A bridge machine fault diagnosis and root cause tracing method

Abstract

This application provides a method for fault diagnosis and root cause tracing of bridge erecting machines, including: optimizing the deployment of multi-physics sensors for key components of the bridge erecting machine, simultaneously collecting multiple related data, and constructing a multi-dimensional dataset by combining equipment structural parameters; constructing a 1:1 digital twin based on tools such as ANSYS and Fluent, integrating geometric models, multi-physics coupling models, and fault evolution physical models to achieve real-time data mapping between physical equipment and virtual models; mining the coupling correlation between multi-physics data, combining with a fault root cause knowledge base, constructing a "type identification-root cause tracing" linkage model, and outputting the fault type and root cause; reproducing the fault evolution process in the digital twin, simulating the repair effect of different maintenance schemes, and selecting the optimal scheme through multi-objective optimization; after the maintenance personnel execute the optimal scheme, they feed back the actual maintenance data and equipment operating status to the cloud, update the digital twin parameters and the fault root cause knowledge base, and continuously improve the diagnostic accuracy and maintenance effectiveness.

Description

Technical Field

[0001] This invention relates to the field of bridge erecting machine fault diagnosis, specifically to a method for fault diagnosis and root cause tracing of bridge erecting machines. Background Technology

[0002] As a core heavy equipment in bridge construction, bridge erecting machines have seen their fault diagnosis needs shift from simple fault type identification to precise root cause tracing. Traditional diagnostic methods, relying solely on single physical field monitoring data, fail to capture the correlation between faults caused by the coupling of multiple physical fields such as mechanics and thermodynamics. This results in low accuracy in root cause tracing, and numerous misdiagnoses lead to frequent recurrence of faults. This not only makes maintenance less targeted and significantly increases costs but also easily results in substantial downtime losses due to ineffective maintenance. Furthermore, traditional diagnostic solutions have many shortcomings. They fail to correlate root cause analysis with equipment structural parameters and operating conditions, lack a virtual-physical verification mechanism supported by physical models, and cannot reproduce the fault evolution process. Moreover, the fragmented use of multi-physical field data and the reliance on experience in developing repair plans easily lead to ineffective repairs, over-maintenance, or secondary damage, further increasing operation and maintenance costs and construction risks.

[0003] Existing technologies for fault diagnosis of bridge erecting machines still have core shortcomings. The relevant algorithms only focus on feature extraction and pattern matching at the data level, without incorporating physical models, and the root cause tracing cannot reach the essence of the fault. Digital twin technology is mostly used for construction posture control and path planning, but it does not build a special model for fault root cause tracing. It can only achieve physical state mapping, without the support of physical mechanisms for fault evolution and tracing logic. At the same time, the industry has not yet formed a closed-loop system of "diagnosis-tracing-verification-optimization". Diagnostic results are not verified by virtual simulation, repair solutions are not predicted for feasibility, and operation and maintenance data cannot be used to optimize the diagnosis and virtual model. As a result, the diagnostic accuracy gradually decreases with use. Moreover, existing technologies have not established a mapping relationship between multi-physics coupling characteristics and fault root causes, sensor deployment is not adapted to coupled fault monitoring needs, and the verification of repair effects still relies on physical trial and error. Many problems have not been effectively solved, and the industry urgently needs an integrated and accurate diagnostic solution. Summary of the Invention

[0004] The purpose of this invention is to solve the technical problems mentioned above, and to propose a method for fault diagnosis and root cause tracing of bridge erecting machines, comprising the following steps: S1. Collect multi-physics field operation data of key components of the bridge erecting machine, and simultaneously acquire structural parameters, historical fault data and historical operation and maintenance data of key components of the bridge erecting machine to build a basic dataset; S2. Based on the aforementioned basic dataset, construct a digital twin corresponding to the physical bridge erecting machine in a 1:1 ratio, and establish a real-time virtual-real data mapping between the physical bridge erecting machine and the digital twin; S3. Perform multi-field coupling correlation analysis on the multi-physics field operation data to extract coupling features related to fault evolution; S4. Based on the extracted coupling features, combined with the structural parameters and historical fault data, fault type identification and fault root cause location are completed simultaneously. S5. Input the fault root cause parameters and the current equipment operating status into the digital twin, and reproduce the fault evolution process through multi-physics field coupling simulation to verify the accuracy of the fault root cause location result; S6. Generate at least two sets of candidate maintenance schemes based on the verified root causes of the faults. Simulate the repair effect of each scheme through the digital twin and optimize to obtain the optimal maintenance scheme. After the maintenance is performed, collect the equipment trial operation feedback data. Update the digital twin, fault diagnosis model and fault root cause knowledge base based on the feedback data to form a closed-loop optimization of the entire process.

