Elevator inverter fault pre-diagnosis and fault-tolerant control system and method based on digital twinning

Abstract

The invention relates to the technical field of elevator control and fault diagnosis, and discloses an elevator inverter fault pre-diagnosis and fault-tolerant control system and method based on digital twinning, and the system comprises a high-fidelity digital twinning body building module which is used for building a digital twinning body of an elevator inverter, the digital twin integrates an electric-thermal-stress coupling model, a device-level aging model, a parasitic parameter influence model and a driving circuit and protection logic model; wherein the electric-thermal-stress coupling model is used for simulating loss, junction temperature fluctuation and thermal stress circulation of a power device, and the device-level aging model is used for representing a time-varying degradation rule of device parameters. Through deep fusion of a digital twin technology, fault diagnosis and fault-tolerant control, a closed-loop system of'perception-pre-diagnosis-decision-control 'is constructed, the safety, reliability and economical efficiency of the elevator inverter are comprehensively improved, and the method has important significance in promoting transformation of the elevator industry to intelligent and unmanned maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator control and fault diagnosis, and more particularly to an elevator inverter fault pre-diagnosis and fault-tolerant control system and method based on digital twinning. BACKGROUND

[0002] The elevator inverter is the core component of the elevator drive system, and its running state directly affects the safety and reliability of the elevator. The traditional elevator inverter fault diagnosis method mainly relies on real-time data collected by sensors for analysis, which has the following shortcomings:

[0003] Fault diagnosis is lagging, often detected after the fault occurs, difficult to early warning, easy to lead to elevator trapped and other safety accidents;

[0004] It is difficult to capture early weak fault characteristics, as the noise interference in the physical system is large and the early fault signal is weak, the traditional method is difficult to effectively extract;

[0005] Lack of accurate simulation of the aging process of the device, unable to accurately predict the remaining useful life of the key power device;

[0006] The fault-tolerant control strategy is single, and it is difficult to achieve smooth switching when the fault occurs, affecting the continuity of the elevator operation.

[0007] Digital twinning technology provides a new way to solve the above problems. The digital twin can accurately map and simulate the physical entity, but the current digital twin model for elevator inverters mostly only considers a single physical field, with low fidelity, and lacks deep integration with fault diagnosis and fault-tolerant control, making it difficult to meet the actual application requirements.

[0008] Based on this, the present application provides an elevator inverter fault pre-diagnosis and fault-tolerant control system and method based on digital twinning to solve the above problems. SUMMARY

[0009] In order to overcome the above-mentioned defects of the prior art, the present application provides an elevator inverter fault pre-diagnosis and fault-tolerant control system and method based on digital twinning to solve the problems existing in the background art.

[0010] The present application provides the following technical solutions: an elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twinning, comprising:

[0011] A high-fidelity digital twin construction module is configured to establish a digital twin of an elevator inverter, the digital twin integrating an electro-thermal-stress coupling model, a device-level aging model, a parasitic parameter influence model, and a drive circuit and protection logic model; wherein the electro-thermal-stress coupling model is configured to simulate power device loss, junction temperature fluctuation, and thermal stress cycle, the device-level aging model is configured to represent time-varying degradation law of device parameters, the parasitic parameter influence model is configured to reflect the influence of PCB layout and stray parameters on inverter switching characteristics, and the drive circuit and protection logic model is configured to reproduce drive chip behavior and protection logic.

[0012] A real-time fault injection and predictive diagnosis module is in communication connection with the high-fidelity digital twin construction module and physical system sensors, configured to inject a preset fault mode on the digital twin, compare ideal state data output by the digital twin with actual sensing data of the physical system to obtain a deviation signal, extract early fault features based on the deviation signal, and identify fault types and predict the remaining service life of key power devices in combination with the device-level aging model.

[0013] A digital twin online adaptive updating module is in communication connection with the high-fidelity digital twin construction module and physical system sensors, respectively, configured to automatically calibrate model parameters of the digital twin according to the deviation between output data of the digital twin and actual sensing data of the physical system, and update the diagnosis model based on physical system data when a new fault mode or a new operating condition is detected.

[0014] An intelligent fault-tolerant control module is in communication connection with the real-time fault injection and predictive diagnosis module and an elevator main control system, configured to execute active fault-tolerant control according to fault pre-diagnosis results, or switch to a preset fault-tolerant control algorithm to achieve seamless fault tolerance when a fault occurs, and cooperates with the elevator main control system to make a decision on an optimal stop scheme.

[0015] An edge-cloud collaborative deployment module is in communication connection with each of the foregoing modules, configured to deploy high-precision digital twin modeling, complex model training, and long-term trend analysis tasks on a cloud or an edge server, deploy lightweight diagnosis models, real-time feature extraction, and basic fault-tolerant control logic on a local elevator controller, and realize data synchronization and model updating between the edge and the cloud through a low-bandwidth communication protocol.

[0016] As a further scheme of the present application: the electro-thermal-stress coupling model is specifically used for: calculating the switching loss and conduction loss of the IGBT or MOSFET, simulating the junction temperature fluctuation in combination with the heat dissipation condition, and calculating the thermal stress cycle based on the junction temperature change amount; in the device-level aging model, the device parameters include the on-resistance Rds(on), the threshold voltage Vth and the thermal resistance Rth(j-c), and the time-varying degradation law is obtained by fitting the parameter degradation curve obtained by experiment; the parasitic parameter influence model is specifically used for: simulating the influence of the stray inductance and stray capacitance introduced by the PCB layout on the voltage overshoot, current oscillation and electromagnetic interference EMI in the switching process; the driving circuit and protection logic model is specifically used for: reproducing the desaturation detection DESAT of the driving chip, the Miller clamp behavior, and the trigger and response process of the overcurrent, overvoltage and overheat protection logic.

[0017] As a further scheme of the present application: the real-time fault injection and predictive diagnosis module comprises:

[0018] a virtual fault injection unit for injecting potential fault modes on the digital twin in parallel, the potential fault modes including single tube open circuit, single tube short circuit, abnormal driving signal, capacitance value attenuation, current sensor drift, current sensor failure and heat dissipation capacity deterioration;

[0019] a deviation feature extraction unit for processing the deviation signal by wavelet packet decomposition or empirical mode decomposition EMD to obtain multi-band components, and extracting early weak fault features in the multi-band components by 1D-CNN, LSTM or Transformer network, the early weak fault features including current waveform slight distortion, switching time sequence slight shift and temperature rise trend anomaly;

[0020] a fault diagnosis and life prediction unit for comparing the early weak fault features with a preset fault feature library to identify the fault type, and predicting the remaining useful life of the IGBT or MOSFET based on the mapping relationship between the parameter degradation law in the device-level aging model and the fault features.

