Magnetic suspension fan emergency speed change working condition bearing protection method and system based on digital twinning

Abstract

The invention provides a magnetic suspension fan emergency variable speed working condition bearing protection method and system based on digital twinning. The magnetic suspension fan emergency variable speed working condition bearing protection method comprises the steps that time sequence data of the fan running state of each magnetic suspension fan in an air duct are obtained; stress distribution of reverse impact on a downstream bearing is analyzed through the spatial change trend of backpressure sudden change, and the dynamic influence range of the reverse impact on the bearing is determined in combination with fan operation parameters; disturbance data and magnetic suspension bearing data of adjacent fans in the dynamic influence range are obtained, the frequency and amplitude of oscillation disturbance are calculated, disturbance frequency components of control signals are filtered out, and a fan control instruction set is obtained; adjusting the pressure gradient of the airflow turbulence area in the air duct in real time through a fan control instruction set in combination with the calculation result of the back pressure sudden change distribution in the air duct to obtain optimized airflow distribution; and if the smoke discharge sudden drop or smoke backflow phenomenon is detected, cooperative control is carried out on the operation time sequence of the multiple fans, and a cooperative operation time sequence scheme is generated.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a bearing protection method and system for a magnetic levitation fan under emergency speed change conditions based on digital twins. Background Art

[0002] Research on digital twin technologies plays a crucial role in modern industry, particularly in building ventilation and fire smoke exhaust systems. Its virtual mapping and real-time simulation significantly improve the operational efficiency and safety of complex systems. Coordinated smoke exhaust from multiple fans in vertical ducts is a critical component in ensuring fire safety in high-rise buildings, directly impacting the safety of life and property and emergency response capabilities. However, existing technologies for handling multiple fans in series and parallel operation are often limited by traditional control methods and static models, making them unable to adapt to the complex coupling effects in dynamic environments, resulting in reduced overall system performance or even failure. In particular, under non-ideal conditions such as signal jitter or network congestion, the chain reaction caused by asynchronous fan startup presents a technical bottleneck that urgently needs to be overcome. Current solutions often rely on independent control of individual fans or preset timing logic. These solutions often exhibit problems with inadequate prediction and delayed response to sudden backpressure changes, sensor disturbances, or airflow disturbances. For example, a downstream fan failure or sudden speed change can cause a sudden change in backpressure. This can cause reverse shock to downstream bearings or oscillation in neighboring fans due to control conflicts, both of which expose the system's lack of adaptability to real-time coupling effects. It can be seen that the limitation of existing methods lies in the lack of dynamic perception and collaborative control capabilities of distributed risks. The core challenges in the research field focus on distributed risk prediction and control optimization under asynchronous startup, which are specifically manifested in three technical factors: first, the transient impact of sudden changes in duct backpressure on the mechanical components of the fan; second, the oscillation of the control system caused by disturbances in adjacent fans; and third, the sudden drop in smoke exhaust efficiency in the refuge layer due to airflow turbulence. These factors have not been effectively addressed, resulting in problems such as smoke backflow or smoke exhaust failure in extreme scenarios. How to build a cross-fan coupling prediction model based on real-time solution of the duct flow field under the digital twin framework and realize dynamic fine-tuning of nanosecond collaborative control timing has become a key issue in improving the robustness and safety of the system. Summary of the Invention

[0003] The present invention provides a bearing protection method for a magnetic levitation fan under emergency speed change conditions based on digital twins, which mainly includes:

[0004] Obtain the time series data of the operating status of each magnetic suspension fan in the air duct;

[0005] Based on the time series data, it is determined whether there is an asynchronous startup phenomenon. If the deviation between the startup time of a certain fan and that of other fans exceeds the preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct.

[0006] By analyzing the spatial variation trend of back pressure mutations, the stress distribution of the downstream bearing under reverse impact is analyzed, and the dynamic impact range of the reverse impact on the bearing is determined in combination with the wind turbine operating parameters;

[0007] Obtain disturbance data from adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculate the frequency and amplitude of the oscillation disturbance, filter out the disturbance frequency component of the control signal, and obtain the wind turbine control instruction set;

[0008] Through the fan control instruction set and the calculation results of the sudden change distribution of back pressure in the air duct, the pressure gradient in the turbulent air flow area in the air duct is adjusted in real time to obtain the optimized air flow distribution;

[0009] If a sudden drop in exhaust gas or smoke backflow is detected, the operation timing of multiple fans will be coordinated and a coordinated operation timing plan will be generated;

[0010] According to the collaborative operation timing plan, the digital twin system conducts real-time simulation of the air duct flow field, solves the pressure and flow changes under the cross-fan coupling effect, and outputs the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans;

[0011] Based on the results of the real-time simulation of the duct flow field by the digital twin system, the parameters of the cross-turbine coupling prediction are updated, and operation optimization instructions are generated based on the real-time changes in back pressure mutations and airflow turbulence as well as the health status of the bearings.

[0012] Furthermore, obtaining time series data on the operating status of each magnetic levitation fan in the air duct includes: acquiring first speed data, first pressure data, and first flow data of the magnetic levitation fan in the air duct from a sensor according to a preset sampling period; performing linear calibration on the first speed data, first pressure data, and first flow data using built-in sensor sensitivity parameters to obtain second speed data, second pressure data, and second flow data; and generating corresponding timestamps based on the data acquisition frequency. To address noise interference in the second speed data, second pressure data, and second flow data, smoothing the signals using a five-point sliding average filter method to obtain third speed data, third pressure data, and third flow data, and removing abnormal data points based on a preset noise threshold. Based on the timestamps in the third speed data, third pressure data, and third flow data, calculating the time interval between adjacent data points. If the time interval exceeds a preset threshold, filling the missing data points using a cubic spline interpolation algorithm to obtain fourth speed data, fourth pressure data, and fourth flow data. Segmenting the fourth speed data, fourth pressure data, and fourth flow data into time windows, and extracting the signal eigenvalues ​​within each time window using a wavelet transform to obtain a first characteristic sequence. A wind turbine operating state vector is constructed based on the first feature sequence, and a time series prediction model is established using a long short-term memory network algorithm. The wind turbine operating state is predicted using the time series prediction model to obtain a second feature sequence. A Gaussian distribution is fitted to the second feature sequence to obtain wind turbine operating state time series data.

[0013] Furthermore, based on the time series data, it is determined whether there is an asynchronous start-up phenomenon. If it is detected that the start-up time of a certain fan deviates from that of other fans by more than a preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the duct, including: obtaining the time series of the start-up moments of multiple fans in the duct from the time series data, calculating the start-up time difference between adjacent fans, and normalizing the start-up time difference using the maximum and minimum normalization method with an interval of 0 to 1 to obtain a first start-up delay sequence. According to the first start-up delay sequence, the correlation between the start-up delays of adjacent fans is calculated using a cross-correlation function, the first start-up delay sequence is segmented using a sliding window of 10 seconds in length, and the comparison result between the delay value and the preset start-up synchronization threshold is calculated in each time window to obtain a second start-up delay sequence. The duct geometric dimension parameters, including duct length, cross-sectional area, and elbow position, are obtained from the duct controller, and the duct space is discretized using a hexahedral mesh division method to obtain duct mesh data. If the delay value in the second startup delay sequence exceeds the preset startup synchronization threshold, a pressure-corrected finite volume algorithm is used to solve the pressure field distribution in the air duct for the air duct grid data, and the first backpressure distribution data is obtained through numerical pressure field calculation. Based on the first backpressure distribution data, the central difference method is used to calculate the pressure gradient between adjacent grid nodes to obtain the second backpressure distribution data. The backpressure mutation area is determined by the pressure gradient threshold. Based on the second backpressure distribution data within the backpressure mutation area, the cubic spline interpolation method is used to calculate the backpressure spatial variation curve to obtain the spatial variation trend data of the backpressure mutation.

[0014] Furthermore, the stress distribution of the downstream bearing under reverse impact is analyzed by analyzing the spatial variation trend of the backpressure mutation, and the dynamic impact range of the reverse impact on the bearing is determined in combination with the wind turbine operating parameters. This method includes: based on the spatial variation trend data of the backpressure mutation, meshing the bearing inner ring, rolling elements, and outer ring structure using tetrahedral elements, setting steel elastic modulus and Poisson's ratio parameters, applying bearing fixed surface constraint boundary conditions, and generating first bearing finite element mesh data. For this first mesh data, the mesh distortion rate is calculated using a mesh quality detection method. Elements with distortion rates exceeding a preset threshold are locally refined to generate second bearing finite element mesh data. A prestressed load is applied to the bearing to obtain first stress distribution data. Based on this first stress distribution data, the transient response of the bearing under the backpressure mutation load is calculated using explicit dynamic equations. First dynamic response data is generated based on the calculated nodal accelerations and displacements. Based on this first dynamic response data, the stress transfer patterns between the bearing components are calculated using a stress transfer path tracing algorithm to generate stress transfer path distribution data. Current speed and power data are obtained from the fan controller. A support vector regression algorithm is used to establish a mapping relationship between speed and power and bearing stress. The optimal mapping parameters are selected through a cross-validation method to obtain second dynamic response data. Based on the second dynamic response data and the stress transfer path distribution data, the comprehensive stress state of the bearing is calculated using a stress superposition criterion. The dynamic influence region of the bearing is determined using the von Mises stress criterion to obtain characteristic data of the dynamic influence range of the bearing.

