficient of the detection equipment is 0.25; Therefore, the maintenance task priority of the punch press is higher and needs to be arranged in priority;

[0232] If the remaining life prediction value of the punch press is 36 hours, which is less than or equal to the preset threshold (such as 48 hours), the system immediately generates an emergency maintenance work order; The content of the work order includes:

[0233] Maintenance time window: it is recommended to stop and maintain within the next 8 hours to avoid the peak period of the production line and minimize production loss;

[0234] Spare part replacement list: clearly list the parts that need to be replaced, such as the main bearing (model SKF 6208) of the punch press and the hydraulic seal (specification Φ50xΦ60x10);

[0235] Temporary capacity allocation scheme: coordinate with other production lines in the same factory area to use spare punch presses to undertake part of the production task during maintenance, ensuring that the production target of 300 vehicles on the same day is not affected;

[0236] When a certain injection molding machine has a predicted remaining life of 72 hours (greater than the preset threshold of 48 hours), but its probability of failure continues to rise from 20% last week to 45% currently, the system generates a progressive maintenance strategy, which includes:

[0237] State monitoring frequency adjustment: increase the vibration monitoring frequency of the injection molding machine from once a day to once every 4 hours, and increase the temperature monitoring frequency from once every 2 hours to once every hour, so as to capture the state change of the equipment more timely;

[0238] Local load reduction operation suggestion: Suggest to temporarily reduce the current load of the injection molding machine from 80% to 60%, reduce equipment wear and delay fault development, and prompt the operator to pay close attention to the equipment operation state to gain time for subsequent maintenance.

[0239] The embodiment provides a kind of based on digital twin electromechanical system fault pre-diagnosis method, by deploying multiple sensor acquisition electromechanical system multi-source data, after cleaning, normalization and feature extraction, digital twin model is constructed and dynamically calibrated;Multi-dimensional fault feature system is established based on simulation data, fusion is input model prediction fault probability and residual life, finally generates maintenance decision and early warning, forms "data acquisition-preprocessing-modeling-calibration-feature analysis-prediction-maintenance" closed loop, realizes electromechanical system early fault accurate identification and maintenance strategy intelligent generation, effectively reduce unplanned downtime and maintenance cost, improve production efficiency and equipment reliability.

[0240] Embodiment two

[0241] Figure 3 It is a function module schematic diagram of the electromechanical system fault pre-diagnosis system based on digital twin of the embodiment of the application, as Figure 3 Shown, an electromechanical system fault pre-diagnosis system based on digital twin, including data acquisition module, data preprocessing module, digital twin modeling module, model dynamic calibration module, fault feature analysis module, fault prediction module and maintenance decision module;

[0242] Data acquisition module, configured to obtain multi-source data in the operation process of electromechanical system;

[0243] Data preprocessing module, configured to preprocess the obtained multi-source data, and preprocessing includes data cleaning, normalization processing and feature extraction;

[0244] Digital twin modeling module, configured to construct digital twin model based on preprocessed multi-source data, fuse electromechanical system design drawing, three-dimensional geometric model, material attribute and dynamics equation;

[0245] Model dynamic calibration module, configured to input real-time acquisition data into digital twin model by adaptive Kalman filtering or particle filtering algorithm during the operation process of electromechanical system, dynamically calibrate model parameters, and output simulation data of real-time operation state of electromechanical system;

[0246] Fault feature analysis module, configured to establish multi-dimensional fault feature index system according to simulation data, fuse and analyze fault features using data fusion algorithm, identify potential fault and generate comprehensive fault feature vector;

[0247] The fault prediction module is configured to input the comprehensive fault feature vector into a pre-trained long short-term memory network or support vector machine prediction model, calculate a fault occurrence probability and a remaining life in combination with historical fault data, and output a fault prediction result;

[0248] The maintenance decision module is configured to generate a maintenance decision suggestion based on a preset maintenance strategy rule library according to the fault prediction result, feed back to an operator through a man-machine interaction interface, and synchronously trigger an early warning mechanism.

[0249] In this embodiment, specifically, the model dynamic calibration module includes a data assimilation unit, an error evaluation unit and a threshold triggering unit;

[0250] The data assimilation unit is configured to fuse real-time sensor data and simulation output of the digital twin model by an adaptive Kalman filtering or particle filtering algorithm to generate calibrated model parameters;

[0251] The error evaluation unit is configured to calculate a root mean square error or a cosine similarity between model output and actual sensor data in real time to generate a model matching degree index;

[0252] The threshold triggering unit is configured to preset a parameter update threshold, and if the model matching degree index exceeds the threshold, the data assimilation unit is automatically triggered to dynamically calibrate the model parameters.