[0021] As a further scheme of the present application: the twin online adaptive updating module comprises:

[0022] a closed-loop parameter calibration unit for continuously collecting the voltage estimation value, current estimation value and temperature estimation value output by the digital twin, comparing with the actual voltage, current and temperature collected by the physical system sensor, calculating the deviation amount, and automatically adjusting the loss calculation parameter, thermal resistance parameter and parasitic parameter in the digital twin according to the deviation amount;

[0023] An incremental learning unit is configured to, when a new fault mode not included in the preset fault mode library is detected, update the diagnostic model parameters in the real-time fault injection and predictive diagnosis module based on the operation data of the new fault mode collected by the physical system through a transfer learning algorithm, so as to expand the fault recognition range of the model.

[0024] As a further scheme of the present application, the intelligent fault-tolerant control module comprises:

[0025] An active fault-tolerant control unit is configured to, when the real-time fault injection and predictive diagnosis module predicts that a key power device is about to fail, perform at least one of the following operations: reducing the output power of the elevator to reduce the load of the device, switching to a redundant phase topology to reconstruct the inverter output, and sending accurate early warning information to a maintenance system.

[0026] A seamless fault-tolerant switching unit is configured to, when a short-circuit or open-circuit fault of the inverter is detected, switch to a preset fault-tolerant control algorithm within milliseconds, the preset fault-tolerant control algorithm comprising a torque compensation algorithm based on specific harmonic injection and a three-phase imbalance suppression algorithm based on reference vector reconstruction.

[0027] A cooperative control unit is configured to interact with the elevator main control system to obtain the current position, running direction and safety circuit state of the elevator, ensure that safety logics such as elevator overspeed protection and door lock monitoring are effective during fault-tolerant operation, and decide on the nearest safe floor based on the above information to achieve smooth landing.

[0028] As a further scheme of the present application, the low-bandwidth communication protocol is specifically used for transmitting only key feature data, fault diagnosis results, model calibration parameters and update instructions between the edge and the cloud, the key feature data including current distortion coefficient, temperature fluctuation amplitude and switching time deviation.

[0029] Further, the elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twinning comprises the following steps:

[0030] S1. A high-fidelity digital twin of the elevator inverter is constructed, the digital twin fusing an electro-thermal-stress coupling model, a device-level aging model, a parasitic parameter influence model and a driving circuit and protection logic model.

[0031] S2. A preset fault mode is injected into the digital twin to obtain ideal state data simulated by the digital twin, the deviation signal is obtained by comparing the actual sensing data of the physical system, the early fault features are extracted based on the deviation signal, the fault type is identified and the remaining useful life of the key power device is predicted.

[0032] S3. automatically calibrate the model parameters of the digital twin according to the deviation of the output data of the digital twin from the actual sensing data of the physical system, and update the diagnostic model based on the physical system data when a new fault mode is detected;

[0033] S4. perform active fault-tolerant control according to the pre-diagnosis result of step S2, or switch to a preset fault-tolerant control algorithm when a fault occurs, and cooperatively decide the optimal stopping scheme with the elevator main control system;

[0034] S5. adopt an edge-cloud collaborative architecture, deploy high-precision twin modeling and complex training tasks in the cloud, deploy lightweight diagnosis and basic fault-tolerant logic in the local controller of the elevator, and realize data synchronization through a low-bandwidth protocol.

[0035] As a further scheme of the application: in step S1, the construction process of the electro-thermal-stress coupling model includes: based on the datasheet parameters and experimental data of IGBT or MOSFET, a loss calculation model is established to obtain switching loss and conduction loss; the loss data is converted into a junction temperature fluctuation curve by combining the thermal resistance network model of the heat dissipation structure; based on the fatigue accumulation theory, the number of thermal stress cycles is calculated according to the amplitude and frequency of the junction temperature fluctuation.

[0036] As a further scheme of the application: in step S2, the process of extracting early fault features includes: wavelet packet decomposition is performed on the deviation signal to obtain sub-signals of different frequency bands, and the energy entropy of each sub-signal is calculated as an initial feature; the initial feature is input into a 1D-CNN network, and deep features are extracted through convolution layers and pooling layers to obtain a feature vector that can represent early weak faults.

[0037] As a further scheme of the application: in step S4, the active fault-tolerant control includes: when it is predicted that the remaining service life of an IGBT of a certain phase is lower than a preset threshold, the elevator is controlled to enter a derated operation mode, the output torque is reduced to 70%-80% of the rated value, and the drive signal of the redundant phase is activated to prepare for topology reconstruction; the seamless fault-tolerant switching includes: when an IGBT short circuit fault is detected, a fault-tolerant modulation strategy based on neutral point offset is switched to within 5ms to compensate for the output loss of the fault phase and maintain the elevator operation to the nearest floor stop.

[0038] The technical effects and advantages of the application are as follows:

[0039] The beneficial effects of the application are as follows:

[0040] (1) Improve fault diagnosis capability

[0041] With the help of high-fidelity digital twin and real-time fault injection technology, the dependence on physical sensor data in traditional diagnosis is broken, and early weak fault characteristics can be identified to issue early warnings at the fault inception stage, avoiding situations such as people being trapped and elevator downtime due to sudden failures, and realizing the transition from passive response to active prevention.

[0042] (2) Accurately grasp the device state and optimize the maintenance mode

[0043] Combined with device-level aging models and multi-physical field coupling analysis, not only can the fault type be identified, but also the remaining use state of key power devices such as IGBT can be accurately judged, so that maintenance work can be changed from regular blind maintenance to accurate maintenance based on the actual state of the device, reducing unnecessary downtime and spare parts consumption.

[0044] (3) Ensure long-term accuracy of the twin and eliminate model bias

[0045] The online adaptive updating module of the twin continuously corrects the model bias caused by device aging and environmental changes through closed-loop parameter calibration and incremental learning, ensuring that the digital twin maintains high fidelity throughout the life cycle of the elevator, providing stable and reliable virtual mapping support for fault diagnosis and fault-tolerant control.

[0046] (4) Improve the safety of the fault state and ensure the elevator experience

[0047] Active fault-tolerant measures based on the diagnosis results (such as reduced capacity operation and topology reconstruction) and millisecond seamless switching when a fault occurs, in cooperation with the elevator control system, can ensure that the elevator can still run smoothly to the nearest safe floor in a fault state, avoiding dangerous situations such as sudden stops and elevator slides, and improving the safety and comfort of the elevator.