[0015] Furthermore, disturbance data from adjacent wind turbines within the dynamic influence range and magnetic bearing data are acquired, the frequency and amplitude of the oscillation disturbance are calculated, and the disturbance frequency component of the control signal is filtered out to obtain a wind turbine control instruction set. This includes: Based on the dynamic influence range, vibration sensors, pressure sensors, acoustic sensors, and magnetic levitation sensors are placed adjacent to the wind turbines. A synchronous sampling mechanism with a sampling frequency of 10 kHz is used to acquire a first vibration signal, a first pressure signal, a first acoustic signal, and a first magnetic levitation signal to obtain a first sensor data sequence. For the first sensor data sequence, signal synchronization is performed using a timestamp alignment method. The sampling frequency is unified using a resampling algorithm based on Shannon's sampling theorem to obtain a second sensor data sequence. Based on the second sensor data sequence, a wavelet decomposition algorithm is used to extract the frequency and amplitude characteristics of the vibration, pressure, and acoustic signals. Anomalous fluctuations are eliminated using a standard deviation-based anomaly detection method to obtain a first disturbance signature sequence. For the first disturbance signature sequence, a fast Fourier transform is used to calculate the spectral distribution, and a peak detection algorithm is used to extract the primary frequency components. The second disturbance signature sequence is then constructed by combining the magnetic bearing temperature and current data. Based on the second disturbance signature sequence, the main disturbance frequency component is determined using an amplitude sorting method. The disturbance type is determined by comparing it with a preset frequency range to generate a first frequency eigenvector. Based on the first frequency eigenvector, an adaptive filter is constructed using a recursive least squares algorithm. The convergence coefficient and forgetting factor are set to filter out the disturbance frequency component in the fan control signal, thereby obtaining a first control instruction sequence. Based on the first control instruction sequence, speed instructions, current instructions, and displacement instructions are generated through proportional transformation to construct a fan control instruction set. Furthermore, the fan control instruction set is combined with the calculated results of the sudden backpressure distribution within the duct to adjust the pressure gradient in the turbulent airflow region within the duct in real time to obtain an optimized airflow distribution. This method includes: based on the fan control instruction set and the sudden backpressure distribution data, the duct space is divided using a hexahedral mesh to generate first computational mesh data. The first velocity field and first pressure field at the nodes of the first computational mesh are calculated using the Reynolds average equation to obtain first flow field distribution data. The mesh quality of the first computational mesh is assessed using a mesh distortion rate criterion. Cells with distortion rates exceeding a preset mesh quality threshold are encrypted to generate second computational mesh data. Based on the first flow field distribution data, the vorticity conservation equation is used on the second computational grid to calculate the airflow rotation intensity. Turbulent airflow regions are identified using a turbulence threshold criterion to obtain a first airflow turbulence index. Based on the first airflow turbulence index, a recursive neural network is used to construct a flow field optimization calculation unit. The number of hidden layer nodes and activation function parameters are set, and the pressure gradient within the turbulent region is optimized to obtain the second flow field distribution data. Based on the second flow field distribution data, a pressure correction method is used to calculate the second velocity field. The degree of computational convergence is determined by the velocity divergence value to obtain the third flow field distribution data.For the third flow field distribution data, the finite volume method is used to construct the flow field control equation, and the airflow field parameters are solved by the mass conservation and momentum conservation equations to obtain the optimized airflow distribution data.

[0016] Furthermore, if a sudden drop in smoke exhaust or backflow is detected, the operation timing of multiple fans is coordinated and controlled to generate a coordinated operation timing plan. This includes: using the differential method to calculate the flue gas flow rate change rate and pressure change rate based on the optimized airflow distribution data, marking areas where the flue gas flow rate falls below 50% of the rated flow rate or where the pressure enters a negative value, and generating first abnormal state data. Based on the first abnormal state data, a flue gas flow parameter curve is calculated using the least squares method. The degree of flue gas flow abnormality is determined by the curve fit, and first flow characteristic data is obtained. Based on the first flow characteristic data, a smoke exhaust anomaly identifier is constructed using a convolutional neural network with a three-layer convolutional structure and two fully connected layers. This classifies abnormal flue gas emission conditions and generates a first fault feature vector. Based on the first fault feature vector, a particle swarm optimization algorithm is used to calculate the multi-fan operation timing compensation value, with a compensation value adjustment step size of 10 nanoseconds, to generate the first operation timing data. Based on the first operation timing data, the parameter constraint criteria are used to verify whether the fan speed adjustment, current adjustment, and displacement adjustment meet the preset range requirements, and generate second operation timing data. For the second operation timing data, a coordinated control parameter matrix is ​​constructed through the wind turbine group controller, and a linear interpolation method is used to calculate the transition sequence of each control parameter to generate a coordinated operation timing solution.

[0017] Furthermore, according to the collaborative operation timing scheme, a digital twin system is used to perform real-time simulation of the duct flow field, solving the pressure and flow changes under the cross-turbine coupling effect, and outputting predicted values ​​of the pressure and flow changes under the coupling effect of multiple magnetic levitation fans. This includes: constructing a digital twin mapping relationship of the duct geometry according to the collaborative operation timing scheme, and generating first virtual scene data including fan positions, pipe dimensions, and boundary conditions. For this first virtual scene data, the Lagrangian particle method is used to divide the flow field calculation region, the grid density is increased in the cross-turbine coupling region, and the pressure and velocity fields are calculated using the Reynolds average equation to obtain first flow field state data. Based on this first flow field state data, a cross-turbine coupling predictor is constructed using a long short-term memory network. The input layer is configured to receive pressure transfer characteristics and flow mutual feedback characteristics, and the hidden layer includes pressure coupling units and flow coupling units to generate a first coupling eigenvector. For this first coupling eigenvector, a recursive training method is used to optimize the cross-turbine coupling predictor parameters. The prediction accuracy is verified using historical data to obtain a second coupling eigenvector. Based on the second coupling eigenvector, the finite volume method is used to solve the governing flow equations within the duct. The flow field parameters in the cross-fan region are iteratively calculated through pressure correction to obtain the second flow field state data. Based on the second flow field state data, an adaptive mesh refinement method is used to perform localized refinement calculations in high-gradient regions. The final mesh size is determined using a mesh convergence criterion to obtain the third flow field state data. Based on the third flow field state data, the least squares method is used to fit the pressure and flow rate variation curves. The cross-fan coupling predictor outputs the predicted pressure and flow rate variations under the coupling effect of multiple magnetic levitation fans.

[0018] Furthermore, based on the results of the digital twin system's real-time simulation of the duct flow field, the parameters of the cross-turbine coupling prediction are updated. Operational optimization instructions are generated based on real-time changes in backpressure and airflow turbulence, as well as the health of the bearings. These instructions include: calculating the pressure transfer coefficient and flow mutual feedback coefficient using the recursive least squares method based on the digital twin simulation results; online updating the coupling coefficient and time lag parameter in the cross-turbine coupling prediction parameters; and calculating parameter corrections using the error feedback method to obtain first parameter optimization data. For the first parameter optimization data, the reliability of the parameter update results is verified using a cross-validation method. Parameter convergence is determined through residual analysis to generate second parameter optimization data. Based on the second parameter optimization data, a coupling parameter optimizer is constructed using a long short-term memory network. Time series predictions are performed on pressure field fluctuations and flow field disturbances. Airflow turbulence characteristics are calculated based on a preset sampling period to obtain a first state prediction sequence. For the first state prediction sequence, vibration signals, temperature signals, and current signals are collected by bearing monitoring sensors, and time-frequency analysis is performed using wavelet transform to generate a first bearing eigenvector. Based on the first bearing eigenvector, the stress superposition criterion is used to calculate bearing stress fluctuations. The bearing operating status is determined using the life prediction curve to obtain first health assessment data. Based on this first health assessment data, a weighted fusion method is used to integrate the airflow turbulence characteristics and the bearing health status. A dynamic programming algorithm is used to calculate the optimal combination of fan speed, current, and displacement commands to generate optimized operating instructions.