[0253] In this embodiment, specifically, the fault feature analysis module includes a feature index construction unit, a cross-modal fusion unit and an anomaly recognition unit;

[0254] The feature index construction unit is configured to establish a multi-dimensional fault feature index system including a vibration signal kurtosis value, a temperature data gradient change rate, a current signal harmonic distortion rate and a pressure data fluctuation entropy value according to simulation data;

[0255] The cross-modal fusion unit is configured to use an improved D-S evidence theory, a neural network algorithm or a fuzzy logic algorithm to fuse and process simulation features from the digital twin model and measured sensor features to generate a comprehensive fault feature vector including time domain, frequency domain and energy features;

[0256] The anomaly recognition unit is configured to identify potential fault features and mark weight coefficients of abnormal dimensions by using a preset fault feature threshold or a statistical process control method.

[0257] In summary, the embodiment provides a kind of electromechanical system fault pre-diagnosis system based on digital twinning, obtains multi-source data by data acquisition module, after cleaning, normalization and feature extraction by data preprocessing module, by digital twinning modeling module, the digital twin of fusion multidimensional model is constructed;Model dynamic calibration module utilizes adaptive Kalman filtering algorithm and the like to calibrate model parameter in real time, fault feature analysis module is based on simulation data and constructs multidimensional feature system and fusion to generate comprehensive vector, fault prediction module exports fault probability and remaining life by LSTM / SVM model, finally maintenance decision module generates maintenance suggestion according to rule and triggers early warning, form the whole process intelligent pre-diagnosis closed loop from data acquisition to maintenance decision, improve electromechanical system fault identification precision and maintenance efficiency.

[0258] Other details of the implementation of the technical solutions of each module in the above-described embodiment of the electromechanical system fault pre-diagnosis system based on digital twinning can be referred to the description in the embodiment of the method for electromechanical system fault pre-diagnosis based on digital twinning.

[0259] Embodiment three

[0260] The embodiment provides an actual application of the method for electromechanical system fault pre-diagnosis based on digital twinning in a complex electromechanical system of a large stamping production line in a certain automobile manufacturing factory, as follows:

[0261] In the stamping production line, the core electromechanical equipment includes multiple high-speed stamping machines, automatic feeding devices and mechanical arms, etc.; in order to obtain multi-source data, various sensors are densely deployed at key positions: vibration sensors are installed at the crankshaft, connecting rod and other positions of the stamping machine, which are used to collect three-direction vibration signals with a frequency range of 0-10 kHz; temperature sensors are installed on the motor housing and driving module, with an accuracy of ±0.3℃; pressure sensors are installed at the transmission chain and guide rail slider of the feeding device, with a range adapted to the operating pressure of the equipment; current and voltage sensors are arranged in the electrical control cabinet to monitor the changes of electrical parameters; these sensors transmit data to the edge computing node in real time through industrial Ethernet, with a transmission delay controlled within 30 ms.

[0262] In the data preprocessing stage, the median filter algorithm (window size of 5) is used to denoise the vibration signal, and remove abnormal noise caused by equipment impact, electromagnetic interference, etc.; Z-score normalization method is used to unify the dimension of temperature, current and other data, so that the mean value is 0 and the standard deviation is 1; Fast Fourier Transform (FFT) is used to extract frequency domain features from vibration signals, wavelet transform is used to obtain energy features of signals, and Empirical Mode Decomposition (EMD) is used to extract time domain statistical features such as mean value, variance, etc.