[0048] (5) Balance technical performance and practical value, and facilitate engineering implementation

[0049] The hierarchical deployment strategy of edge-cloud cooperation allocates complex computing tasks to the cloud and deploys lightweight logic on the local controller, combined with low-bandwidth communication protocols, to reduce the computing power requirements of the local hardware of the elevator while ensuring real-time response, and can adapt to new and old elevator scenarios, solving the problem of high precision and practicality in the industrial application of digital twin technology, and facilitating large-scale promotion.

[0050] (6) Break through the limitations of traditional simulation and improve virtual mapping capabilities

[0051] Compared with traditional simulation that only focuses on circuit characteristics, the invention integrates multi-dimensional factors such as electro-thermal-stress coupling, parasitic parameters, and drive protection logic, so that the twin can reproduce the fault transient process and aging cumulative effect, improving the simulation capability of the real operating state of the inverter, and providing a more realistic virtual platform for fault mechanism analysis and control strategy verification. BRIEF DESCRIPTION OF DRAWINGS

[0052] The application will be further described below with reference to the drawings.

[0053] Figure 1 is the system block diagram of the elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twinning of the application;

[0054] Figure 2 is the flow chart of the elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twinning of the application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0056] Please refer to Figure 1 The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twinning, as shown in the figure, comprises:

[0057] The high-fidelity digital twinning construction module is used to establish the digital twinning of the elevator inverter, and the digital twinning integrates an electro-thermal-stress coupling model, a device-level aging model, a parasitic parameter influence model, and a driving circuit and protection logic model. The electro-thermal-stress coupling model is used to simulate the loss, junction temperature fluctuation and thermal stress cycle of the power device, the device-level aging model is used to represent the time-varying degradation law of the device parameter, the parasitic parameter influence model is used to reflect the influence of PCB layout and stray parameters on the switching characteristics of the inverter, and the driving circuit and protection logic model is used to reproduce the behavior of the driving chip and the protection logic.

[0058] In actual application, the high-fidelity digital twinning construction module constructs the digital twinning of the elevator inverter through a multi-physical field modeling tool (such as ANSYS, PLECS):

[0059] The electro-thermal-stress coupling model: based on the IGBT data table parameters (such as switching time, on-state voltage drop) to calculate the switching loss and on-state loss, combined with the three-dimensional thermal resistance network model of the heat sink (including thermal grease, substrate thermal resistance) to convert the loss into junction temperature fluctuation, and through the Coffin-Manson formula to calculate the influence of thermal stress cycle on the device life;

[0060] Device-level aging model: Obtain the degradation curves of IGBT's on-state resistance Rds(on) and threshold voltage Vth with cycle number through accelerated aging experiments, and embed the curves fitted into mathematical expressions (such as exponential decay model) into the twin body;

[0061] Parasitic parameter influence model: Extract the stray inductance (such as bus inductance, gate inductance) and stray capacitance (such as inter-device distributed capacitance) of PCB layout through AltiumDesigner, and add them to the twin body as circuit elements to simulate voltage overshoot (such as voltage spike during IGBT turn-off) during switching process;

[0062] Drive circuit and protection logic model: Based on the data manual of the drive chip (such as 2SC0435T), reproduce the threshold voltage, response time of desaturation detection (DESAT), and the clamping effect of the Miller clamp circuit, and implant the trip threshold (such as 2 times of the rated current) and delay logic of the overcurrent protection;

[0063] Real-time fault injection and predictive diagnosis module, which is connected with high-fidelity digital twin body construction module and physical system sensors, is used to inject preset fault modes on the digital twin body, compare the ideal state data output by the digital twin body with the actual sensor data of the physical system to obtain the deviation signal, extract early fault features based on the deviation signal, and identify fault types and predict the remaining useful life of key power devices in combination with device-level aging model;

[0064] In actual application, real-time fault injection and predictive diagnosis module communicates with twin body module and physical system sensors (current sensor, temperature sensor, voltage sensor) through Ethernet:

[0065] The preset fault mode library contains 20 typical faults such as single tube open circuit and capacitor value attenuation (such as 20% reduction of electrolytic capacitor capacity);

[0066] Compare the ideal current waveform output by the twin body with the actual current waveform collected by the Hall sensor to calculate the deviation signal (such as instantaneous difference, effective value deviation);

[0067] Twin body online adaptive updating module, which is connected with high-fidelity digital twin body construction module and physical system sensors respectively, is used to automatically calibrate the model parameters of the digital twin body according to the deviation between the output data of the digital twin body and the actual sensor data of the physical system, and update the diagnosis model based on the physical system data when new fault mode or new operating condition is detected;

[0068] In actual application, twin body online adaptive updating module communicates with sensors through PLC, and parameter calibration is triggered every 10 minutes:

[0069] If the IGBT junction temperature predicted by the twin (calculated based on the loss model) deviates from the measured value by more than 5℃, automatically adjust the thermal resistance parameter (such as Rth(j-c));

[0070] When a new fault is detected (such as current sensor zero drift exceeding 0.5% FS), update the classifier parameters of the diagnostic model through transfer learning;

[0071] Intelligent fault-tolerant control module, in communication connection with real-time fault injection and predictive diagnosis module and elevator main control system, for executing active fault-tolerant control according to fault pre-diagnosis result, or switching to preset fault-tolerant control algorithm to realize seamless fault-tolerant when fault occurs, and cooperating with elevator main control system to decide optimal stop scheme;

[0072] In actual application, intelligent fault-tolerant control module communicates with elevator main control system (such as Mitsubishi HOPE-II main control board) through CAN bus:

[0073] Active fault-tolerant: when predicting that the remaining life of certain phase IGBT is less than 10%, send derating instruction to main control system to limit output power to 80% of rated value;

[0074] Seamless switching: when detecting IGBT short circuit, switch to preset three-phase four-leg fault-tolerant topology within 5ms to compensate fault phase through redundant bridge arm;

[0075] Edge-cloud collaborative deployment module, in communication connection with each of the foregoing modules, for deploying high-precision twin modeling, complex model training and long-term trend analysis tasks to cloud or edge server, deploying lightweight diagnostic model, real-time feature extraction and basic fault-tolerant control logic to elevator local controller, and realizing data synchronization and model updating between edge and cloud through low-bandwidth communication protocol;

[0076] In actual application, edge-cloud collaborative deployment module:

[0077] Cloud (Aliyun ECS server) deploys twin reconstruction, long-term trend analysis (such as monthly aging report) tasks;

[0078] Elevator local controller (STM32H743) deploys lightweight 1D-CNN model (input is 128-point current sequence, output is fault probability);

[0079] Adopt MQTT protocol to transmit data, only upload fault features (such as current distortion coefficient) and diagnostic results, and data volume is less than 10MB per hour.