[0019] The present invention provides a bearing protection system for emergency speed change of a magnetic levitation fan based on digital twin, which mainly includes:

[0020] A data acquisition module is used to obtain time series data of the operating status of each magnetic suspension fan in the air duct;

[0021] The asynchronous start detection module is used to determine whether there is an asynchronous start phenomenon based on time series data. If the deviation between the start time of a certain fan and that of other fans exceeds a preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct;

[0022] The back pressure mutation analysis module is used to analyze the stress distribution of the downstream bearing under reverse impact based on the spatial variation trend of the back pressure mutation, and determine the dynamic impact range of the reverse impact on the bearing in combination with the wind turbine operating parameters;

[0023] The bearing stress analysis module is used to obtain disturbance data of adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculate the frequency and amplitude of the oscillation disturbance, filter out the disturbance frequency component of the control signal, and obtain the wind turbine control instruction set;

[0024] The disturbance data processing module is used to adjust the pressure gradient in the turbulent airflow area in the air duct in real time through the fan control instruction set and the calculation results of the sudden change distribution of back pressure in the air duct to obtain the optimized airflow distribution;

[0025] The airflow optimization module is used to coordinate the operation timing of multiple fans and generate a coordinated operation timing plan if a sudden drop in smoke exhaust or smoke backflow is detected;

[0026] The collaborative control module is used to simulate the duct flow field in real time through the digital twin system according to the collaborative operation timing plan, solve the pressure and flow changes under the cross-fan coupling effect, and output the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans;

[0027] The simulation and optimization module is used to update the parameters of cross-turbine coupling prediction based on the results of real-time simulation of the duct flow field by the digital twin system, and generate operation optimization instructions based on the real-time changes of back pressure mutations and airflow turbulence as well as the health status of bearings.

[0028] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0029] The present invention discloses a bearing protection method and system for emergency speed change conditions of magnetic levitation fans based on digital twins. By collecting fan operation data in real time, the back pressure mutation caused by asynchronous startup is analyzed, and the influence range of the bearing subjected to reverse impact is calculated. On this basis, sensors are deployed to obtain disturbance data, and an adaptive filtering algorithm is used to generate fan control instructions to adjust the airflow distribution. In response to the phenomenon of sudden drop in exhaust gas or backflow of smoke, the present invention adjusts the operating timing of multiple fans at the nanosecond level, and uses digital twin technology to simulate the air duct flow field in real time to construct a cross-fan coupling prediction model. Finally, the model parameters are updated according to the simulation results, and operation optimization instructions are generated to realize the coordinated control of multiple magnetic levitation fans, effectively solving problems such as asynchronous startup and back pressure mutation, and improving the stability and efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of a bearing protection method for a magnetic levitation fan under emergency speed change conditions based on digital twins of the present invention.

[0031] Figure 2 This is a structural schematic diagram of a bearing protection system for an emergency speed change condition of a magnetic levitation fan based on digital twins of the present invention. DETAILED DESCRIPTION

[0032] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] like Figure 1-2 In this embodiment, a bearing protection method for a magnetic levitation fan under emergency speed change conditions based on digital twins may specifically include:

[0034] S101 obtains the operating status data of each fan in the air duct, and performs data calibration and feature extraction to form time series data of the fan operating status.

[0035] In this embodiment, the digital twin system monitors the operating status of multiple magnetically suspended fans within the duct in real time, capturing potential anomalies such as speed fluctuations exceeding the rated value by ±2%, pressure deviations from the set value by ±3%, or flow deviations from the rated value by ±5%. Relying on a high-precision sensor network, it collects key operating parameters and, through multi-stage data processing, generates reliable time series data, providing a foundation for analysis.

[0036] S1011 uses sensors to collect first speed data, first pressure data and first flow data of each magnetic levitation fan in the air duct at a preset sampling frequency, and performs linear calibration according to sensor calibration parameters to generate second speed data, second pressure data and second flow data.

[0037] In this embodiment, the sampling frequency is set to once every 100 milliseconds, and the sensor ranges are 0 to 10,000 rpm for speed, 0 to 5,000 Pa for pressure, and 0 to 100 cubic meters per minute for flow. The calibration process utilizes sensor sensitivity parameters, such as 0.1 volts per rpm for the speed sensor, 0.001 volts per Pa for the pressure sensor, and 0.01 volts per cubic meter per minute for the flow sensor, to eliminate hardware bias. The calibrated secondary data accurately reflects the wind turbine's operating status and is suitable for subsequent processing.

[0038] S1012 uses a sliding average filtering method to suppress noise for the second speed data, the second pressure data and the second flow data to generate third speed data, third pressure data and third flow data, and uses a cubic spline interpolation algorithm to fill in data points with abnormal timestamp intervals to generate fourth speed data, fourth pressure data and fourth flow data.

[0039] In the embodiment of the present application, a sliding average filter uses a window of five data points to effectively smooth out noise caused by electromagnetic interference or mechanical vibration. For example, the speed data sequence at a certain moment is 4521, 4498, 4607, 4492, and 4533 revolutions per minute. After filtering, a smoothed value of 4530 revolutions per minute is obtained. The noise threshold is set to plus or minus 5% of the average value to remove outliers. For missing data with a timestamp interval of more than 200 milliseconds, cubic spline interpolation is used, based on three data points before and after, to ensure data continuity and smoothness.

[0040] S1013 divides the fourth speed data, the fourth pressure data and the fourth flow data into time windows, uses wavelet transform to extract signal eigenvalues ​​to generate a first characteristic sequence, performs time series prediction through a long short-term memory network algorithm to generate a second characteristic sequence, and generates time series data of the fan operating status through Gaussian distribution fitting.

[0041] In the embodiment of the present application, the time window is set to 10 seconds, and the statistical features within the window, including the mean, variance and kurtosis, are extracted by wavelet transform to generate a first characteristic sequence that reflects the dynamic characteristics of the wind turbine operation. Based on the first characteristic sequence, a long short-term memory network model is constructed. The hidden layer contains 128 memory units, the training data accounts for 80%, the verification data accounts for 20%, and the training rounds are 1000 rounds. The predicted second characteristic sequence is fitted by Gaussian distribution, and the mean and standard deviation are calculated to generate time series data. If the operating status at a certain moment exceeds the range of 3 times the standard deviation, it is judged to be abnormal.

[0042] In the embodiments of this application, multi-level data processing and feature extraction are used to generate high-precision time series data on wind turbine operating status, providing a reliable basis for backpressure mutation analysis and oscillation disturbance detection. Compared to traditional static models, this method can dynamically adapt to data fluctuations under complex operating conditions, significantly improving the robustness and accuracy of data processing, thereby laying a solid foundation for bearing protection and airflow optimization. It can effectively reduce the risk of misjudgment due to inaccurate data and provide more efficient technical support for the coordinated operation of multiple wind turbines.

[0043] S102 determines whether there is an asynchronous startup phenomenon, and calculates the back pressure mutation distribution in the air duct based on the flow field solution module to generate spatial variation trend data of the back pressure mutation.

[0044] In an embodiment of the present application, potential asynchronous startup is detected by analyzing the time series data of the fan operating status. Asynchronous startup may be caused by deviations in the fan startup timing or external disturbances, resulting in a sudden change in backpressure in the air duct, which in turn affects the mechanical stability of the bearing. Through high-precision flow field solution, the spatial distribution characteristics of backpressure are accurately captured, providing a key basis for bearing protection and airflow optimization. This method can effectively deal with flow field distortion under complex working conditions and significantly improve the robustness of system operation.

[0045] S1021 obtains the startup time series data of multiple magnetic levitation fans from the air duct controller, calculates the startup time difference between adjacent fans using the maximum and minimum normalization method, generates a first startup delay sequence, and analyzes the startup delay correlation through the cross-correlation function to generate a second startup delay sequence.

[0046] In an embodiment of the present application, the startup time series data is recorded in real time by the air duct controller, including the startup time of each fan. The startup time difference of adjacent fans is normalized by the maximum and minimum normalization method in the range of 0 to 1 to generate a first startup delay sequence. For example, in a system containing three fans, the startup times are 0 seconds, 2.5 seconds and 4.8 seconds respectively, and the time differences are calculated to be 2.5 seconds and 2.3 seconds, which are 0.52 and 0.48 respectively after normalization. Based on the first startup delay sequence, the cross-correlation function is used to calculate the temporal correlation of the startup delays of adjacent fans, and the data is divided by a 10-second sliding window. In each window, it is compared with the preset synchronization threshold of 0.3 to generate a second startup delay sequence. This process can effectively identify the timing characteristics of asynchronous startup and lay the foundation for flow field analysis.

[0047] S1022 obtains the duct geometric parameters for the delay value that exceeds the preset synchronization threshold in the second startup delay sequence, discretizes the duct space using the hexahedral mesh division method, generates duct mesh data, and solves the pressure field distribution through the finite volume algorithm to generate the first back pressure distribution data.

[0048] In an embodiment of the present application, when it is detected that the delay value in the second startup delay sequence exceeds 0.3, it is determined that there is an asynchronous startup phenomenon, and the geometric parameters are then obtained from the air duct controller, including the air duct length of 12 meters, the cross-sectional area of ​​0.5 square meters, and the position of the 90-degree elbow at 6 meters. The air duct space is discretized using a hexahedral grid, divided into 240 nodes along the way and 10×10 grids horizontally, generating a total of 24,000 grid units. Based on the air duct grid data, the system uses a finite volume algorithm based on pressure correction to iteratively solve the pressure field in the air duct and generate the first back pressure distribution data. The calculation results show that the pressure value fluctuates between 800 Pa and 1200 Pa, and a significant pressure gradient is formed at the elbow due to the sudden change in air flow direction. This method can efficiently simulate the flow field characteristics in complex air ducts and provide high-precision data support for back pressure mutation analysis.

