Intelligent monitoring method and system for locking state of bottom door of coal hopper car

Abstract

The invention discloses a coal hopper car bottom door locking state intelligent monitoring method and system, and relates to the technical field of railway freight equipment state monitoring, and the method comprises the steps: collecting bottom door multi-source monitoring data, and preprocessing the data to obtain a standardized data stream; outputting a locking state judgment result and confidence thereof through a multi-source information fusion model; constructing a digital twinborn model, and carrying out online unification processing by taking a judgment result and confidence as boundary conditions; executing advanced simulation based on the unification model, and outputting a health state index and a residual life prediction value; and a maintenance strategy is generated by integrating the information, and closed-loop optimization is realized through sensitivity analysis. According to the invention, the integration of accurate sensing, reliability evaluation and life prediction of the locking state of the bottom door of the coal hopper car is realized, and meanwhile, a'monitoring-prediction-optimization 'closed-loop system is constructed by utilizing advanced simulation and sensitivity feedback, so that the accuracy of state judgment and the perspectiveness of operation and maintenance decision are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of railway freight equipment condition monitoring technology, specifically to an intelligent monitoring method and system for the locking status of the bottom door of a coal hopper car. Background Technology

[0002] In the railway freight sector, reliable monitoring of the locking status of coal hopper car bottom doors is crucial for ensuring transportation safety and improving loading and unloading efficiency. Current technologies primarily revolve around single-sensor mechanisms, including mechanical position detection based on displacement sensors, motion feature recognition through vibration signal analysis, and image comparison methods using machine vision for locking pins. In recent years, with the penetration of IoT technology into railway equipment, a multi-sensor data acquisition framework has been initially established. Simultaneously, the application of the digital twin concept in mechanical equipment health management has provided new insights into condition monitoring. Existing research generally focuses on enhancing system reliability by improving the accuracy of single detection channels or introducing redundant sensor arrays, but a closed-loop decision-making system integrating physical mechanisms and data-driven approaches has yet to be formed.

[0003] Existing technologies suffer from three main shortcomings: First, monitoring methods based on a single information source struggle to cope with complex operating conditions. Displacement sensors are susceptible to transient false signals caused by mechanical vibrations; vision systems experience a significant drop in recognition rate under low light or dust-covered conditions; and vibration analysis methods cannot effectively distinguish between normal driving vibrations and abnormal collisions of the locking mechanism. This limitation in perception dimensions leads to persistently high false alarm and false negative rates in real-world operating environments, failing to meet high reliability requirements. Second, existing methods generally lack the ability to quantitatively assess the reliability of monitoring results. Traditional binary judgment outputs (locked / unlocked) fail to provide a level of confidence for decision-making. When sensor data is contradictory or subject to noise interference, maintenance personnel cannot determine the reliability of the monitoring results, introducing uncertainty into safety decisions. Most critically, existing technologies suffer from a disconnect between perception and prediction—condition monitoring and lifespan assessment are separate systems. Traditional monitoring methods can only provide a current state assessment and cannot predict the performance degradation trend of the locking mechanism based on the current state, nor can they feed the predicted information back to the monitoring stage for model optimization. This one-way information flow causes maintenance strategies to always lag behind actual needs, making it impossible to achieve predictive maintenance or continuously improve the accuracy of status identification through long-term operational data. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing methods for monitoring the condition of coal hopper car bottom doors suffer from insufficient reliability due to the reliance on a single monitoring method, a lack of credible quantitative evaluation of monitoring results, a disconnect between condition judgment and health prediction, and the problem of how to achieve accurate condition monitoring and predictive maintenance closed-loop management.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide an intelligent monitoring method for the locking status of the bottom door of a coal hopper car, including collecting multi-source monitoring data of the bottom door of the coal hopper car and preprocessing it to obtain a standardized data stream; The standardized multi-source data stream is input into a preset multi-source information fusion decision model for fusion inference, and a binary judgment result is output. At the same time, a first confidence level characterizing the reliability of the binary judgment result is also output. A digital twin virtual model corresponding to the physical locking mechanism is constructed. The binary judgment result and the first confidence level are used as the primary boundary conditions, and the feature parameters in the standardized data stream are used as auxiliary driving data. They are all input into the digital twin virtual model, and online consistency processing is performed to obtain a consistent virtual model. The standardized virtual model is used to perform advanced simulation, calculate and output the health status index and the predicted value of remaining useful life; By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated. Data sensitivity information is obtained based on the analysis of the health status index and the predicted remaining useful life, and then fed back to the multi-source information fusion decision model.