[0080] As a further aspect of the present application: the electro-thermal-stress coupling model is particularly used for: calculating the switching loss and conduction loss of IGBT or MOSFET, simulating the junction temperature fluctuation combined with the heat dissipation condition, and calculating the thermal stress cycle based on the junction temperature change amount; in the device-level aging model, the device parameters include the on-resistance Rds(on), the threshold voltage Vth and the thermal resistance Rth(j-c), and the time-varying degradation law is obtained by fitting the parameter degradation curve obtained by experiment; the parasitic parameter influence model is particularly used for: simulating the influence of the stray inductance and stray capacitance introduced by the PCB layout on the voltage overshoot, current oscillation and electromagnetic interference EMI in the switching process; the driving circuit and protection logic model is particularly used for: reproducing the desaturation detection DESAT, the Miller clamp behavior of the driving chip, and the triggering and response process of the overcurrent, overvoltage and overheat protection logic;

[0081] In actual application, the electro-thermal-stress coupling model:

[0082] The switching loss calculation adopts the energy method: the turn-on energy Eon and the turn-off energy Eoff of the IGBT at a switching frequency of 10 kHz are simulated by twin simulation, and the instantaneous switching loss (Psw=(Eon+Eoff)×fsw) is calculated combined with the current and voltage values; the conduction loss is calculated by the product of the conduction voltage drop Vce(sat) and the conduction current Ic (Pcond=Vce(sat)×Ic×D, D is the duty cycle).

[0083] The heat dissipation condition simulation: the wind speed (0.5 m / s) and the environmental temperature (30℃) of the heat dissipation fin are input into the thermal resistance network to calculate the junction temperature Tj=Ptotal×(Rth(j-c)+Rth(c-s)+Rth(s-a))+Ta, wherein Ptotal is the total loss.

[0084] The thermal stress cycle calculation: based on the junction temperature fluctuation ΔTj (such as from 60℃ to 120℃), the fatigue damage is accumulated by the Miner rule (D=Σni / Ni, ni is the current cycle number, and Ni is the life cycle number corresponding to ΔTj).

[0085] The device-level aging model:

[0086] An accelerated aging experiment (junction temperature fluctuation ΔTj=100℃, cycle frequency 0.1 Hz) is performed on 10 IGBTs of the same type, and the parameters are measured every 1000 cycles:

[0087] Rds(on) degrades from the initial 20mΩ to 30mΩ (10 million cycles), which is fitted as Rds(on)=20+0.0001×N (N is the cycle number);

[0088] The thermal resistance Rth(j-c) rises from 0.5 K / W to 0.7 K / W, and is fitted as Rth(j-c) = 0.5 + 2e-9 x N 2 .

[0089] Parasitic parameter impact model:

[0090] The stray inductance of the PCB layout (such as the DC bus inductance Lbus = 200 nH) is measured by an impedance analyzer, and is connected to the bus loop of the twin body to simulate the voltage overshoot (Vovershoot = Lbus x di / dt, di / dt is the off current change rate) when the IGBT is turned off;

[0091] The stray capacitance (such as the IGBT collector-emitter capacitance Cce = 100 pF) is used to simulate high-frequency oscillation (oscillation frequency f = 1 / (2π√(Lbus x Cce))).

[0092] Drive circuit and protection logic model:

[0093] Desaturation detection: when the IGBT is turned on, if the collector voltage exceeds 7V (DESAT threshold) and lasts for 1μs, the protection signal is triggered to turn off the IGBT;

[0094] Overcurrent protection: when the phase current is detected to exceed 300A (rated current 150A) and lasts for 50μs, the soft turn-off logic of the drive chip is triggered (turn-off time 2μs).

[0095] As a further scheme of the application: the real-time fault injection and predictive diagnosis module comprises:

[0096] A virtual fault injection unit is configured to inject potential fault modes on the digital twin in parallel, the potential fault modes including single tube open circuit, single tube short circuit, abnormal drive signal, capacitance value attenuation, current sensor drift, current sensor failure, and heat dissipation capacity deterioration;

[0097] A deviation feature extraction unit is configured to process the deviation signal by wavelet packet decomposition or empirical mode decomposition (EMD) to obtain multi-band components, and extract early weak fault features in the multi-band components by 1D-CNN, LSTM or Transformer network, the early weak fault features including current waveform slight distortion, switch timing slight shift, and temperature rise trend anomaly;

[0098] A fault diagnosis and life prediction unit is configured to compare the early weak fault features with a preset fault feature library to identify the fault type, and predict the remaining service life of the IGBT or MOSFET based on the mapping relationship between the parameter degradation law in the device-level aging model and the fault features;

[0099] In actual application, the virtual fault injection unit:

[0100] Inject faults in parallel in the twins via Python scripts:

[0101] Single tube open circuit: Set the IGBT on-resistance to 1MΩ;

[0102] Abnormal drive signal: reduce the gate voltage from 15V to 12V (resulting in an increase of 50ns in turn-on delay);

[0103] Deterioration of heat dissipation capacity: Increase the heat sink thermal resistance Rth(sa) from 10K / W to 20K / W (simulating the effect of dust accumulation).

[0104] Deviation feature extraction unit:

[0105] Perform wavelet packet decomposition on the current deviation signal (the difference between the ideal current and the actual current of the twin):

[0106] The number of decomposition layers is 3, and 8 frequency band components are obtained (0-5kHz, 5-10kHz...35-40kHz);

[0107] The energy proportion of each component is calculated as the initial feature and input into a 1D-CNN network (including 2 convolutional layers: 32 3×1 convolution kernels, stride 1; pooling layer: 2×1 maximum pooling; fully connected layer: 64 neurons), which outputs 16-dimensional deep features to characterize slight distortions of the current waveform (such as 1% peak deviation).

[0108] Fault diagnosis and life prediction unit:

[0109] Fault identification: 16-dimensional features are input into the SVM classifier, compared with the preset feature library (containing feature templates of 20 types of faults), and the fault type is output (such as "IGBT1 open circuit" with an accuracy rate of ≥95%).