[0049] S1023 uses the central difference method to calculate the pressure gradient between adjacent grid nodes for the first backpressure distribution data to generate the second backpressure distribution data, and uses the pressure gradient threshold to determine the range of the backpressure mutation area. The cubic spline interpolation method is used to generate the spatial variation trend data of the backpressure mutation.

[0050] In an embodiment of the present application, based on the first backpressure distribution data, the central difference method is used to calculate the pressure gradient between adjacent grid nodes to generate the second backpressure distribution data. The pressure gradient threshold is set to 50 Pa per meter. When the local gradient exceeds the threshold, it is determined to be a backpressure mutation area. For example, the pressure on the inner wall of the duct elbow reaches 1180 Pa, and the outer wall drops to 860 Pa. The pressure gradient is as high as 75 Pa per meter, indicating that there is a significant backpressure mutation. For the second backpressure distribution data in the mutation area, the cubic spline interpolation method is used to fit the discrete pressure points to generate a continuous backpressure spatial variation curve, which intuitively reflects the distribution law of the backpressure along the axial direction of the duct. The calculation results show that a pressure fluctuation band is formed within 3 meters downstream of the elbow, and the pulsation amplitude reaches 15% of the average pressure, which is easy to induce vortex and aerodynamic coupling effects. Through refined flow field analysis, the backpressure mutation area and its impact on the upstream fan can be accurately located, providing a reliable basis for bearing protection.

[0051] In the embodiment of the present application, through multi-level timing analysis and flow field solution, it is possible to dynamically detect the back pressure mutation caused by asynchronous startup and generate high-precision spatial variation trend data. Compared with traditional static control methods, this method can capture the flow field distortion and pressure pulsation characteristics in the air duct in real time, effectively reducing the risk of bearing reverse impact caused by back pressure mutation. At the same time, refined grid division and interpolation fitting improve the calculation accuracy of back pressure distribution, provide a solid data foundation for airflow optimization and coordinated control, and significantly improve the operational stability and safety of the multi-fan system.

[0052] S103 uses finite element analysis to calculate the stress distribution of the downstream magnetic levitation fan bearing under reverse impact, and combines the fan operating parameters such as speed and power to determine the dynamic impact range of the reverse impact on the bearing.

[0053] In the embodiments of this application, transient shocks caused by sudden changes in back pressure pose a significant threat to the mechanical stability of the magnetic levitation wind turbine bearings. The digital twin system simulates the stress response of the bearings under complex loads through finite element analysis. Combining real-time speed and power data, it constructs a dynamic mapping relationship between stress distribution and operating status. Compared to traditional static analysis methods, it can efficiently identify weak areas of the bearing, providing a high-precision basis for optimizing protection strategies, and significantly improving the predictive capability and reliability under dynamic conditions.

[0054] S1031 uses tetrahedral elements to mesh the bearing inner ring, rolling elements, and outer ring, sets the mechanical parameters of the bearing steel material, applies boundary constraints, and generates initial finite element mesh data.

[0055] In the embodiment of the present application, the bearing structure includes an inner ring with a diameter of 120 mm, an outer ring with a diameter of 180 mm and 16 rolling elements. GCr15 bearing steel is selected, with an elastic modulus of 210 GPa and a Poisson's ratio of 0.3. The system uses tetrahedral elements for meshing, with an initial mesh size of 2 mm, generating approximately 85,000 elements. In the boundary condition setting, the connection surface between the outer ring and the casing is fixed, and a speed load is applied to the connection surface between the inner ring and the rotating shaft. To ensure the calculation accuracy, the mesh distortion rate is calculated using a mesh quality detection algorithm. For 15% of the elements with a distortion rate exceeding 0.85 in the transition area of ​​the raceway curvature, local encryption is performed, and the mesh size is reduced to 1 mm to generate optimized high-precision mesh data. This process effectively reduces the numerical calculation error and provides a reliable geometric basis for stress analysis.

[0056] S1032 applies a sudden back pressure load to the optimized mesh data, calculates the transient response of the bearing using explicit dynamic equations, generates dynamic response data, and analyzes the stress distribution law through a stress transfer path tracing algorithm.

[0057] In the embodiment of the present application, the sudden change in back pressure load is manifested as circumferential non-uniform pressure with a maximum value of 1200 Pa and an impact duration of 0.2 seconds. Based on the optimized grid data, the explicit dynamic equation is used to iteratively solve the response of the bearing under transient load, calculate the acceleration and displacement of each node, and generate dynamic response data. The results show that the maximum radial displacement of the inner ring is 0.08 mm, and the maximum acceleration of the outer ring is 50 meters per square second. Further, through the stress transfer path tracing algorithm, the transmission law of the stress wave from the inner ring through the rolling element to the outer ring is analyzed, and it is found that there is stress concentration in the contact area of ​​the rolling element, the maximum vonMises stress reaches 420 MPa, and the transfer path is inclined at 45 degrees. This algorithm reveals the formation mechanism of stress concentration by tracking the stress gradient between nodes, providing theoretical support for identifying weak areas of the bearing.

[0058] S1033 obtains speed and power data from the fan controller, uses the support vector regression algorithm to establish the mapping relationship between operating parameters and stress distribution, and combines the stress superposition criterion and the von Mises criterion to determine the dynamic influence range of the bearing.

[0059] In the embodiment of the present application, the rated speed of the wind turbine is 15,000 revolutions per minute and the power is 200 kilowatts. The system uses the support vector regression algorithm and the Gaussian kernel function as the basis to construct a nonlinear mapping model of speed, power and bearing stress. The algorithm optimizes the parameters through 5-fold cross-validation, selects a penalty factor of 100 and a kernel function parameter of 0.01, and predicts that the basic life of the bearing under rated conditions is 50,000 hours. Combining the dynamic response data with the distribution of the stress transfer path, the system uses the stress superposition criterion to calculate the comprehensive stress state of the bearing and determines the dynamic influence range through the vonMises criterion. The results show that the affected area is concentrated in the contact area of ​​the rolling element on the load-bearing side, covering an angle range of 120 degrees, and the stress fluctuation amplitude of the inner ring surface is 35% of the average stress. This area is highly consistent with the actual wear pattern, which verifies the accuracy of the method. Through multi-physics field coupling analysis, the accuracy of the bearing force prediction is significantly improved, providing data support for extending the bearing life.

[0060] In this application's examples, refined finite element modeling and dynamic stress analysis comprehensively capture the impact of sudden changes in back pressure on bearings. Compared to traditional methods, this approach offers significant advantages in mesh optimization, stress transfer analysis, and operating parameter mapping. It can identify stress concentration areas and predict potential damage risks in real time, providing efficient and reliable technical support for bearing protection in magnetic levitation wind turbines under emergency speed-shifting conditions.

[0061] S104 deploys multiple types of sensors to collect disturbance data from nearby wind turbines, extracts characteristic parameters of oscillation disturbances, and uses adaptive filtering algorithms to optimize control signals and establish a wind turbine control instruction set.

[0062] In an embodiment of the present application, during the operation of a magnetic levitation fan, oscillation disturbances between adjacent fans may be transmitted through vibration, pressure pulsation, or acoustic noise, affecting bearing stability and system efficiency. The digital twin system captures disturbance characteristics in real time through high-frequency synchronous sampling and multi-dimensional signal processing, combines adaptive filtering technology to filter out harmful frequency components, and generates optimized control instructions. This method can significantly reduce bearing vibration and control conflicts caused by disturbances, improve the stability and safety of the coordinated operation of multiple fans, and provide reliable protection for fire smoke exhaust systems in high-rise buildings.

[0063] S1041 deploys vibration, pressure, acoustic, and magnetic levitation sensors within the dynamic impact range, acquires multi-source signal data through high-frequency synchronous sampling, and generates unified time series data through timestamp alignment and resampling processing.

[0064] In an embodiment of the present application, sensors are deployed in the dynamic impact area of ​​the bearing, including a vibration sensor with a range of 0 to 100 meters per square second, a pressure sensor with a range of 0 to 2000 Pa, an acoustic sensor with a frequency response of 20 Hz to 20 kHz, and a magnetic levitation sensor covering a displacement of plus or minus 1 mm, a temperature of 0 to 150 degrees Celsius, and a current of 0 to 20 amps. The sampling frequency is set to 10 kHz to meet the Shannon sampling theorem's requirements for capturing high-frequency signals. The time delay caused by hardware differences in the sensors, such as a vibration signal delay of 0.5 milliseconds and a pressure signal delay of 0.8 milliseconds, is corrected using a timestamp alignment method. The signals are unified to a consistent time base through a resampling algorithm based on the Shannon sampling theorem to generate a second sensor data sequence. This process ensures the temporal consistency of multi-source signals and provides a high-precision data foundation for subsequent feature extraction.