[0005] In the preferred embodiment, before collecting multiphysics operation data in step S1, the high-incidence fault locations and multiphysics coupling sensitive areas of the key components of the bridge erecting machine are analyzed by finite element simulation. The location of the measuring points is determined by combining fault mode influence analysis (FMEA). A multiphysics sensor group is deployed at each measuring point. The multiphysics sensor group includes at least a mechanical load sensor, an infrared temperature sensor, a current and voltage sensor, and a fluid flow velocity sensor.

[0006] In the preferred embodiment, in step S1, each sensor is triggered to synchronously collect data at a preset frequency through the edge computing gateway, and the timestamp synchronization of all collected data is achieved by using the NTP network time protocol, with a synchronization error ≤1ms; for the collected raw data, singular values ​​are removed by the 3σ criterion, environmental noise is filtered by the adaptive wavelet denoising algorithm, and then normalized to the [0,1] interval to form standardized multiphysics operation data.

[0007] In the preferred embodiment, in step S2, the constructed digital twin is a three-layer architecture of "geometry-physics-fault". Specifically, it involves: constructing a 1:1 geometric model based on the Unity3D platform that is completely consistent with the structural dimensions and assembly relationships of the key components of the physical bridge erecting machine; integrating mechanical models, heat conduction and fluid models, and electromagnetic models on the basis of the geometric model to construct a multi-physics coupled physical model; training a fault evolution model based on historical fault data to clarify the dynamic evolution law of faults caused by different fault root causes, and finally forming a three-layer integrated digital twin.

[0008] In the preferred embodiment, in step S2, the preprocessed multi-physics operation data at the edge is transmitted in real time to the digital twin via a 4G / 5G network, and the QoS 2 level MQTT message queue telemetry transmission protocol is used to ensure the reliability of data transmission. The physical model parameters of the digital twin are calibrated according to a preset period using the measured operation data of the physical bridge erecting machine to ensure that the state mapping error between the virtual model and the physical device is ≤3%.

[0009] In the preferred embodiment, in step S3, an improved coupling feature extraction algorithm that integrates mutual information entropy and finite element analysis results is used to calculate the mutual information entropy between different physical field parameters in the multi-physics field operation data, quantify the coupling correlation between multi-physics fields, and extract at least 12 core coupling features that are strongly correlated with fault evolution by combining the finite element simulation analysis results, and remove redundant feature information.

[0010] In the preferred embodiment, in step S4, the extracted coupling features are input into a lightweight convolutional neural network (CNN) model to complete fault type identification. The input layer dimension of the lightweight CNN model matches the coupling feature dimension, and the hidden layer is set to 2 layers. An association rule engine is constructed based on the fault root cause knowledge base. Combining the equipment structural parameters, coupling features, and fault type identification results, multi-dimensional evidence matching is performed. When the matching degree is ≥95%, the corresponding root cause of the fault and supporting evidence are output.

[0011] In the preferred embodiment, in step S6, the evaluation indicators for each candidate maintenance scheme in the simulation evaluation include at least the fault repair rate, equipment life extension cycle, maintenance cost and downtime. A multi-objective optimization algorithm is adopted to minimize maintenance cost, minimize downtime and maximize equipment remaining life as optimization objectives to select the optimal maintenance scheme, and push the operation steps, tool list, parameter standards and safety precautions of the optimal maintenance scheme to the operation and maintenance terminal.

[0012] In the preferred embodiment, step S6, the update based on feedback data, specifically involves: adjusting the physical model parameters of the digital twin based on the feedback data from the equipment trial operation, optimizing the accuracy of multi-physics coupling simulation; adding the type, root cause, maintenance plan, and repair effect of this fault to the fault root cause knowledge base, and using an incremental learning algorithm to update the parameters of the fault diagnosis model and the association rule engine to continuously improve the accuracy of fault diagnosis and root cause tracing.

[0013] In the preferred scheme, after completing a single fault diagnosis and maintenance loop, the system enters a normal operation state, collects multi-physics field operation data and equipment status of the bridge erecting machine in real time, and repeatedly executes virtual-real mapping calibration, coupling feature extraction, fault monitoring and diagnosis, virtual verification and closed-loop optimization steps to realize fault diagnosis and dynamic optimization of operation and maintenance throughout the entire life cycle of the bridge erecting machine.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Innovative integration of multi-physics field coupling analysis and digital twin technology breaks through the limitations of traditional methods that only identify fault phenomena, and achieves accurate positioning of the root cause of faults, thereby solving the industry problems of misjudgment of faults and treating symptoms but not the root cause from the source.

[0015] (2) Construct a digital twin for fault diagnosis of dedicated bridge erecting machine, integrate the physical model of fault evolution, and virtually reproduce the fault development process to verify the effectiveness of the diagnosis results, so that the fault diagnosis has the support of a physical model and improves the reliability of the diagnosis.