[0110] Lifetime prediction: Based on the temperature rise trend in the characteristics (such as the junction temperature rise rate increasing from 0.5°C / min to 1°C / min), combined with the Rth(jc) degradation curve of the device-level aging model, the remaining life is calculated (for example, if the current Rth(jc) = 0.6K / W, it is predicted that it will take another 3000 hours to reach the failure threshold of 0.8K / W).

[0111] As a further solution of the present invention: the twin online adaptive update module includes:

[0112] A closed-loop parameter calibration unit, which continuously collects estimated voltage, current, and temperature values ​​output by the digital twin, compares them with the actual voltage, current, and temperature collected by the physical system sensors, calculates the deviation, and automatically adjusts the loss calculation parameters, thermal resistance parameters, and parasitic parameters in the digital twin based on the deviation.

[0113] An incremental learning unit is configured to, when a new fault mode not included in the preset fault mode library is detected, update the diagnostic model parameters in the real-time fault injection and predictive diagnosis module based on the operation data of the new fault mode collected by the physical system, and expand the fault recognition range of the model through a transfer learning algorithm.

[0114] In actual application, the closed-loop parameter calibration unit:

[0115] Data is collected every 5 minutes: DC bus voltage estimation value Udc_sim output by the twin, actual sensor measurement value Udc_meas;

[0116] The deviation amount ΔUdc = |Udc_sim-Udc_meas| is calculated, and if ΔUdc>2V (exceeding the allowable error 1%), the bus parasitic resistance in the twin is adjusted (from 0.1Ω to 0.15Ω);

[0117] Temperature calibration: if the junction temperature estimation value Tj_sim and the infrared temperature Tj_meas deviate by >5℃, the thermal resistance parameter is adjusted in proportion (Rth(j-c)=Rth(j-c)×(Tj_sim-Ta) / (Tj_meas-Ta), Ta is the ambient temperature).

[0118] The incremental learning unit:

[0119] When a new fault (such as a 5% linearity deviation of the current sensor, which is not in the preset library) is detected, data collection is triggered: the current, voltage, and temperature data 10 minutes before the fault occurs are recorded (sampling rate 10kHz);

[0120] A transfer learning algorithm (such as Fine-tuning) is used: the first two layers of weights of the 1D-CNN are frozen, only the parameters of the fully connected layer are updated, and the model is trained with new fault data (500 samples) to add the recognition ability of the model to the fault (the recognition accuracy is improved from 0 to more than 90%).

[0121] As a further scheme of the application, the intelligent fault-tolerant control module includes:

[0122] The active fault-tolerant control unit is configured to, when the real-time fault injection and predictive diagnosis module predicts that a key power device is about to fail, perform at least one of the following operations: reducing the output power of the elevator to reduce the load of the device, switching to a redundant phase topology to reconstruct the inverter output, and sending accurate early warning information to the maintenance system;

[0123] The seamless fault-tolerant switching unit is configured to, when a short circuit or open circuit fault of the inverter is detected, switch to a preset fault-tolerant control algorithm within milliseconds, and the preset fault-tolerant control algorithm includes a torque compensation algorithm based on specific harmonic injection and a three-phase imbalance suppression algorithm based on reference vector reconstruction.

[0124] A cooperative control unit is used to interact with the elevator main control system to obtain the current position, running direction and safety circuit state of the elevator, ensure the effectiveness of safety logic such as elevator overspeed protection and door lock monitoring during fault-tolerant operation, and decide on the nearest safe floor to achieve smooth stopping based on the above information;

[0125] In actual application, the active fault-tolerant control unit:

[0126] When it is predicted that the remaining life of IGBT2 is less than 500 hours, an instruction is sent to the elevator main control:

[0127] Output power limitation: reduce the maximum torque from 300 N·m to 240 N·m;

[0128] Redundant phase switching: activate the standby IGBT bridge arm (usually in standby state), and switch to four-phase topology through contactor;

[0129] Maintenance warning: send information (including fault location "IGBT2" and recommended replacement time "within 7 days") to the maintenance platform through the 4G module.

[0130] Seamless fault-tolerant switching unit:

[0131] When A-phase IGBT is detected to be open (A-phase current is detected to be 0 by current sensor for 200 μs), switch to specific harmonic injection algorithm within 3 ms:

[0132] Inject 3rd harmonic (amplitude is 1 / 6 of fundamental wave) in reference voltage vector to compensate torque ripple caused by three-phase imbalance (from ±5% to ±1%);

[0133] When short-circuit fault occurs: after triggering DESAT protection, switch to reference vector reconstruction algorithm within 1 ms to project the voltage vector of the fault phase to the remaining two phases to maintain balanced output voltage.

[0134] Cooperative control unit:

[0135] Read the position signal (such as current floor "5th floor" and running direction "up") and safety circuit state (door lock closed and speed limiter normal) of the elevator main control through CAN bus;

[0136] Decide on the nearest safe floor: calculate the running time to the 6th floor (10 m away) and the 4th floor (8 m away), and select the 4th floor as the stopping target;

[0137] During fault-tolerant operation, real-time feedback output torque to the main control (to ensure that it does not exceed the speed limiter threshold), and monitor the door lock signal (if disconnected, immediately trigger emergency braking).

[0138] As a further scheme of the present application: the low-bandwidth communication protocol is specifically used for transmitting only key feature data, fault diagnosis results, model calibration parameters and update instructions between the edge and the cloud, and the key feature data includes a current distortion coefficient, a temperature fluctuation amplitude and a switching time deviation amount;

[0139] In actual application, the low-bandwidth communication protocol adopts a self-defined message format based on MQTT:

[0140] Key feature data:

[0141] Current distortion coefficient (THD): 2 decimal places are reserved (such as 3.25%), and 1 byte is used for transmission;

[0142] Temperature fluctuation amplitude: maximum-minimum value of IGBT junction temperature (such as "60-120℃"), and 2 bytes are used for transmission;

[0143] Switching time deviation: difference between actual turn-on time and ideal value (such as "+50ns"), and 1 byte is used for transmission.

[0144] Fault diagnosis result: encoded as 2 bytes (high 8 bits represent fault type, such as 0x03 representing IGBT open circuit; low 8 bits represent confidence, such as 0x9A representing 97%).

[0145] Model calibration parameter: adjustment value of thermal resistance Rth(j-c) (such as "+0.05K / W"), and 2 bytes are used for transmission.