[0065] S1042 uses wavelet decomposition algorithm to extract frequency and amplitude features for unified time series data, eliminates noise interference through anomaly detection, generates a preliminary disturbance feature sequence, and further extracts the main frequency components in combination with fast Fourier transform.

[0066] In this embodiment, the system applies a four-layer wavelet decomposition to the second sensor data sequence, focusing on the 50 Hz to 2000 Hz range for vibration signals, the 20 Hz to 500 Hz range for pressure signals, and the 100 Hz to 5000 Hz range for acoustic signals. Frequency and amplitude characteristics are extracted for each frequency band. Anomaly detection is based on a standard deviation method, with a threshold set at three standard deviations of the mean. This method eliminates abnormal fluctuations caused by transient interference or sensor failure, generating a first disturbance signature sequence. Furthermore, a fast Fourier transform (FFT) combined with a 1024-point Hanning window function is used to calculate the signal spectral distribution. A peak detection algorithm is used to extract key frequency components, such as the 300 Hz rotation frequency and its 600 Hz multiple of the vibration signal, and the 450 Hz leaf frequency of the pressure signal. Magnetic bearing data shows temperature fluctuations between 35 and 45 degrees Celsius and current fluctuations within ±15% of the rated value. Spectral analysis is then used to generate a second disturbance signature sequence. This method, through multi-scale signal decomposition and spectral analysis, accurately captures disturbance characteristics, providing a reliable basis for filter optimization.

[0067] Based on the disturbance characteristic sequence, S1043 uses the recursive least squares algorithm to construct an adaptive filter to filter out harmful frequency components in the control signal and generate an optimized control instruction set, including speed, current and displacement instructions.

[0068] In an embodiment of the present application, for the second disturbance characteristic sequence, the main disturbance frequencies of 300 Hz, 450 Hz and 600 Hz are determined by amplitude sorting, and the disturbance type is judged by comparing with the preset frequency range to generate a frequency characteristic vector. A 32nd-order recursive least squares algorithm is used to construct an adaptive filter, and the convergence coefficient is set to 0.01 and the forgetting factor is set to 0.98 to achieve a 35-decibel attenuation for the main disturbance frequency. The filtered control signal is converted into a first control instruction sequence through proportional transformation, which includes a speed instruction of 12,000 to 18,000 revolutions per minute, a current instruction of 5 to 15 amperes, and a displacement instruction of plus or minus 0.5 mm, and finally a fan control instruction set is constructed. This algorithm dynamically adjusts the filter parameters to suppress the interference of disturbances on the control system in real time, ensuring the stability and accuracy of the instructions.

[0069] In the embodiments of this application, multi-sensor collaborative sampling and adaptive filtering technology are used to comprehensively capture the dynamic coupling effects between adjacent wind turbines. For example, the correlation coefficient of vibration signals between wind turbines 2 meters apart reaches 0.85, and the transmission loss is approximately 6 decibels per meter, reflecting the inherent vibration characteristics of the supporting structure. This method can effectively reduce the speed fluctuations and bearing overload risks caused by disturbances, optimize the execution efficiency of control instructions, thereby improving the coordination of multi-wind turbine systems and the effectiveness of fire and smoke exhaust, and providing efficient technical support for equipment protection under complex working conditions.

[0070] S105 optimizes the pressure gradient in the turbulent airflow area in the air duct in real time through the fan control instruction set combined with the back pressure mutation distribution data to obtain the airflow distribution data.

[0071] In the embodiments of this application, during the operation of a magnetic levitation fan group, sudden changes in back pressure within the air duct often cause airflow turbulence, leading to eddies and pressure pulsations, which in turn affect bearing stability and smoke exhaust efficiency. Through high-precision flow field calculation and neural network optimization, the airflow distribution is dynamically adjusted, significantly reducing the turbulence intensity and pressure fluctuations in the turbulent areas. This method, through multi-scale grid division and recursive calculation, not only improves the accuracy of flow field prediction, but also effectively reduces the impact of airflow turbulence on downstream equipment, providing higher reliability and safety for fire smoke exhaust systems.

[0072] Based on the fan control instruction set and the sudden change distribution of back pressure, S1051 uses hexahedral grids to divide the air duct space, combines the Reynolds average equation to calculate the velocity field and pressure field, generates initial flow field distribution data, and generates a high-precision calculation grid through grid quality optimization.

[0073] In the embodiment of the present application, the air duct is 12 meters long, with a cross-sectional area of ​​0.5 square meters, and includes a 90-degree elbow. The flow field is divided into hexahedral grids, 240 nodes are set along the way, and 10×10 grids are divided horizontally, generating a total of 24,000 grid units. Based on the control instruction set, the velocity field and pressure field of the grid nodes are iteratively solved by the Reynolds average equation to generate the first flow field distribution data, which shows that the mainstream velocity is about 15 meters per second, and the local eddy velocity in the elbow area is reduced to 8 meters per second. In order to ensure the accuracy of the calculation, the grid distortion rate criterion is used to evaluate the grid quality, and the aspect ratio threshold is set to 5 and the distortion angle threshold is set to 60 degrees. The grid units with excessive distortion in the elbows and turbulent areas are encrypted, and the size is reduced to 1 / 2 of the original, and the total number of grids is increased to 35,000. The optimized grid can more accurately capture the characteristics of complex flow fields, providing a reliable basis for turbulence analysis.

[0074] S1052 uses the vorticity conservation equation to calculate the airflow rotation intensity based on the optimized grid data, identifies the turbulent area through the turbulence threshold, and uses a recursive neural network to optimize the pressure gradient in the turbulent area to generate optimized flow field data.

[0075] In an embodiment of the present application, based on the second calculation grid data, the vortex conservation equation is used to calculate the airflow rotation intensity, and it is found that there is a significant vortex street structure within 3 meters downstream of the elbow. The area with local turbulence exceeding 20% ​​is marked as a turbulent area, involving about 5,000 grid cells, accounting for 15% of the total grid. For these turbulent areas, a three-layer recursive neural network is constructed, the input layer matches the number of grid nodes, the hidden layer contains 128 neurons, and the hyperbolic tangent activation function is used. The network is trained with standard flow field data, and the parameters are optimized through back propagation. The calculation results reduce the pressure gradient in the turbulent area from 75 Pa per meter to 45 Pa per meter. Further, through the pressure correction method, the second velocity field is iteratively calculated based on the velocity divergence value. When the adjacent iteration error is less than 0.1%, it is determined to converge and generate the second flow field distribution data. This process effectively weakens the secondary flow and vortex intensity in the turbulent area and improves the flow field stability.

[0076] Based on the optimized flow field data, S1053 uses the finite volume method combined with the mass conservation and momentum conservation equations to solve the airflow parameters, generate optimized airflow distribution data, and significantly improve the uniformity of pressure distribution.

[0077] In an embodiment of the present application, the system constructs the flow field control equation based on the second flow field distribution data using the finite volume method, constrains the airflow density change to less than 1% through mass conservation, and controls the pressure fluctuation within a preset range through momentum conservation. The calculation results generate the third flow field distribution data, which shows that the maximum deviation of the airflow velocity field is reduced from 45% to 25%, the pressure on the outer wall of the elbow is reduced from 1200 Pa to 980 Pa, and the inner wall is increased from 860 Pa to 920 Pa, and the pressure distribution is more uniform. In the optimized flow field, the vortex street structure is more regular, the pressure pulsation amplitude is reduced, and the axial force fluctuation is reduced to 65% of the original. This method significantly reduces the impact load of airflow turbulence on the magnetic levitation bearing through multi-physical field constraints and refined solutions, providing technical support for the long-term stable operation of the system.

[0078] In this embodiment, multi-level flow field optimization and grid densification are used to adjust the airflow distribution within the duct in real time, significantly improving the flow field characteristics in turbulent areas. Compared to traditional static adjustment methods, this method can dynamically adapt to sudden changes in backpressure and complex geometric conditions, reducing vibration and pressure pulsation caused by eddy currents, thereby reducing mechanical stress on bearings, improving the efficiency of multi-fan coordinated operation, and the overall performance of the fire smoke exhaust system.

[0079] If S106 detects a sudden drop in exhaust efficiency or smoke backflow in the optimized airflow distribution data, it adjusts the multi-fan operation timing precisely at the nanosecond level through the collaborative control module to generate a collaborative operation timing plan.

[0080] In this application example, a sudden drop in smoke velocity or backflow from a magnetically levitated fan cluster during a fire smoke exhaust scenario could potentially pose a safety hazard to the refuge floor. The digital twin system monitors smoke flow parameters in real time, combining deep learning with optimization algorithms to rapidly identify abnormal conditions and adjust fan operation timing. This method utilizes nanosecond-level control accuracy to significantly improve the system's responsiveness to dynamic disturbances, ensuring smoke exhaust efficiency and system safety. Compared to traditional static sequence control, it significantly reduces the risk of smoke backflow, providing highly effective protection for high-rise building firefighting.