[0016] (3) Establish a virtual simulation and optimization mechanism for maintenance plans to replace the traditional experience-based maintenance method. This can predict the implementation effect of different maintenance plans in advance, avoid secondary damage to equipment caused by blind maintenance, and improve the scientific nature of maintenance decisions.

[0017] (4) Create a closed-loop technology system for the entire process of “diagnosis-traceability-verification-repair-optimization”. The operation and maintenance data can be used to iteratively optimize the digital twin, fault diagnosis model and knowledge base, so that the diagnosis system has the ability to self-evolve and adapt to complex and ever-changing operating conditions and new fault types.

[0018] (5) To realize the systematic acquisition of multi-physics field data and in-depth mining of coupling characteristics, change the status quo of isolated analysis of traditional single physical field data, capture the essential characteristics of fault occurrence, and greatly improve the diagnostic effectiveness of complex coupled faults.

[0019] (6) The technical solution of the present invention does not require modification of the existing bridge erecting machine hardware structure. It can be implemented by simply expanding the deployment of sensors and digital twin modeling. It has strong compatibility and can be quickly adapted to various bridge erecting machine models, reducing the technical upgrade and application costs of existing equipment.

[0020] (7) Establish a real-time data link between the physical equipment of the bridge erecting machine and the virtual model to achieve accurate mapping and dynamic calibration of the virtual and real states, so that fault diagnosis and maintenance simulation are based on the actual operating state of the equipment, thereby improving the accuracy of the entire process of operation and maintenance. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical process of the present invention.

[0022] Figure 2 This is a schematic diagram of the digital twin construction process of the present invention.

[0023] Figure 3 This is a schematic diagram of the fault root cause tracing process of the present invention.

[0024] Figure 4 This is a schematic diagram of the virtual verification and closed-loop optimization process of the present invention. Detailed Implementation

[0025] Example 1 like Figures 1-4 As shown, this invention proposes a method for fault diagnosis and root cause tracing of bridge erecting machines, including the following steps: Step 1: Planning of Multiphysics Measurement Points and Selection of Sensors for Key Components: Using finite element simulation (ANSYS software), analyze the high-fault locations and multiphysics coupling sensitive areas of key components such as drive motors, reducers, and hoisting mechanisms. Combined with Failure Mode and Effects Analysis (FMEA), determine the location of measurement points. Deploy multiphysics sensors at each measurement point: mechanical load sensor (range 0-200kN, accuracy 0.1% FS), infrared temperature sensor (measurement range -40℃~200℃, error ±0.1℃), current and voltage sensor (accuracy 0.05% FS), and fluid velocity sensor (range 0-10m / s, resolution 0.01m / s), ensuring full coverage of multiphysics data.

[0026] Step 2: Multiphysics Data Synchronization Acquisition and Preprocessing: The sensor is triggered by the edge computing gateway to collect data at a preset frequency. The sampling frequency is 256Hz for mechanical data, 10Hz for temperature data, 100Hz for electromagnetic parameters, and 5Hz for fluid data. The NTP protocol is used to achieve timestamp synchronization (synchronization error ≤1ms). The collected data is preprocessed by removing outliers using the 3σ criterion, filtering environmental noise using an improved adaptive wavelet denoising algorithm, and normalizing to the [0,1] interval to form a standardized multiphysics dataset.

[0027] Step 3: Equipment structural parameters and historical data collection: Collect structural parameters of key components of the bridge erecting machine (such as gear module, number of teeth, bearing clearance, motor winding resistance, heat dissipation channel cross-sectional area, etc.), design manual data and historical fault records (including fault type, maintenance plan, root cause analysis results), and build a basic information database to provide data support for digital twin modeling and root cause tracing.

[0028] Step 4: Layered construction of digital twin: Based on the Unity3D platform, a 1:1 geometric model is built to accurately restore the structural dimensions and assembly relationships of key components; the ANSYS mechanical model, Fluent thermal conduction and fluid model, and Maxwell electromagnetic model are integrated to construct a multiphysics coupling model; historical fault data is used to train the fault evolution model to clarify the dynamic process of faults caused by different root causes, and finally a three-layer digital twin of "geometry-physics-fault" is formed.

[0029] Step 5: Real-time mapping and calibration of virtual and physical data: Real-time transmission of pre-processed data at the edge and the digital twin is achieved through 4G / 5G networks, and the MQTT protocol (QoS 2 level) is used to ensure the reliability of data transmission; the virtual model parameters are calibrated every hour using measured data from physical devices (such as temperature and vibration amplitude of key parts) to ensure the consistency between the virtual model and the physical devices, with a mapping error of ≤3%.

[0030] Step 6: Multiphysics coupling feature extraction: Call the improved coupling feature extraction algorithm to calculate the mutual information entropy between multiphysics data, and extract 12-dimensional core coupling features by combining the finite element analysis results, and remove redundant information.