[0146] Data transmission frequency: once every 30 seconds in normal operation, real-time transmission in fault, single message length <10 bytes, and daily data volume <5MB (adapted to the weak network environment of the elevator shaft).

[0147] Further, the elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twinning includes the following steps:

[0148] S1. Constructing a high-fidelity digital twin of the elevator inverter, the digital twin fuses an electro-thermal-stress coupling model, a device-level aging model, a parasitic parameter influence model and a driving circuit and protection logic model;

[0149] S2. Injecting a preset fault mode into the digital twin to obtain ideal state data simulated by the digital twin, comparing actual sensing data of the physical system to obtain a deviation signal, extracting early fault features based on the deviation signal, identifying the fault type and predicting the remaining service life of the key power device;

[0150] S3. Automatically calibrating model parameters of the digital twin according to the deviation between output data of the digital twin and actual sensing data of the physical system, and updating the diagnosis model based on physical system data when a new fault mode is detected;

[0151] S4. Perform active fault-tolerant control according to the pre-diagnosis result of step S2, or switch to the preset fault-tolerant control algorithm when a fault occurs, and make a decision on the optimal stopping scheme in cooperation with the elevator main control system;

[0152] S5. Adopt an edge-cloud collaborative architecture, deploy high-precision twin modeling and complex training tasks in the cloud, and deploy lightweight diagnosis and basic fault-tolerant logic on the local controller of the elevator, and realize data synchronization through a low-bandwidth protocol;

[0153] In actual application, step S1:

[0154] A "multi-domain model fusion" method is used to construct the digital twin:

[0155] Circuit domain: Build the main circuit of the inverter (including IGBT, freewheeling diode, DC bus capacitor) in PLECS, input rated parameters (such as DC voltage 380V, switching frequency 10kHz);

[0156] Thermal domain: Establish a three-dimensional heat dissipation model of the IGBT module (including fin size 100×80×20mm, fin pitch 5mm) in ANSYS Icepak, and input the loss data calculated by the circuit domain as the heat source;

[0157] Aging domain: Compile the parameter degradation formula obtained from the accelerated aging experiment (such as Vth=5-0.0001×N) into C code, and embed it into the twin through the S function interface of MATLAB / Simulink;

[0158] Drive and protection domain: Build a behavioral model of the drive chip (such as IR2110) in Simulink, and reproduce the hardware logic of DESAT detection (including comparator threshold, delay circuit).

[0159] Step S2:

[0160] Fault injection: Inject a "30% attenuation of capacitor capacitance" fault through the API interface of the twin, and output an ideal three-phase current waveform (amplitude 50A, THD=1% of sinusoidal degree);

[0161] Deviation calculation: Collect the Hall sensor current data of the physical system (sampling rate 20kHz), and calculate the instantaneous deviation from the ideal waveform (such as deviation 0.5A at t=10ms);

[0162] Feature extraction: Perform EMD decomposition on the deviation signal to obtain 5 intrinsic mode functions (IMF), and calculate the energy entropy of each IMF as a feature;

[0163] Diagnosis and prediction: Input the features into the LSTM network, and output the fault type "capacitor aging" and the remaining life "1200 hours".

[0164] Step S3:

[0165] Parameter calibration: Start calibration at 2 a.m. (low load period of elevator), if the estimated value of twin voltage deviates from the actual value by >2%, adjust the bus parasitic resistance (from 0.05Ω to 0.06Ω);

[0166] New fault learning: When detecting "drive optocoupler delay increase" (not in the preset library), collect 100 sets of fault data, update the classification layer parameters of the 1D-CNN model through incremental learning, and realize the identification of the fault.

[0167] Step S4:

[0168] Active fault tolerance: If step S2 predicts "B-phase IGBT remaining life <500 hours", control the elevator to enter "light load mode" (load limit 80%), and notify the maintenance at the same time;

[0169] Seamless switching: If a sudden A-phase IGBT short circuit occurs, trigger the hardware protection within 0.5ms, switch to the fault-tolerant modulation strategy (such as based on zero sequence voltage injection) within 2ms, and cooperate with the main control to select the nearest 3rd floor stop (currently at the 5th floor and descending).

[0170] Step S5:

[0171] Cloud deployment: Aliyun server runs twin reconstruction program (model is updated every morning), trains LSTM life prediction model (once a week);

[0172] Edge deployment: Local PLC (such as Siemens S7-1200) of the elevator runs lightweight diagnosis model (inference time <10ms);

[0173] Communication: Use LoRa protocol to transmit data, only send fault codes (such as "F03" represents IGBT open circuit) and characteristic values (such as temperature fluctuation amplitude "5℃"), every 30 seconds, single data volume <50 bytes.

[0174] As a further scheme of the application: in step S1, the construction process of the electro-thermal-stress coupling model includes: based on the datasheet parameters and experimental data of IGBT or MOSFET, a loss calculation model is established to obtain switching loss and conduction loss; combining the thermal resistance network model of the heat dissipation structure, the loss data is converted into a junction temperature fluctuation curve; based on the fatigue accumulation theory, the thermal stress cycle number is calculated according to the junction temperature fluctuation amplitude and frequency;

[0175] In practical application,

[0176] Loss calculation model construction:

[0177] Switching loss: According to the Eon-Ic, Eoff-Vce curve in the IGBT datasheet (such as Infineon FF450R12ME4), the switching energy calculation formula is obtained by interpolation fitting: Eon = 0.002 * Ic + 0.0001 * Vce (Ic unit A, Vce unit V), and the switching loss Psw = (Eon + Eoff) * fsw (fsw = 10 kHz);

[0178] Conduction loss: Based on the conduction voltage drop Vce(sat) = 1.2 + 0.001 * Ic (measured data fitting), the Pcond = Vce(sat) * Ic * D (D is the duty cycle, take 0.5) is calculated.

[0179] Thermal resistance network model construction:

[0180] Establish a three-level thermal resistance network: junction to shell Rth(j-c) = 0.5 K / W (datasheet value), shell to heat sink Rth(c-s) = 0.2 K / W (thermal grease), heat sink to environment Rth(s-a) = 10 K / W (measured);

[0181] Junction temperature calculation: Tj = (Psw + Pcond) * (Rth(j-c) + Rth(c-s) + Rth(s-a)) + Ta (Ta = 30℃), the junction temperature fluctuation curve (such as from 60℃ to 110℃, period 10 seconds) is obtained.