[0081] Based on the optimized airflow distribution data, S1061 uses the differential method to calculate the flue gas flow velocity and pressure change rate, mark abnormal areas, generate abnormal state data, and fit the flow parameter curve through the least squares method to quantify the degree of abnormality.

[0082] In this application embodiment, the system applies the differential method to the flue gas flow rate and pressure data, calculates its time rate of change, marks the area where the flow rate is lower than 50% of the rated value, that is, 7.5 meters per second, or where negative pressure occurs, and generates the first abnormal state data. For example, under a certain abnormal operating condition, the flue gas flow rate drops from 15 meters per second to 4 meters per second within 5 seconds, and the pressure drops from 1000 Pa to -200 Pa, indicating a serious smoke exhaust anomaly. For the first abnormal state data, the system collects 100 sampling points, uses the least squares method to fit the flue gas flow parameter curve, and calculates the degree of fit to evaluate the degree of fluctuation. If the degree of fit is lower than 0.85, it is determined to be a severe flow anomaly, and the first flow characteristic data is generated. This method quantifies the dynamic characteristics of flue gas flow through mathematical fitting, providing accurate data support for subsequent anomaly classification.

[0083] S1062 uses a convolutional neural network to build a smoke exhaust anomaly identifier, classify abnormal conditions, generate fault feature vectors, and calculate the fan operation timing compensation value through the particle swarm optimization algorithm to generate an initial timing adjustment plan.

[0084] In this application embodiment, a convolutional neural network based on the VGG structure is used, which includes 3 convolutional layers and 2 fully connected layers. The convolution kernel size is 3×3, and the pooling layer uses maximum pooling. The input is the flue gas flow parameter matrix. The network is trained with 1,000 sets of smoke exhaust plummets and 500 sets of smoke backflow historical data, and the anomaly recognition accuracy rate reaches 95%. Based on the first flow feature data, the network classification generates the first fault feature vector, which reflects the type and degree of the anomaly. For this vector, the system applies the particle swarm optimization algorithm, sets 50 particles, iterates 200 rounds, and limits the search space to plus or minus 100 nanoseconds. The fitness function evaluates the effect of timing adjustment on the recovery of flow rate and pressure. The optimization result generates the first running timing data. For example, the startup delay compensation values ​​of adjacent fans are 85 nanoseconds and -65 nanoseconds, respectively. This process achieves efficient timing adjustment through intelligent optimization, laying the foundation for collaborative control.

[0085] S1063 verifies the parameter constraints of the initial timing adjustment plan, constructs a collaborative control parameter matrix, generates a smooth transition sequence through linear interpolation, and outputs a collaborative operation timing plan.

[0086] In this application embodiment, the first operating sequence data is verified to ensure that the speed adjustment does not exceed 5% of the rated speed, the current adjustment does not exceed 10% of the rated current, and the bearing displacement does not exceed 0.5 mm. After verification, the second operating sequence data is generated, showing that the speed adjustment of the three fans is plus or minus 750 revolutions per minute, the current adjustment is plus or minus 1.5 amperes, and the displacement adjustment is plus or minus 0.3 mm. The system integrates the speed, current, and displacement parameters of each fan through a 6×6 collaborative control parameter matrix, and uses linear interpolation to calculate the transition sequence of 100 control time points to generate a smooth collaborative operation sequence plan. After adjustment, the flue gas flow rate returned to 13 meters per second, the pressure returned to 900 Pa, and the system operated stably. This method uses matrix control and interpolation smoothing to achieve efficient coordination of multiple fans and significantly improve the dynamic response capability of the smoke exhaust system.

[0087] Through multi-level anomaly detection and timing optimization, we can quickly respond to smoke exhaust anomalies and ensure the stable operation of the fire smoke exhaust system. Compared with traditional methods, this method combines deep learning with nanosecond-level control to greatly improve the accuracy of anomaly identification and the efficiency of timing adjustment, effectively preventing safety hazards such as smoke backflow and providing reliable technical support for the coordinated operation of multiple fans.

[0088] S107 simulates the air duct flow field in real time, solves the coupling effect between multiple magnetic levitation fans, builds a cross-fan coupling prediction model, and generates predicted values ​​of pressure and flow changes to optimize system operation.

[0089] In the embodiment of the present application, during the coordinated operation of a group of magnetically levitated fans, the aerodynamic coupling effect between fans significantly affects the flow field stability and smoke exhaust efficiency. The digital twin system simulates the interaction between cross-fan pressure and flow in real time through virtual mapping and high-precision flow field solution, and generates accurate prediction values ​​by combining the long-short-term memory network prediction coupling characteristics. This method can not only dynamically capture complex flow field changes, but also provide data support for the coordinated control of multiple fans. Compared with traditional static simulation, it significantly improves the prediction accuracy and system robustness, providing efficient protection for fire smoke exhaust scenarios.

[0090] S1071 builds a virtual mapping model based on the geometric structure of the air duct and the collaborative operation timing scheme, divides the flow field grid using the Lagrangian particle method, calculates the initial flow field state through the Reynolds average equation, and generates the first flow field state data.

[0091] In the embodiment of the present application, the air duct is 12 meters long, with a cross-sectional area of ​​0.5 square meters, including a 90-degree elbow at 6 meters. According to the position of the fan and the size of the pipe, a digital twin mapping model with a 1:1 ratio to the physical system is constructed, and boundary conditions such as inlet velocity and outlet pressure are set to generate the first virtual scene data. The Lagrangian particle method is used to divide the flow field, the standard segment grid size is 50 mm, the cross-fan coupling area is encrypted to 25 mm, the impeller area uses a rotating grid, and the internal and external interfaces use sliding grid technology. The total number of grids reaches 480,000, of which the coupling area accounts for 35%. The first flow field state data is generated by iteratively solving the Reynolds average equation, which shows that the airflow velocity is about 15 meters per second and the pressure distribution is between 800 and 1200 Pa. This method ensures the high accuracy of the initial flow field through fine grids and dynamic boundary processing, laying the foundation for coupling analysis.

[0092] Based on the initial flow field state, S1072 constructs a long short-term memory network predictor, extracts the pressure and flow coupling characteristics, generates coupling feature vectors, and optimizes model parameters through recursive training to improve prediction accuracy.

[0093] In an embodiment of the present application, a long short-term memory network is used, and a multi-layer structure is designed. The input layer receives the pressure transmission and flow mutual feedback characteristics, and the hidden layer contains 128 memory units, of which the pressure and flow coupling units have 64 neurons each. The training data includes 1,000 sets of historical operation records, covering normal and disturbed working conditions. The network parameters are optimized through recursive training, and the prediction accuracy of the validation set reaches 92%. The network outputs the first coupling eigenvector, which reflects the interaction law between pressure and flow between fans, such as the pressure transmission coefficient of adjacent fans is 0.35 and the flow mutual feedback coefficient is 0.28. Recursive training dynamically adjusts the weights through gradient descent, enhances the adaptability of the model to nonlinear coupling effects, and generates a second coupling eigenvector. This process significantly improves the model's predictive ability for dynamic working conditions and provides reliable feature support for flow field optimization.

[0094] Based on coupled eigenvectors, S1073 uses the finite volume method to solve the flow field control equations, combines adaptive grid refinement and pressure correction to generate flow field state data, and outputs pressure and flow prediction values ​​through least squares fitting.

[0095] For the second coupling eigenvector, the finite volume method is used to solve the flow field control equations, and 250 monitoring points are set up across the wind turbine area to record the instantaneous pressure and flow. The pressure correction method is iteratively calculated, and when the error of the adjacent iterative pressure field is less than 0.1%, it is judged to be converged, and the second flow field state data is generated. For high-gradient areas such as downstream of the elbow, the pressure gradient threshold is set to 50 Pa per meter, the grid size is refined to 10 mm, the total number of grids is increased to 500,000, and the grid convergence change is less than 1%, generating the third flow field state data. The pressure and flow change curves are fitted by the least squares method, and the predicted values ​​are output, showing that when the upstream wind turbine speed fluctuates by 10%, the downstream pressure fluctuation is 65% of the upstream, the flow fluctuation is 58%, and the pressure transfer coefficient in the elbow area increases to 0.42. This method accurately captures the spatial distribution of the coupling effect through adaptive grid and multi-physics field solution, providing high-precision data for optimization.

[0096] Through real-time flow field simulation and intelligent prediction, the aerodynamic coupling characteristics between multiple fans are comprehensively analyzed. Compared with traditional methods, this method has significant advantages in dynamic grid optimization, coupling feature extraction, and predictive model construction. It can effectively monitor the operating status of fan groups, reduce vibration and efficiency losses caused by coupling effects, and provide solid technical support for the stable operation of fire smoke exhaust systems in high-rise buildings.

[0097] S108 obtains the real-time simulation results of the duct flow field by the digital twin system. It updates the parameters of the cross-fan coupling prediction model based on the changes in pressure and flow, and generates operation optimization instructions based on the back pressure mutation, airflow turbulence and bearing health status.