[0263] When building the digital twin model, first, a three-dimensional fine model of each device in the stamping production line is established using CAD software, and key component sizes, assembly relationships, and other information are labeled in detail. Then, the model is imported into professional simulation software for geometry repair and simplification to generate a parameterized geometric model. Based on physical laws such as material mechanics and dynamics, a stress distribution model of mechanical components, a thermal model of motors and transmission systems, and an electromechanical coupling model are constructed to accurately describe the physical processes during device operation. With the help of neural networks (such as LSTM networks) and state-space models, combined with years of historical operation data of the production line, an action model is constructed to simulate the dynamic response characteristics of the device under different stamping frequencies and load conditions, effectively capturing the nonlinear laws in device operation. At the same time, according to industry expert experience and past fault cases, a rule model is constructed to define diagnostic rules such as "vibration amplitude exceeds a certain threshold and duration exceeds 5 seconds → mechanical component loosening fault exists" and form a fault knowledge graph. Finally, through a mature model integration framework, the parameterized geometric model, multi-domain physical model, action model, and rule model are deeply coupled to form a unified digital twin model that accurately reflects the actual operating conditions of the stamping production line.

[0264] During the operation of the production line, the real-time collected sensor data is fused with the simulation output of the digital twin model through an adaptive Kalman filter algorithm.

[0265] Specifically, the algorithm dynamically adjusts the weight according to the uncertainty of the data, continuously updates the model parameters, and ensures that the model closely follows the changes in the actual operating state of the device. By calculating the root mean square error (RMSE) between the model output and the actual sensor data in real time, an RMSE threshold of 0.03g (g is the acceleration of gravity) is set. When the error exceeds the threshold, the model parameter update process is automatically triggered. At the same time, the cosine similarity between the model output and the measured data is calculated. When the similarity is less than 0.9, the calibration mechanism is also started to ensure the high matching degree and accuracy of the model.

[0266] According to the simulation data output by the digital twin model, a multi-dimensional fault feature index system is established. In addition to covering the kurtosis value of vibration signals, the gradient change rate of temperature data, the harmonic distortion rate of current signals, and the fluctuation entropy value of pressure data, the system also combines stamping process parameters such as stamping speed change rate and mold closing time deviation. An improved D-S evidence theory is used to fuse these fault features, reasonably assigning weights to each feature, such as setting the vibration kurtosis value weight to 0.35 and the temperature gradient change rate weight to 0.25, and calculating the trustworthiness of each feature to identify potential faults and generate a comprehensive fault feature vector containing time-domain, frequency-domain, energy, and process-related features.

[0267] The comprehensive fault feature vector is input into a long short-term memory (LSTM) prediction model pre-trained, the model combines historical fault data, and uses a complex time series analysis algorithm to calculate the fault occurrence probability and the remaining life; when training the LSTM model, a large number of historical data samples covering normal operation, early fault, severe fault and other different states are used, and the model parameters are optimized through multiple iterations, so that the model has strong fault prediction capability; the fault prediction result output by the model, such as the remaining life prediction, is accurate to hours, and the fault occurrence probability is presented in percentage form.

[0268] Based on the fault prediction result, a maintenance decision suggestion is generated according to a preset maintenance strategy rule base; for example, when the remaining life prediction value of a certain punch press is less than 48 hours and the fault occurrence probability is more than 80%, an emergency maintenance work order is immediately generated, the work order lists the spare parts list (such as severely worn molds, aged motor brushes, etc.) that need to be replaced in detail, plans a maintenance time window (such as downtime maintenance within the next 8-12 hours), and synchronously triggers a production scheduling system to start a temporary capacity allocation scheme to transfer part of the production tasks to other standby punch presses to reduce the impact on overall production; if the device remaining life prediction value is greater than 48 hours but the fault occurrence probability shows a gradually increasing trend, a gradual maintenance strategy is generated, suggesting to increase the state monitoring frequency of the device, such as increasing the vibration monitoring from once a day to once every 4 hours, and appropriately reducing the punch speed according to the device operating condition to reduce the device load and delay the development of the fault; the maintenance decision suggestion is intuitively fed back to the operator through a specially developed human-computer interaction interface, the interface displays device health status, fault type, maintenance measures and other information in the form of charts and text combination, and synchronously triggers an early warning mechanism through sound and light alarms to ensure that the operator can respond in time and take appropriate measures.

[0269] After a period of actual operation verification, the punch production line reduces the unplanned downtime by 68% and reduces the equipment maintenance cost by 35% after applying the digital twin-based mechanical and electrical system fault pre-diagnosis method proposed in the application, the production efficiency is significantly improved, which fully proves the effectiveness and practicality of the application in the field of complex mechanical and electrical system fault pre-diagnosis.