[0182] Thermal stress cycle calculation:

[0183] Based on the Coffin-Manson model: Nf = A * (ΔTj)^B (A = 1e10, B = -2.5, experimental fitting);

[0184] Calculate the number of daily thermal stress cycles: the elevator starts and stops 10 times per hour, each time causing ΔTj = 50℃, 240 times per day, and the cumulative damage D = 240 / Nf.

[0185] As a further scheme of the application: in step S2, the process of extracting early fault features includes: wavelet packet decomposition of the deviation signal to obtain sub-signals of different frequency bands, and calculation of the energy entropy of each sub-signal as the initial feature; input the initial feature into the 1D-CNN network, extract the deep layer feature through the convolution layer and the pooling layer, and obtain the feature vector capable of representing the early weak fault;

[0186] In practical application, wavelet packet decomposition:

[0187] The deviation signal is a current difference sequence (length 1024 points, sampling rate 10 kHz), and db4 wavelet basis is used for 3-layer decomposition to obtain 8 sub-signals (frequency band 0-1.25kHz to 8.75-10kHz);

[0188] Calculate the energy entropy of each sub-signal: H = -∑(pi x log2(pi)), where pi is the energy proportion of the ith sub-signal (such as the 3rd sub-signal energy proportion 0.2, H = 0.72), to obtain an 8-dimensional initial feature.

[0189] 1D-CNN feature extraction:

[0190] Network structure: input layer (8-dimensional feature) → convolutional layer 1 (16 3x1 convolutional kernels, ReLU activation) → pooling layer 1 (2x1 max pooling) → convolutional layer 2 (32 3x1 convolutional kernels) → pooling layer 2 (2x1 max pooling) → fully connected layer (16 neurons);

[0191] Output: 16-dimensional deep feature vector (such as [0.12, 0.35,..., 0.08]), where the 5th feature has a sensitivity of 90% to "IGBT slight aging" (5% increase in on-resistance).

[0192] As a further scheme of the application: in step S4, the active fault-tolerant control includes: when it is predicted that the remaining service life of a certain phase IGBT is lower than a preset threshold, controlling the elevator to enter a derated operation mode, reducing the output torque to 70%-80% of the rated value, and activating the drive signal of the redundant phase to prepare for topology reconstruction; the seamless fault-tolerant switching includes: when an IGBT short-circuit fault is detected, switching to a neutral point offset-based fault-tolerant modulation strategy within 5ms, compensating for the output loss of the fault phase, and maintaining the elevator operation to the nearest floor stop;

[0193] In actual application, the active fault-tolerant control:

[0194] The preset threshold is set to "remaining life < 800 hours", and when it is predicted that the life of the C-phase IGBT is 750 hours:

[0195] Derated operation: by modifying the modulation ratio of SVPWM (from 0.9 to 0.72), the output torque is reduced from 250 N·m to 200 N·m (80% of the rated value);

[0196] Redundant phase activation: send an enable signal to the drive chip of the standby IGBT (raise the gate voltage from 0V to 15V), close the redundant phase contactor (response time < 100ms), and wait for the topology switching instruction.

[0197] Seamless fault-tolerant switching:

[0198] Short-circuit detection: when the DESAT pin voltage > 7V (for 1μs), it is determined that the IGBT is short-circuited, and the gate drive is immediately turned off (within 0.5ms);

[0199] Fault tolerant modulation: switch to neutral point shift strategy within 3ms, distribute the reference voltage of the fault phase (e.g. phase A) to phases B and C (shifted by 1 / 2 of the fault phase voltage), compensate for the unbalance of the output voltage;

[0200] Docking control: communicate with the elevator master, get the current position (7th floor) and speed (1.5m / s), calculate the deceleration curve (acceleration -0.5m / s 2 ), ensure smooth docking to the 6th floor (the nearest safe floor) within 5 seconds, and disconnect the main contactor after docking.

[0201] It should be declared that all the data collected in this application are collected with the consent and authorization of the user, and the use of the data is legal and compliant, and the use and processing of the data comply with the relevant laws, regulations and standards of the relevant region. The above formulas are dimensionless to calculate the numerical value, the formula is obtained by software simulation of a large number of data to obtain the formula of the nearest real situation, and the preset parameters and threshold in the formula are set by the person skilled in the art according to the actual situation.

[0202] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. Elevator inverter fault prediction and fault-tolerant control system based on digital twin, characterized by: include: A high-fidelity digital twin construction module for building a digital twin of an elevator inverter. The digital twin integrates an electro-thermal-stress coupling model, a device-level aging model, a parasitic parameter impact model, and a drive circuit and protection logic model. The electro-thermal-stress coupling model simulates power device losses, junction temperature fluctuations, and thermal stress cycles. The device-level aging model characterizes the time-varying degradation of device parameters. The parasitic parameter impact model reflects the impact of PCB layout and stray parameters on inverter switching characteristics. The drive circuit and protection logic model replicates driver chip behavior and protection logic. A real-time fault injection and predictive diagnosis module, in communication with the high-fidelity digital twin building module and the physical system sensors, is configured to inject a preset fault mode into the digital twin, compare the ideal state data output by the digital twin with the actual sensor data of the physical system to obtain a deviation signal, extract early fault characteristics based on the deviation signal, and identify the fault type and predict the remaining useful life of key power devices in combination with the device-level aging model; A twin online adaptive update module, which is in communication with the high-fidelity digital twin construction module and the physical system sensors, and is used to automatically calibrate the model parameters of the digital twin based on the deviation between the output data of the digital twin and the actual sensor data of the physical system, and update the diagnostic model based on the physical system data when a new fault mode or new operating condition is detected; an intelligent fault-tolerant control module, in communication with the real-time fault injection and predictive diagnosis module and the elevator master control system, configured to perform active fault-tolerant control based on fault pre-diagnosis results, or switch to a preset fault-tolerant control algorithm to achieve seamless fault tolerance when a fault occurs, and collaborate with the elevator master control system to determine an optimal parking plan; The edge-cloud collaborative deployment module communicates with each of the aforementioned modules and is used to deploy high-precision twin modeling, complex model training, and long-term trend analysis tasks on the cloud or edge server, deploy lightweight diagnostic models, real-time feature extraction, and basic fault-tolerant control logic on the elevator local controller, and achieve data synchronization and model updates between the edge and cloud through a low-bandwidth communication protocol.