[0031] Step 7: Fault type identification and root cause tracing: Input the coupling features into a lightweight CNN model (input layer dimension 12, hidden layer 2, output layer dimension 8) to identify the fault type; based on the fault root cause knowledge base (containing 1000+ historical cases), build an association rule engine, and combine equipment structural parameters and coupling features to locate the root cause of the fault.

[0032] Step 8: Virtual Reproduction of Fault Evolution: Input the root cause parameters of the fault and the current equipment status data into the digital twin, start the multiphysics coupling simulation, reproduce the complete evolution process of the fault from its inception to its manifestation, visualize the impact of multiphysics coupling on the fault development, and verify the accuracy of the root cause tracing results.

[0033] Step 9: Virtual simulation and optimization of maintenance plan: Generate more than 3 candidate maintenance plans based on the root cause tracing results, simulate the implementation effect of each plan in the digital twin, and evaluate the indicators including fault repair rate, equipment life extension cycle, maintenance cost, and downtime. Select the optimal plan through multi-objective optimization algorithm.

[0034] Step 10: Push and execute the optimal maintenance plan: Push the optimal plan (including operation steps, tool list, parameter standards, and safety precautions) to the operation and maintenance personnel through the Web / APP platform. The operation and maintenance personnel perform maintenance operations according to the plan and record the operation process and key data in a synchronous manner.

[0035] Step 11: Physical verification and data feedback of maintenance effect: After maintenance is completed, start the equipment trial operation, collect multiphysics field data during the trial operation phase, and verify whether the fault has been completely repaired; record the maintenance effect through the field terminal and upload it to the cloud database to form a feedback dataset.

[0036] Step 12: Iterative optimization of digital twin and knowledge base: Adjust the physical model parameters of digital twin based on feedback dataset to optimize the accuracy of multi-physics coupling simulation; input new fault cases (including type, root cause, maintenance plan, and effect) into fault root cause knowledge base, and use incremental learning algorithm to update the parameters of "type recognition-root cause tracing" linkage model to continuously improve diagnostic accuracy.

[0037] Step 13: Continuous monitoring and dynamic optimization: Once the system enters normal operation, collect multi-physics data and equipment operating status in real time. Repeat steps 5-12 to dynamically update the digital twin and knowledge base, ensuring adaptability to new fault types and achieving continuous optimization throughout the entire lifecycle.

[0038] Example 2 This embodiment applies to the JQ900A type box girder bridge erecting machine (a commonly used whole-span bridge erecting machine for railway passenger dedicated lines, with a rated lifting capacity of 900t and suitable for 32m standard box girders). The project it supports is the Guangzhou North Second Ring Expressway reconstruction and expansion project. The target fault scenario is: the drive motor of the left-span lifting mechanism of the bridge erecting machine (model Y2-400L-4, rated power 315kW, rated voltage 380V, rated current 578A) continuously alarms for over-temperature of the stator winding during operation. Traditional maintenance misjudges the problem as aging of the stator winding insulation based on single data such as vibration and temperature. This invention aims to use the method to complete the entire process of fault diagnosis, root cause tracing, maintenance optimization, and closed-loop iteration.

[0039] Complete implementation steps: Step 1: Planning of Multiphysics Measurement Points and Selection of Sensors for Key Components 1. First, fault simulation analysis was performed on the target drive motor and its matching reducer using ANSYS 2024R1 finite element software. The high-risk fault locations were identified as the motor stator windings, front and rear bearings, cooling duct inlet, and reducer input shaft gear. Then, Fault Mode and Effects Analysis (FMEA) was used to clarify the multiphysics coupling sensitive areas of these locations. Finally, six core measurement points were determined, and a multiphysics sensor group was deployed at each point. The specific selection and deployment are as follows:

[0040] 2. All sensors are connected to the edge computing gateway (model Advantech UNO-2484G) via shielded cables, providing the hardware foundation for synchronous acquisition of multi-physics data.

[0041] Step 2: Synchronous Acquisition and Preprocessing of Multiphysics Data 1. Trigger all sensors to synchronously collect data at preset frequencies via an edge computing gateway. The sampling frequencies are set as follows: 256Hz for mechanical load / vibration data, 100Hz for electromagnetic parameters, 10Hz for temperature data, and 5Hz for fluid velocity data. Use the NTP network time protocol to synchronize the timestamps of all collected data, with a synchronization error of no more than 1ms, to ensure the spatiotemporal consistency of multi-physics data.