[0270] Embodiment Four

[0271] Based on the embodiment one, the application provides a digital twin-based mechanical and electrical system fault pre-diagnosis method, comprising the following steps:

[0272] Based on the preprocessed multi-source data, the mechanical and electrical system design drawings, three-dimensional geometric models, material properties and dynamic equations are fused to construct a digital twin model;

[0273] Based on the Bayesian rule update framework, the dynamic adjustment of the rule confidence is realized:

[0274] The initial rule confidence is set by expert experience, and after each fault occurs, the confidence is updated by Bayesian posterior probability according to the matching degree of the actual fault type and the rule triggering result;

[0275] During the operation of the mechanical and electrical system, real-time collected data is input into the digital twin model through adaptive Kalman filtering or particle filtering algorithm, the model parameters are dynamically calibrated, and simulation data of the real-time operation state of the mechanical and electrical system is output;

[0276] An exponential relationship model of the vibration kurtosis value and the bearing stiffness degradation coefficient is established, and when the kurtosis value exceeds the threshold value, the stiffness parameter is automatically updated, replacing the traditional fixed threshold calibration;

[0277] According to the simulation data, a multi-dimensional fault feature index system is established, and a data fusion algorithm is used to fuse and analyze the fault features, identify potential faults and generate a comprehensive fault feature vector;

[0278] The real-time monitoring of the process parameter change rate is realized through the separation of the process disturbance component and the fault component by the variational mode decomposition;

[0279] The dynamic calibration of the exponential relationship model of the vibration kurtosis value and the bearing stiffness degradation coefficient adopts a wear-parameter mapping function ΔK=α·e^(β·Kurtosis)+γ, wherein α, β and γ are calibrated through the accelerated life test.

[0280] It should be noted that each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts of each embodiment can be referred to. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0281] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting faults in electromechanical systems based on digital twins, characterized in that: The following steps are involved: Acquire multi-source data during the operation of the electromechanical system; pre-process the acquired multi-source data; Build a digital twin model based on pre-processed multi-source data, integrating electromechanical system design drawings, 3D geometric models, material properties, and dynamic equations; During the operation of the electromechanical system, real-time collected data is input into the digital twin model through an adaptive Kalman filter or particle filter algorithm, model parameters are dynamically calibrated, and simulation data of the real-time operating status of the electromechanical system is output. The method for constructing the digital twin model includes the following steps: Use modeling tools to build a 3D model of the electromechanical system, and import it into simulation software for geometric repair and simplification, generate a parametric geometric model, and map the structural features of mechanical components; Building a multi-domain physical model based on physical laws, wherein the multi-domain physical model includes stress distribution of mechanical components, temperature field changes, and electromechanical coupling relationships; Based on neural network or state space model, combined with historical operation data, behavioral model is constructed to simulate the dynamic response characteristics of the system under different working conditions and capture nonlinear operation laws; Build a rule model based on expert experience and historical fault data, define fault diagnosis rules and maintenance decision logic, and form a knowledge graph; The parametric geometric model, multi-domain physical model, behavioral model, and rule model are coupled through the model integration framework to form a unified digital twin model. A multi-dimensional fault feature index system is established based on simulation data. The data fusion algorithm is used to fuse and analyze the fault features, identify potential faults, and generate a comprehensive fault feature vector. The comprehensive fault feature vector is input into the pre-trained long short-term memory network or support vector machine prediction model, and the fault probability and remaining life are calculated in combination with historical fault data, and the fault prediction result is output.

2. A method for predicting faults in electromechanical systems based on digital twins according to claim 1, characterized in that: The model parameters are dynamically calibrated, and the matching degree of the digital twin model is evaluated in real time by calculating the root mean square error or cosine similarity between the model output and the actual sensor data, thereby triggering the parameter update mechanism.

3. The electromechanical system fault pre-diagnosis method based on digital twin according to claim 1 is characterized in that: The multi-dimensional fault characteristic indicator system includes the kurtosis value of the vibration signal, the gradient change rate of the temperature data, the harmonic distortion rate of the current signal and the fluctuation entropy value of the pressure data.

4. A method for predicting faults in electromechanical systems based on digital twins according to claim 1, characterized in that: The multi-source data is collected through vibration sensors, temperature sensors, pressure sensors and current sensors deployed at key parts of the electromechanical system; the multi-source data includes vibration signals, temperature data, pressure data and current and voltage data.