2. The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twin according to claim 1 is characterized in that: The electro-thermal-stress coupling model is specifically used to calculate the switching loss and conduction loss of the IGBT or MOSFET, simulate junction temperature fluctuations in combination with heat dissipation conditions, and calculate thermal stress cycles based on the junction temperature change; in the device-level aging model, the device parameters include on-resistance Rds(on), threshold voltage Vth and thermal resistance Rth(jc), and the time-varying degradation law is obtained by fitting the parameter degradation curve obtained through experiments; the parasitic parameter influence model is specifically used to simulate the influence of stray inductance and stray capacitance introduced by the PCB layout on voltage overshoot, current oscillation and electromagnetic interference EMI during the switching process; the drive circuit and protection logic model is specifically used to reproduce the desaturation detection DESAT and Miller clamping behavior of the drive chip, as well as the triggering and response process of the overcurrent, overvoltage and overheating protection logic.

3. The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twin according to claim 1 is characterized in that: The real-time fault injection and predictive diagnosis module includes: A virtual fault injection unit, configured to inject potential fault modes in parallel on the digital twin, the potential fault modes including single transistor open circuit, single transistor short circuit, abnormal driving signal, capacitor capacitance attenuation, current sensor drift, current sensor failure, and deterioration of heat dissipation capability; a deviation feature extraction unit, configured to process the deviation signal using wavelet packet decomposition or empirical mode decomposition (EMD) to obtain multi-band components, and extract early weak fault features from the multi-band components using a 1D-CNN, LSTM, or Transformer network. The early weak fault features include slight distortion of the current waveform, slight shift in the switching timing, and abnormal temperature rise trends; A fault diagnosis and life prediction unit is used to compare the early weak fault characteristics with a preset fault characteristic library to identify the fault type, and predict the remaining service life of the IGBT or MOSFET based on the mapping relationship between the parameter degradation law and the fault characteristics in the device-level aging model.

4. The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twin according to claim 1 is characterized in that: The twin online adaptive update module includes: A closed-loop parameter calibration unit, configured to continuously collect voltage, current, and temperature estimates output by the digital twin, compare them with the actual voltage, current, and temperature collected by the physical system sensors, calculate deviations, and automatically adjust loss calculation parameters, thermal resistance parameters, and parasitic parameters in the digital twin based on the deviations; The incremental learning unit is used to update the diagnostic model parameters in the real-time fault injection and predictive diagnosis module through a transfer learning algorithm based on the operating data under the new fault mode collected by the physical system when a new fault mode that is not included in the preset fault mode library is detected, so as to expand the fault recognition scope of the model.

5. The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twin according to claim 1 is characterized in that: The intelligent fault-tolerant control module includes: An active fault-tolerant control unit, configured to, when the real-time fault injection and predictive diagnostic module predicts that a key power device is about to fail, perform at least one of the following operations: reduce elevator output power to reduce device load, switch to a redundant phase topology to reconfigure inverter output, or send accurate early warning information to a maintenance system; A seamless fault-tolerant switching unit, configured to switch to a preset fault-tolerant control algorithm within milliseconds when a short-circuit or open-circuit fault is detected in the inverter. The preset fault-tolerant control algorithm includes a torque compensation algorithm based on specific harmonic injection and a three-phase imbalance suppression algorithm based on reference vector reconstruction; The collaborative control unit is used to interact with the elevator main control system to obtain the current position, running direction and safety circuit status of the elevator, ensure the effectiveness of safety logic such as elevator overspeed protection and door lock monitoring during fault-tolerant operation, and decide the nearest safe floor based on the above information to achieve smooth parking.

6. The elevator inverter fault pre-diagnosis and fault-tolerant control system based on digital twin according to claim 1 is characterized in that: The low-bandwidth communication protocol is specifically used to transmit only key feature data, fault diagnosis results, model calibration parameters, and update instructions between the edge and the cloud. The key feature data includes current distortion coefficient, temperature fluctuation amplitude, and switching time deviation.

7. Elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twin, characterized by: The following steps are involved: S1. Build a high-fidelity digital twin of the elevator inverter, integrating an electrical-thermal-stress coupling model, a device-level aging model, a parasitic parameter impact model, and a drive circuit and protection logic model. S2. Injecting a preset fault mode into the digital twin, obtaining ideal state data simulated by the digital twin, comparing it with the actual sensor data of the physical system to obtain a deviation signal, extracting early fault characteristics based on the deviation signal, identifying the fault type, and predicting the remaining useful life of key power devices; S3. Automatically calibrate the model parameters of the digital twin based on the deviation between the output data of the digital twin and the actual sensor data of the physical system, and update the diagnostic model based on the physical system data when a new failure mode is detected; S4. Execute active fault-tolerant control based on the pre-diagnosis results of step S2, or switch to a preset fault-tolerant control algorithm when a fault occurs, and collaborate with the elevator master control system to decide the optimal docking plan; S5. Using an edge-cloud collaborative architecture, high-precision twin modeling and complex training tasks are deployed in the cloud, lightweight diagnosis and basic fault-tolerant logic are deployed in the local elevator controller, and data synchronization is achieved through a low-bandwidth protocol.

8. The elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twin according to claim 7 is characterized in that: In step S1, the construction process of the electrical-thermal-stress coupling model includes: establishing a loss calculation model based on the datasheet parameters and experimental data of the IGBT or MOSFET to obtain the switching loss and conduction loss; combining the thermal resistance network model of the heat dissipation structure to convert the loss data into a junction temperature fluctuation curve; based on the fatigue accumulation theory, calculating the number of thermal stress cycles according to the junction temperature fluctuation amplitude and frequency.

9. The elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twin according to claim 7 is characterized in that: In step S2, the process of extracting early fault features includes: performing wavelet packet decomposition on the deviation signal to obtain sub-signals in different frequency bands, and calculating the energy entropy of each sub-signal as the initial feature; inputting the initial feature into the 1D-CNN network, extracting deep features through the convolution layer and the pooling layer, and obtaining a feature vector that can characterize early weak faults.

10. The elevator inverter fault pre-diagnosis and fault-tolerant control method based on digital twin according to claim 7, characterized in that: In step S4, the active fault-tolerant control includes: when it is predicted that the remaining service life of a phase IGBT is lower than a preset threshold, controlling the elevator to enter a derated operation mode, reducing the output torque to 70%-80% of the rated value, and activating the drive signal of the redundant phase to prepare for topology reconstruction; the seamless fault-tolerant switching includes: when an IGBT short-circuit fault is detected, switching to a fault-tolerant modulation strategy based on neutral point offset within 5ms, compensating for the output loss of the faulty phase, and maintaining the elevator operation to the nearest floor.

Images