[0042] 2. Perform preprocessing on the collected raw data: Outlier removal: The 3σ criterion is used to identify outliers that exceed the mean ± 3 standard deviations and remove them. Missing data are filled in using linear interpolation. Noise filtering: An improved adaptive wavelet denoising algorithm is adopted. Addressing the shortcomings of traditional soft thresholding functions (constant bias) and hard thresholding functions (discontinuities), a continuously differentiable adaptive thresholding function is used. The threshold is dynamically adjusted according to the noise level of the wavelet decomposition level. When the absolute value of the wavelet decomposition coefficients is greater than or equal to the adaptive threshold, the coefficients are retained and adaptively shrunk according to the threshold. When the absolute value of the wavelet decomposition coefficients is less than the adaptive threshold, the coefficients are set to zero to filter noise. The wavelet basis is selected as db4 wavelet, and the decomposition level is set to 5 levels to filter environmental noise and construction interference noise. Data normalization: The extreme value normalization method is used to linearly map all preprocessed data to the interval between 0 and 1. The mapping rule is: the historical minimum value of a single set of parameters is the starting point of the mapping and the historical maximum value is the ending point of the mapping. The original data is proportionally converted to the interval between 0 and 1 to form a standardized multiphysics dataset.

[0043] Step 3: Equipment structural parameters and historical data acquisition 1. Collect the structural parameters of key components of the target bridge erecting machine: motor stator winding resistance 0.012Ω / phase, heat dissipation duct cross-sectional area 0.08m², fan rated speed 1480r / min, reducer gear module 12mm, number of teeth 83, bearing rated clearance 0.03mm; 2. Collect equipment design manuals, factory inspection reports, and 32 historical fault records (including fault types, maintenance plans, root cause analysis results, and operation and maintenance data) of the same type of bridge erecting machine in the North Second Ring Road project to build a basic information database and provide data support for digital twin modeling and fault root cause tracing.

[0044] Step 4: Layered Construction of Digital Twin Based on the Unity3D 2022 platform, a three-layer digital twin of the physical bridge-building machine, corresponding to the geometry, physics, and fault layers, was constructed. The specific construction process is as follows: 1. Geometric Model Layer: 1:1 3D models of drive motors and reducers are drawn using SolidWorks and imported into the Unity3D platform to accurately reproduce the structural dimensions, assembly relationships, and spatial positions of the components. The number of model faces is controlled within 500,000 to balance accuracy and operating efficiency. 2. Multiphysics Coupled Physical Model Layer: Based on the geometric model, a multiphysics coupled simulation model is integrated, and bidirectional coupling rules are set. Electromagnetic model: Ansys Maxwell was used to construct a two-dimensional transient electromagnetic field model of the motor to calculate electromagnetic losses and winding current distribution; Thermal-fluid model: Ansys Fluent was used to build a fluid-heat conduction coupling model of the motor cooling duct, and the duct velocity, heat dissipation efficiency and temperature field distribution were calculated. Mechanical Model: Ansys Workbench was used to build a structural mechanical model of the motor-reducer to calculate load distribution and vibration response; Coupling rules: The electromagnetic model and the thermal model are bidirectionally coupled, with electromagnetic loss serving as the heat source input for the thermal model, and temperature changes providing feedback to correct the winding resistance parameters of the electromagnetic model; the thermal model and the fluid model are bidirectionally coupled, with the fluid heat dissipation coefficient serving as the boundary condition for the thermal model, and temperature field distribution providing feedback to correct the fluid's physical properties; the coupling time step is set to 0.1s, and the convergence criterion is that the calculation residuals of each physical field do not exceed 1e-6. 3. Fault Evolution Model Layer: Based on historical fault data and finite element simulation data, a damage evolution model for motor overheating faults is constructed. The input parameters are duct blockage rate, ambient temperature, and motor load rate, and the outputs are winding temperature, insulation aging rate, and fault occurrence time. The training dataset consists of 24 sets of historical data on duct blockage faults of the same model of equipment and 100 sets of multiphysics simulation data. The Levenberg-Marquardt algorithm is used for training, and the convergence condition is that the mean square error of the training set does not exceed 1e-4. This clarifies the dynamic evolution law of faults from initiation to over-temperature alarm under different duct blockage rates.

[0045] Step 5: Real-time mapping and calibration of virtual and real data 1. Real-time transmission of standardized multi-physics field data at the edge and digital twin in the cloud is achieved through 5G network. The MQTT protocol (QoS level 2) is used to ensure the reliability and orderliness of data transmission, and the data transmission latency does not exceed 200ms. 2. Perform virtual model parameter calibration every hour. The core calibration parameters include winding resistance, heat dissipation coefficient, and load damping ratio. The calibration method is least squares parameter fitting. The calibration benchmark is the measured temperature, vibration amplitude, and flow velocity data of the physical equipment. The calculation rule for the state mapping error between the virtual model and the physical equipment is as follows: use the rated value of the corresponding parameter as the calculation benchmark, calculate the absolute difference between the virtual simulation value and the actual measured value on site, and then divide it by the rated value of the parameter to obtain the relative mapping error. After each calibration, ensure that the relative mapping error of all core parameters does not exceed 3%.