5. The electromechanical system fault pre-diagnosis method based on digital twin according to claim 1 is characterized in that: The preprocessing includes data cleaning, normalization and feature extraction. The data cleaning uses median filtering or Gaussian filtering algorithm to remove abnormal noise; the normalization process unifies the data dimension through minimum-maximum normalization or Z-score normalization; the feature extraction uses fast Fourier transform, wavelet transform or empirical mode decomposition to extract frequency domain features, energy features and time domain statistical features.

6. A method for predicting faults in electromechanical systems based on digital twins according to claim 1, characterized in that: The method further includes generating maintenance decision suggestions based on the fault prediction results and a preset maintenance strategy rule base, feeding back the suggestions to the operator through a human-computer interaction interface, and simultaneously triggering an early warning mechanism. The process of generating maintenance decision suggestions includes: Based on the remaining life prediction value in the fault prediction results and the weight of the impact of equipment downtime on the production plan, the priority coefficient of the maintenance task is calculated; If the remaining life prediction value is less than or equal to the preset threshold, an emergency maintenance work order is generated, including a repair time window, a spare parts replacement list, and a temporary capacity allocation plan. If the remaining life prediction value is greater than the preset threshold but the probability of failure continues to increase, a progressive maintenance strategy is generated, including adjustment of condition monitoring frequency and local load reduction operation suggestions.

7. A mechatronic system fault pre-diagnosis system based on digital twins, applied to the mechatronic system fault pre-diagnosis method based on digital twins according to claim 1, characterized in that: It includes data acquisition module, data preprocessing module, digital twin modeling module, model dynamic calibration module, fault feature analysis module, fault prediction module and maintenance decision module; The data acquisition module is configured to acquire multi-source data during the operation of the electromechanical system; The data preprocessing module is configured to preprocess the acquired multi-source data, wherein the preprocessing includes data cleaning, normalization and feature extraction; The digital twin modeling module is configured to construct a digital twin model based on preprocessed multi-source data, integrating electromechanical system design drawings, three-dimensional geometric models, material properties and dynamic equations; The model dynamic calibration module is configured to input real-time collected data into the digital twin model through an adaptive Kalman filter or a particle filter algorithm during the operation of the electromechanical system, dynamically calibrate the model parameters, and output simulation data of the real-time operating status of the electromechanical system; The fault feature analysis module is configured to establish a multi-dimensional fault feature indicator system based on simulation data, use a data fusion algorithm to fuse and analyze fault features, identify potential faults and generate a comprehensive fault feature vector; The fault prediction module is configured to input the comprehensive fault feature vector into a pre-trained long short-term memory network or support vector machine prediction model, calculate the fault probability and remaining life in combination with historical fault data, and output the fault prediction result; The maintenance decision module is configured to generate maintenance decision suggestions based on the fault prediction results and a preset maintenance strategy rule library, and feed them back to the operator through the human-computer interaction interface, and simultaneously trigger the early warning mechanism.

8. The electromechanical system fault prediction system based on digital twin according to claim 7 is characterized in that: The model dynamic calibration module includes a data assimilation unit, an error evaluation unit and a threshold triggering unit; The data assimilation unit is configured to fuse the real-time sensor data with the simulation output of the digital twin model through an adaptive Kalman filter or a particle filter algorithm to generate calibrated model parameters; The error evaluation unit is configured to calculate the root mean square error or cosine similarity between the model output and the actual sensor data in real time to generate a model matching index; The threshold trigger unit is configured to preset a parameter update threshold. If the model matching index exceeds the threshold, the data assimilation unit is automatically triggered to dynamically calibrate the model parameters.

9. The electromechanical system fault prediction system based on digital twin according to claim 7, characterized in that: The fault feature analysis module includes a feature index construction unit, a cross-modal fusion unit and an anomaly identification unit; The characteristic index construction unit is configured to establish a multi-dimensional fault characteristic index system including the vibration signal kurtosis value, the temperature data gradient change rate, the current signal harmonic distortion rate, and the pressure data fluctuation entropy value based on the simulation data; The cross-modal fusion unit is configured to use an improved DS evidence theory, a neural network algorithm, or a fuzzy logic algorithm to fuse the simulation features from the digital twin model with the measured sensor features to generate a comprehensive fault feature vector including time domain, frequency domain, and energy features; The abnormality identification unit is configured to identify potential fault features through a preset fault feature threshold or a statistical process control method, and mark the weight coefficient of the abnormal dimension.

Images