[0098] In the embodiments of this application, during the operation of a multi-magnetic levitation fan system, cross-fan coupling effects and airflow turbulence pose a continuous threat to bearing health. The digital twin system dynamically optimizes coupling model parameters through real-time simulation and multi-source data fusion, comprehensively assesses flow field disturbances and bearing status, and generates precise control instructions. This method utilizes recursive updates and intelligent predictions to significantly improve the system's adaptability to complex operating conditions, reduce the risk of dynamic bearing loads and airflow turbulence, and provide a solid guarantee for the efficient operation of fire smoke exhaust systems.

[0099] Based on the simulation results of the digital twin system, S1081 uses the recursive least squares method to update the pressure transfer and flow feedback coefficients, generates optimized parameter data, and ensures parameter convergence through cross-validation and residual analysis.

[0100] In an embodiment of the present application, the initial pressure transfer coefficient of 0.35, the flow mutual feedback coefficient of 0.28, and the time lag parameter of 0.15 seconds are extracted from the simulation results, and the parameter correction amount is calculated by recursive least squares method to obtain the first optimized parameter data, and the correction values ​​are 0.08 and 0.05 respectively. The error feedback method ensures the stability of parameter updates by iteratively minimizing the prediction error. To verify the reliability, the system applies 5-fold cross-validation, and the historical data are grouped and rotated for testing. The residual analysis shows that the prediction error is less than 5%, confirming the parameter convergence, generating the second optimized parameter data, and the coefficients are updated to 0.43 and 0.33, and the time lag is reduced to 0.12 seconds. This method enhances the adaptability of the coupling model to the dynamic flow field through online updating and verification.

[0101] S1082 uses optimized parameter data to build a long short-term memory network predictor, calculates pressure field and flow field disturbances, generates a state prediction sequence, and extracts time-frequency features through bearing sensor data to evaluate bearing health status.

[0102] In this embodiment, a three-layer long short-term memory (LSTM) network is employed. The input layer receives the second optimization parameter data and flow field data, while the hidden layer contains 128 memory cells with a sampling period of 10 milliseconds, focusing on monitoring areas of turbulent airflow. The network predicts pressure field fluctuations and flow field disturbances, generating a first state prediction sequence, showing that the turbulence level decreases from 25% to 15% under sudden changes in backpressure. Simultaneously, the bearing monitoring sensor collects vibration signals with a range of 0 to 100 meters per square second, temperature of 0 to 150 degrees Celsius, and current of 0 to 20 amperes. A four-layer wavelet transform is used to extract features in the 50 Hz to 2000 Hz frequency band to generate the first bearing eigenvector. Stress superposition criterion calculations show that the dynamic load fluctuation of the bearing under turbulent conditions reaches 35% of the rated value. B10 life predictions indicate that life decreases significantly when the load exceeds the rated value by 40%, with a temperature increase of 5 to 8 degrees Celsius and a current fluctuation of 15%. This process, through time series prediction and multidimensional feature analysis, accurately captures the dynamic interaction between the flow field and the bearing, providing comprehensive data for optimized control.

[0103] S1083 integrates the airflow turbulence characteristics and bearing health status, and uses weighted fusion and dynamic programming algorithms to calculate the optimal combination of speed, current, and displacement instructions to generate the final operation optimization instructions.

[0104] In an embodiment of the present application, the system adopts a weighted fusion method, assigning a weight of 0.6 to the airflow turbulence feature and a weight of 0.4 to the bearing health status, and combines the first state prediction sequence with the first bearing feature vector to generate health assessment data. The dynamic programming algorithm is based on multi-objective constraints, optimizes the speed adjustment amount to be controlled within the rated speed of plus or minus 5%, the current adjustment amount to plus or minus 10%, and the displacement amount to plus or minus 0.3 mm, calculates the optimal instruction combination, and generates operation optimization instructions. After optimization, the airflow turbulence is reduced to below 10%, the dynamic load fluctuation of the bearing is reduced to 25% of the rated value, and the temperature rise is controlled at 3 to 5 degrees Celsius. This method achieves a synergistic improvement of airflow stability and bearing life through multi-objective optimization and weight balance, significantly enhancing the system operation reliability.

[0105] In this application's embodiment, high-precision operational optimization instructions are generated through online parameter updates, flow field disturbance prediction, and bearing condition monitoring. Compared to traditional control methods, this method offers significant advantages in multi-source data fusion and dynamic optimization. It can effectively address sudden changes in backpressure and airflow disturbances, ensuring stable operation and efficient firefighting and smoke extraction for multi-fan systems under complex operating conditions.

[0106] The present invention provides a bearing protection system for emergency speed change of a magnetic levitation fan based on digital twin, which mainly includes:

[0107] A data acquisition module is used to obtain time series data of the operating status of each magnetic suspension fan in the air duct;

[0108] The asynchronous start detection module is used to determine whether there is an asynchronous start phenomenon based on time series data. If the deviation between the start time of a certain fan and that of other fans exceeds a preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct;

[0109] The back pressure mutation analysis module is used to analyze the stress distribution of the downstream bearing under reverse impact based on the spatial variation trend of the back pressure mutation, and determine the dynamic impact range of the reverse impact on the bearing in combination with the wind turbine operating parameters;

[0110] The bearing stress analysis module is used to obtain disturbance data of adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculate the frequency and amplitude of the oscillation disturbance, filter out the disturbance frequency component of the control signal, and obtain the wind turbine control instruction set;

[0111] The disturbance data processing module is used to adjust the pressure gradient in the turbulent airflow area in the air duct in real time through the fan control instruction set and the calculation results of the sudden change distribution of back pressure in the air duct to obtain the optimized airflow distribution;

[0112] The airflow optimization module is used to coordinate the operation timing of multiple fans and generate a coordinated operation timing plan if a sudden drop in smoke exhaust or smoke backflow is detected;

[0113] The collaborative control module is used to simulate the duct flow field in real time through the digital twin system according to the collaborative operation timing plan, solve the pressure and flow changes under the cross-fan coupling effect, and output the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans;

[0114] The simulation and optimization module is used to update the parameters of cross-turbine coupling prediction based on the results of real-time simulation of the duct flow field by the digital twin system, and generate operation optimization instructions based on the real-time changes of back pressure mutations and airflow turbulence as well as the health status of bearings.

[0115] The above only lists some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and modifications can be made. As long as the improvements and modifications are made on the basis of the basic principles of the present invention, they should be considered to fall within the scope of protection of the present invention.

Claims

1. A bearing protection method for a magnetic levitation fan under emergency speed change conditions based on digital twin, characterized in that: The method comprises: Obtain the time series data of the operating status of each magnetic suspension fan in the air duct; Based on the time series data, it is determined whether there is an asynchronous startup phenomenon. If the deviation between the startup time of a certain fan and that of other fans exceeds the preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct. By analyzing the spatial variation trend of back pressure mutations, the stress distribution of the downstream bearing under reverse impact is analyzed, and the dynamic impact range of the reverse impact on the bearing is determined in combination with the wind turbine operating parameters; Obtain disturbance data from adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculate the frequency and amplitude of the oscillation disturbance, filter out the disturbance frequency component of the control signal, and obtain the wind turbine control instruction set; Through the fan control instruction set and the calculation results of the sudden change distribution of back pressure in the air duct, the pressure gradient in the turbulent air flow area in the air duct is adjusted in real time to obtain the optimized air flow distribution; If a sudden drop in exhaust gas or smoke backflow is detected, the operation timing of multiple fans will be coordinated and a coordinated operation timing plan will be generated; According to the collaborative operation timing plan, the digital twin system conducts real-time simulation of the air duct flow field, solves the pressure and flow changes under the cross-fan coupling effect, and outputs the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans; Based on the results of real-time simulation of the air duct flow field by the digital twin system, operation optimization instructions are generated based on the real-time changes of back pressure mutations and airflow turbulence as well as the health status of the bearings.

2. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: The obtaining of time series data of the operating status of each magnetic suspension fan in the air duct includes: Acquire first speed data, first pressure data, and first flow data of the magnetic levitation fan in the air duct from the sensor according to a preset sampling period, perform linear calibration on the first speed data, first pressure data, and first flow data using the built-in sensor sensitivity parameter to obtain second speed data, second pressure data, and second flow data, and perform signal smoothing using a sliding average filtering method to obtain third speed data, third pressure data, and third flow data; For the timestamp marks in the third speed data, the third pressure data, and the third flow data, if the interval time exceeds a preset threshold, a cubic spline interpolation algorithm is used to fill the missing data points to obtain fourth speed data, fourth pressure data, and fourth flow data; The fourth speed data, the fourth pressure data and the fourth flow data are subjected to wavelet transform to extract signal feature values ​​to obtain a first feature sequence, a long short-term memory network algorithm is used to predict the fan operating state through the first feature sequence to obtain a second feature sequence, and a Gaussian distribution fitting is performed on the second feature sequence to obtain the fan operating state time series data.

3. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: The determination of whether there is an asynchronous startup phenomenon based on the time series data is performed. If the deviation between the startup time of a certain fan and that of other fans exceeds a preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct, including: Obtain fan startup time series data from the air duct controller, and calculate the startup time difference between adjacent fans using a maximum and minimum normalization method for the time series data to obtain a first startup delay sequence; Calculating a startup delay correlation degree using a cross-correlation function based on the first startup delay sequence, and comparing the delay value with a preset startup synchronization threshold using a sliding window to obtain a second startup delay sequence; For the delay value exceeding the preset startup synchronization threshold in the second startup delay sequence, the air duct space is discretized using a hexahedral meshing method to obtain air duct mesh data; According to the air duct grid data, the pressure field distribution is solved by using a finite volume algorithm based on pressure correction, and the pressure gradient is calculated by the central difference method to obtain the back pressure mutation area range and spatial variation trend data.

4. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: The method of analyzing the stress distribution of the downstream bearing subjected to the reverse impact by the spatial variation trend of the back pressure mutation and determining the dynamic impact range of the reverse impact on the bearing in combination with the wind turbine operating parameters includes: Use tetrahedron elements to mesh the bearing inner ring, rolling elements, and outer ring structures, set the steel elastic modulus and Poisson's ratio parameters, and generate the first mesh data of the bearing finite element method; For the first mesh data, the mesh distortion rate is calculated using a mesh quality detection method, and the cells whose distortion rate exceeds a preset threshold are locally encrypted to obtain the second mesh data of the bearing finite element; Calculating the transient response of the bearing under the sudden back pressure load using an explicit dynamic equation based on the second grid data, and generating first dynamic response data by calculating the node acceleration and displacement; For the first dynamic response data, the stress transfer law between the bearing components is calculated by the stress transfer path tracing method to obtain the stress transfer path distribution data, and the dynamic influence area of ​​the bearing is determined using the stress superposition criterion.

5. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: The method includes obtaining disturbance data of adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculating the frequency and amplitude of the oscillation disturbance, filtering out the disturbance frequency component of the control signal, and obtaining a wind turbine control instruction set, including: According to the layout of the wind turbine, a vibration sensor, a pressure sensor, an acoustic sensor and a magnetic suspension sensor are used to perform synchronous sampling to obtain a first sensor data sequence; Processing the first sensor data sequence using a timestamp alignment method to obtain a second sensor data sequence; Extracting signal features from the second sensor data sequence and eliminating abnormal fluctuations using an anomaly detection method based on standard deviation to obtain a first disturbance feature sequence; For the first disturbance characteristic sequence, the main frequency components are extracted, and an adaptive filter is constructed in combination with the magnetic bearing data to obtain a fan control instruction set.

6. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: The fan control instruction set is combined with the calculation results of the sudden change distribution of back pressure in the air duct to adjust the pressure gradient in the turbulent air flow area in the air duct in real time to obtain the optimized air flow distribution, including: The air duct space is divided according to the fan control instruction set, and the velocity field and pressure field of the grid nodes are calculated by the Reynolds average equation to obtain the first flow field distribution data; For the first flow field distribution data, a quality assessment is performed on the computational grid using a grid distortion rate criterion. If the grid unit distortion rate exceeds a preset grid quality threshold, the grid unit is encrypted to obtain second computational grid data. Calculating the airflow rotation intensity value using the vorticity conservation equation based on the second calculation grid data, identifying the airflow turbulence area using the turbulence threshold criterion, and obtaining a first airflow turbulence index; For the first airflow turbulence index, a recursive neural network is used to construct a flow field optimization calculation unit, and the airflow parameters in the turbulent area are calculated through the mass conservation and momentum conservation equations to obtain the optimized airflow distribution.

7. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: If a sudden drop in exhaust gas or smoke backflow is detected, the operation timing of multiple fans is coordinated and controlled to generate a coordinated operation timing plan, including: Calculating the flue gas flow rate change rate and pressure change rate using a differential method based on the flue gas flow rate data, marking areas where the flue gas flow rate is lower than a rated flow rate threshold, and obtaining the first abnormal state data; Calculating a smoke flow parameter curve using a least squares method for the first abnormal state data, determining the degree of smoke flow abnormality based on the curve fitting, and obtaining first flow characteristic data; A smoke exhaust anomaly identifier is constructed based on the first flow characteristic data using a three-layer convolutional structure and two fully connected layers to classify and identify abnormal smoke emission states and obtain a first fault feature vector. Calculating a multi-wind turbine operation timing compensation value based on the first fault characteristic vector using a particle swarm optimization algorithm, setting a compensation value adjustment step, and generating first operation timing data; Verify, based on the first operating time series data, whether the fan speed adjustment amount, the current adjustment amount, and the displacement adjustment amount meet the preset range requirements through parameter constraint criteria, and obtain second operating time series data; With respect to the second operation timing data, a transition sequence of each control parameter is calculated to generate a coordinated operation timing solution.

8. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: According to the collaborative operation timing scheme, the digital twin system is used to perform real-time simulation of the air duct flow field, solve the pressure and flow changes under the cross-fan coupling effect, and output the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans, including: A digital twin mapping model is established according to the geometric structure parameters of the air duct, and first virtual scene data is generated through boundary condition constraints, where the first virtual scene data includes fan position information and pipe size information; Dividing the flow field grid using the Lagrangian particle method for the first virtual scene data, increasing the grid density in the cross-wind turbine coupling region, and calculating the first flow field state data using the Reynolds average equation; constructing an inter-turbine coupling predictor based on the first flow field state data, wherein the input layer of the predictor receives pressure transmission characteristics and flow mutual feedback characteristics and outputs a coupling feature vector; The least square method is used to fit the pressure and flow change curves for the coupling eigenvector, and the pressure and flow change prediction values ​​under the coupling effect of multiple magnetic levitation fans are output through the cross-fan coupling predictor.

9. The method for protecting bearings in emergency speed change conditions of a digital twin magnetic levitation wind turbine according to claim 1 is characterized in that: Based on the results of the real-time simulation of the air duct flow field by the digital twin system, operation optimization instructions are generated for the real-time changes of back pressure mutations and airflow turbulence as well as the health status of the bearings, including: The pressure transfer coefficient and flow mutual feedback coefficient are calculated based on the digital twin simulation results, and the parameter correction amount is obtained using the recursive least squares method to obtain the first optimized parameter data; For the first optimization parameter data, a coupled parameter optimizer is constructed using a long short-term memory network, and the pressure field fluctuation value and the flow field disturbance value are obtained through time series prediction to obtain a first state prediction sequence; According to the first state prediction sequence, a bearing monitoring sensor is used to collect vibration signals, temperature signals, and current signals, and time-frequency eigenvalues ​​are obtained through wavelet transform to obtain a first bearing eigenvector; For the first bearing characteristic vector, a weighted fusion method is used to process the airflow turbulence characteristics and the bearing health status, and the combined values ​​of the fan speed instruction, current instruction and displacement instruction are calculated through a dynamic programming algorithm to obtain the operation optimization instruction.

10. A magnetic levitation fan emergency speed change working condition bearing protection system based on digital twin, characterized in that: The system comprises: A data acquisition module is used to obtain time series data of the operating status of each magnetic suspension fan in the air duct; The asynchronous start detection module is used to determine whether there is an asynchronous start phenomenon based on time series data. If the deviation between the start time of a certain fan and that of other fans exceeds a preset threshold, the spatial variation trend of the backpressure mutation is obtained by calculating the backpressure mutation distribution in the air duct; The back pressure mutation analysis module is used to analyze the stress distribution of the downstream bearing under reverse impact based on the spatial variation trend of the back pressure mutation, and determine the dynamic impact range of the reverse impact on the bearing in combination with the wind turbine operating parameters; The bearing stress analysis module is used to obtain disturbance data of adjacent wind turbines within the dynamic influence range and magnetic bearing data, calculate the frequency and amplitude of the oscillation disturbance, filter out the disturbance frequency component of the control signal, and obtain the wind turbine control instruction set; The disturbance data processing module is used to adjust the pressure gradient in the turbulent area of ​​the air duct in real time through the fan control instruction set and the calculation results of the sudden change distribution of back pressure in the air duct to obtain the optimized air flow distribution; The airflow optimization module is used to coordinate the operation timing of multiple fans and generate a coordinated operation timing plan if a sudden drop in smoke exhaust or smoke backflow is detected; The collaborative control module is used to simulate the duct flow field in real time through the digital twin system according to the collaborative operation timing plan, solve the pressure and flow changes under the cross-fan coupling effect, and output the predicted values ​​of pressure and flow changes under the coupling effect of multiple magnetic levitation fans; The simulation and optimization module is used to generate operation optimization instructions based on the results of real-time simulation of the air duct flow field by the digital twin system, targeting the real-time changes of back pressure mutations and airflow turbulence as well as the health status of the bearings.

Images