[0046] Step 6: Multiphysics Coupling Feature Extraction An improved coupled feature extraction algorithm integrating mutual information entropy and finite element analysis is used to complete the coupled feature extraction. The specific process is as follows: 1. The mutual information entropy method is used to calculate the coupling correlation between multi-physics parameters. By statistically analyzing the joint distribution probability and marginal distribution probability of two sets of time series data of different physics fields, the nonlinear correlation between the two sets of data is quantified. The higher the value, the stronger the coupling correlation between the two sets of physics parameters. 2. Based on the parameter correlation obtained from finite element simulation analysis, a weight allocation was set: the mutual information entropy calculation result accounted for 60% of the weight, and the correlation of finite element simulation accounted for 40%. Finally, 12 core coupling features strongly correlated with fault evolution were extracted, including: radial load-vibration acceleration coupling coefficient; electromagnetic loss-stator winding temperature correlation; cooling duct fluid velocity-winding heat dissipation efficiency coefficient; bearing clearance-vibration peak value correlation; motor load rate-current harmonic distortion rate coupling coefficient; ambient temperature-winding temperature rise rate correlation; gear meshing frequency-vibration acceleration coupling coefficient; winding temperature-insulation resistance attenuation rate correlation; duct pressure difference-fluid velocity coupling coefficient; voltage imbalance-electromagnetic loss correlation; axial load-bearing temperature coupling coefficient; and speed fluctuation-vibration amplitude correlation.

[0047] 3. Remove redundant feature information that is irrelevant to fault evolution and complete the construction of the coupled feature dataset.

[0048] Step 7: Fault Type Identification and Root Cause Tracing 1. Fault Type Identification: 12-dimensional coupled features are input into a lightweight CNN fault identification model. The specific model structure is as follows: Input layer (12-dimensional, corresponding to 12 coupled features) → First hidden layer (convolutional layer, 3×1 kernel, stride 1, 8 output channels, ReLU activation function) → Second hidden layer (fully connected layer, 16 neurons, ReLU activation function, Dropout rate 0.2) → Output layer (8-dimensional, corresponding to 8 common faults: motor overheating, gear noise, bearing wear, winding insulation degradation, duct blockage, load overload, outrigger deformation, brake failure); the model outputs a fault type identification result of motor overheating fault, with a confidence level of 99.2%. 2. Fault Root Cause Tracing: An association rule engine is built based on a fault root cause knowledge base of 1200+ historical cases. It combines equipment structural parameters, coupling features, and fault type identification results to perform multi-dimensional evidence matching. The calculation rule for the fault root cause matching degree is as follows: the total matching degree is composed of two weighted components, of which evidence item matching accounts for 60% and coupling feature correlation accounts for 40%. The evidence item matching degree is the ratio of the number of successfully matched evidence items to the total number of evidence items. There are a total of 10 evidence items, which are divided into three categories: structural parameter compliance, coupling feature compliance, and historical case compliance. When the final matching degree reaches 95% or above, the corresponding root cause of the fault and supporting evidence are output.

[0049] After matching, the final output result is: the root cause of the fault is blockage of the motor cooling air duct (blockage rate of about 85%), with a matching degree of 97.8%, which meets the threshold requirements. The supporting evidence is as follows: ① The velocity difference between the air duct inlet and outlet is not less than 2.8 m / s, and the velocity-heat dissipation efficiency coefficient is less than the threshold of 0.3; ② The correlation between winding temperature and electromagnetic loss is only 0.2, ruling out overheating caused by electromagnetic loss; ③ The winding insulation resistance meets the standard, ruling out insulation aging; ④ The matching degree of historical cases of air duct blockage faults under the same operating conditions is 100%.

[0050] Step 8: Virtual Reproduction of Fault Evolution Input the root cause parameters of the fault (air duct blockage rate 85%) and the current equipment operating parameters (load rate 75%, ambient temperature 32℃) into the digital twin, and start the multi-physics field coupled transient simulation to reproduce the complete evolution process of the fault from its inception to the over-temperature alarm: air duct blockage → insufficient cooling airflow → decreased winding heat dissipation efficiency → continuous heat accumulation → continuous increase in winding temperature → over-temperature alarm. Visualize the impact of multi-physics field coupling on the fault development. The error between the simulation-obtained steady-state winding temperature and the measured value is no more than 2.2%, which verifies the accuracy of the fault root cause tracing results.

[0051] Step 9: Virtual simulation and optimization of maintenance plan 1. Generate 3 candidate maintenance schemes based on the root cause tracing results: Option 1: Disassemble the motor end cover, thoroughly clean the dust and debris in the heat dissipation air duct, and replace the air duct sealing strip; Option 2: Replace the motor stator windings and insulation materials, and clean the air ducts simultaneously; Option 3: Replace with a brand new drive motor of the same model; 2. The implementation effects of the three schemes were simulated in the digital twin. The evaluation indicators included fault repair rate, equipment life extension period, maintenance cost, and downtime. The NSGA-II multi-objective optimization algorithm was adopted with the optimization objectives of "minimizing maintenance cost, minimizing downtime, and maximizing equipment remaining life". The optimal scheme was finally selected as Scheme 1. The simulation results show that after the implementation of Scheme 1, the winding temperature can be restored to the rated range, the fault repair rate is 100%, the maintenance cost is only 12% of Scheme 2 and 3% of Scheme 3, and the downtime can be controlled within 4 hours.

[0052] Step 10: Push and execute the optimal maintenance plan The optimal maintenance plan is pushed synchronously to the mobile APP of the operation and maintenance personnel through the web management platform. The plan includes: operation steps, special tool list, air duct cleaning parameter standards, power outage and interlock safety precautions, and acceptance standards. The operation and maintenance personnel strictly follow the plan to perform maintenance operations, and record that the total amount of air duct debris cleaned is 1.2kg, and the air duct blockage cleaning completion rate is 98%.

[0053] Step 11: Entity verification and data feedback of maintenance results After maintenance, the bridge erecting machine was put into no-load and rated load trial operation for 2 hours. Multi-physics field data were collected simultaneously during the trial operation. The verification results were as follows: the steady-state temperature of the motor stator winding dropped to 68℃ (rated allowable temperature 120℃), the airflow velocity in the air duct returned to the design value of 5.2m / s, the overheat alarm was completely eliminated, and the fault was completely repaired. The maintenance process data, trial operation data, and maintenance effect verification results were uploaded to the cloud database through the field terminal to form a feedback dataset.

[0054] Step 12: Iterative Optimization of Digital Twin and Knowledge Base Based on the feedback dataset, the physical model parameters of the digital twin, such as the heat dissipation coefficient and airflow resistance, were adjusted. After optimization, the multiphysics coupling simulation error was reduced to within 1.8%. The fault case (fault type, root cause, maintenance plan, and repair effect) was added to the fault root cause knowledge base. The parameters of the lightweight CNN model and the association rule engine were updated using an incremental learning algorithm. The model fault identification accuracy improved by 0.3 percentage points compared with before the update.

[0055] Step 13: Continuous Monitoring and Dynamic Optimization Once the system enters a normal operating state, it collects multi-physics field operation data and equipment status of the bridge erecting machine in real time, repeats steps 5-12, and dynamically updates the digital twin and fault knowledge base to achieve fault diagnosis and dynamic optimization of operation and maintenance throughout the entire life cycle of the bridge erecting machine.

[0056] III. Verification of the Effects of the Examples This embodiment fully realizes the entire process of fault diagnosis for bridge erecting machines, including "type identification, root cause tracing, virtual verification, efficient repair, and closed-loop optimization." It accurately identifies motor overheating faults and precisely locates the root cause as air duct blockage, completely avoiding the problem of "misjudging winding insulation aging" in traditional diagnosis. The root cause tracing accuracy is close to 100%. Compared with traditional maintenance solutions, the maintenance trial and error costs are significantly reduced, downtime is significantly shortened, and there is no secondary damage to any equipment, verifying the technical effectiveness of the method of this invention.

[0057] IV. Comparison with Traditional Diagnostic Methods For the same fault scenario, the traditional bridge erecting machine fault diagnosis method is used: only motor vibration and winding temperature data are collected, the fault type is identified as motor overheating by LSTM algorithm, the root cause of the fault is determined to be aging of stator winding insulation based on feature matching, and the recommended maintenance solution is to replace the motor winding. The maintenance cost of this solution is 8 times that of the optimal solution of this invention, the downtime is 24 hours, and the overheating fault is not eliminated after replacing the winding. In the end, it is still necessary to clean the air duct, resulting in ineffective maintenance costs and additional downtime losses, which further verifies the significant advantages of this invention over traditional technology.

[0058] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing and tracing the root cause of a fault in a bridge-launching machine, characterized in that, Includes the following steps: S1. Collect multi-physics field operation data of key components of the bridge erecting machine, and simultaneously acquire structural parameters, historical fault data and historical operation and maintenance data of key components of the bridge erecting machine to build a basic dataset; S2. Construct a digital twin corresponding to the physical bridge erecting machine in a 1:1 ratio based on the basic dataset, and establish a real-time virtual-real data mapping between the physical bridge erecting machine and the digital twin; S3. Perform multi-field coupling correlation analysis on multi-physics field operation data to extract coupling features related to fault evolution; S4. Based on the extracted coupling features, combined with structural parameters and historical fault data, fault type identification and root cause location are completed simultaneously. S5. Input the fault root cause parameters and the current equipment operating status into the digital twin, and reproduce the fault evolution process through multi-physics field coupling simulation to verify the accuracy of the fault root cause location results. S6. Generate at least two candidate maintenance schemes based on the verified root causes of the faults. Simulate the repair effect of each scheme through a digital twin and optimize to obtain the optimal maintenance scheme. After the maintenance is performed, collect the equipment trial operation feedback data. Update the digital twin, fault diagnosis model and fault root cause knowledge base based on the feedback data to form a closed-loop optimization of the entire process.

2. The method according to claim 1, wherein, In step S1, before collecting multiphysics operation data, the high-incidence fault locations of key components of the bridge erecting machine and the multiphysics coupling sensitive areas are analyzed by finite element simulation. The location of the measuring points is determined by combining fault mode influence analysis (FMEA). A multiphysics sensor group is deployed at each measuring point. The multiphysics sensor group includes at least a mechanical load sensor, an infrared temperature sensor, a current and voltage sensor, and a fluid flow velocity sensor.

3. The method according to claim 2, wherein, In step S1, each sensor is triggered to collect data synchronously at a preset frequency through the edge computing gateway. The NTP network time protocol is used to realize the timestamp synchronization of all collected data, with a synchronization error of ≤1ms. For the collected raw data, singular values ​​are removed by the 3σ criterion, environmental noise is filtered by the adaptive wavelet denoising algorithm, and then normalized to the [0,1] interval to form standardized multiphysics operation data.

4. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 1, characterized in that, In step S2, the constructed digital twin adopts a three-layer architecture of "geometry-physics-fault". Specifically, it is as follows: a 1:1 geometric model is constructed based on the Unity3D platform, which is completely consistent with the structural dimensions and assembly relationship of the key components of the physical bridge erecting machine; a mechanical model, a heat conduction and fluid model, and an electromagnetic model are integrated on the basis of the geometric model to construct a multi-physics field coupled physical model; a fault evolution model is obtained by training based on historical fault data to clarify the dynamic evolution law of faults caused by different fault root causes, and finally a three-layer integrated digital twin is formed.

5. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 4, characterized in that, In step S2, the pre-processed multi-physics operation data at the edge is transmitted in real time to the digital twin via 4G / 5G network. The MQTT message queue telemetry transmission protocol with QoS level 2 is used to ensure the reliability of data transmission. The physical model parameters of the digital twin are calibrated according to the measured operation data of the physical bridge erecting machine at a preset period to ensure that the state mapping error between the virtual model and the physical device is ≤3%.

6. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 1, characterized in that, In step S3, an improved coupling feature extraction algorithm that integrates mutual information entropy and finite element analysis results is used to calculate the mutual information entropy between different physical field parameters in the multi-physics field operation data, quantify the coupling correlation between multi-physics fields, and extract at least 12 core coupling features that are strongly correlated with fault evolution by combining the finite element simulation analysis results, and remove redundant feature information.

7. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 1, characterized in that, In step S4, the extracted coupling features are input into a lightweight convolutional neural network (CNN) model to complete fault type identification. The input layer dimension of the lightweight CNN model matches the coupling feature dimension, and the hidden layer is set to 2 layers. An association rule engine is built based on a fault root cause knowledge base. It combines equipment structural parameters, coupling features and fault type identification results to perform multi-dimensional evidence matching. When the matching degree is ≥95%, the corresponding root cause of the fault and supporting evidence are output.

8. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 1, characterized in that, In step S6, the evaluation indicators for each candidate maintenance scheme in the simulation evaluation include at least the fault repair rate, equipment life extension cycle, maintenance cost and downtime. A multi-objective optimization algorithm is adopted to minimize maintenance cost, shortest downtime and maximize equipment remaining life as optimization objectives to select the optimal maintenance scheme, and push the operation steps, tool list, parameter standards and safety precautions of the optimal maintenance scheme to the operation and maintenance terminal.

9. The method for fault diagnosis and root cause tracing of a bridge erecting machine according to claim 1, characterized in that, In step S6, the update based on feedback data specifically involves: adjusting the physical model parameters of the digital twin based on the feedback data from the equipment trial operation, and optimizing the accuracy of multi-physics coupling simulation; adding the type, root cause, maintenance plan, and repair effect of this fault to the fault root cause knowledge base, and using incremental learning algorithms to update the parameters of the fault diagnosis model and the association rule engine to continuously improve the accuracy of fault diagnosis and root cause tracing.

10. A method for fault diagnosis and root cause tracing of a bridge erecting machine according to any one of claims 1 to 9, characterized in that, After completing a single fault diagnosis and maintenance loop, the system enters a normal operating state, collecting multi-physics field operation data and equipment status of the bridge erecting machine in real time, and repeatedly executing virtual-real mapping calibration, coupling feature extraction, fault monitoring and diagnosis, virtual verification and closed-loop optimization steps to realize fault diagnosis and dynamic optimization of operation and maintenance throughout the entire life cycle of the bridge erecting machine.

Images