[0007] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car described in this invention, the multi-source monitoring data includes force and strain data reflecting the locking force state. Displacement, clearance, and angle data reflecting the relative state of the door; Magnetic field, magnetic flux, and inductance data reflecting the state of the locked magnetic circuit; Actuation process data characterizing the operating conditions of the actuator include: pressure, stroke and valve position, flow rate, drive current and voltage; Thermal data reflecting the effects of sealing and icing include: temperature, heat flux, and heat distribution information; Environmental and operating parameters include: ambient temperature, humidity, icing and snow accumulation indicators, dust concentration, vehicle speed, and load information.

[0008] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car according to the present invention, the preprocessing includes performing filtering and noise reduction and timestamp synchronization and alignment operations on the data with time series characteristics in the multi-source monitoring data. Image enhancement and distortion correction operations are performed on the heat distribution information in the multi-source monitoring data; A quality assessment operation is performed on the multi-source monitoring data, and a data quality flag is generated for each data source according to predefined rules; Based on the data quality flag, invalid data is removed, and dimensional normalization is performed on valid numerical data to form the standardized data stream.

[0009] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car described in this invention, the preset multi-source information fusion decision model is a probabilistic reasoning model constructed based on DS evidence theory. The fusion reasoning includes: transforming each data source in the standardized data stream into an independent evidence source, and constructing a basic probability assignment for each evidence source for the {locked, unlocked} proposition; Calculate the dynamic weight of the evidence source corresponding to each data source based on the data quality flag bit. The basic probability allocation of the corresponding evidence sources is corrected using the dynamic weights, and the Dempster combination rule is used to synthesize all the corrected evidence sources to obtain the reliability and similarity of the synthesized proposition. The confidence level is used as an intermediate probability value and compared with a preset judgment threshold to generate the binary judgment result; The first confidence level is obtained by weighted calculation based on the interval width formed by confidence and similarity, and the relative distance between the intermediate probability value and the judgment threshold.

[0010] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car described in this invention, the online consistency processing includes constructing a digital twin virtual model that includes the geometric structure, material physical properties and multibody dynamics characteristics of the physical locking mechanism. The binary determination result is used as the initial state constraint condition of the digital twin virtual model; The first confidence level is used as a control factor for the calibration intensity of model parameters; The displacement, gap, and angle data in the standardized data stream are used as the geometric motion driving input for the digital twin virtual model; Force and strain data, as well as pressure, flow rate, driving current and voltage data from the standardized data stream, are used as dynamic state observation inputs for the digital twin virtual model; The dynamic response of the calculation model is verified online with the actual state of the physical mechanism. When the response deviation exceeds the preset threshold, the friction coefficient and connection stiffness parameters in the model are adaptively corrected based on the control factor of the first confidence level. When the response deviation drops to within the preset threshold, the digital twin virtual model is determined to have completed online consistency processing, and the model with the corrected parameters at this time is output as the consistent virtual model.

[0011] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car described in this invention, the calculation and output of the health status index and the predicted value of the remaining service life includes setting a preset future operating condition that includes load spectrum, operating frequency and environmental conditions. The uniform virtual model is driven to perform advanced numerical simulation under the preset future operating conditions to simulate the performance evolution of the locking mechanism within a preset time range; The key performance parameters representing the health status of the mechanism are extracted from the simulation process. These key performance parameters include the peak locking contact force, locking time, and mechanism transmission efficiency. Based on the degradation trajectory of the key performance parameters, the percentage of performance degradation at the current moment relative to the initial healthy state is calculated to obtain the health state index. The simulation data of key performance parameters are compared with the preset failure threshold. The simulation time point when the performance parameter first exceeds the failure threshold is taken as the end point of the lifespan. The time from the current moment to the end point of the lifespan is calculated to obtain the predicted value of the remaining lifespan.

[0012] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car described in this invention, the generation of comprehensive monitoring conclusions including maintenance strategies includes establishing a three-level decision matrix including a real-time status layer, a health assessment layer, and a prediction layer. The real-time status layer is divided into two states: successful locking and failed locking, based on the binary judgment result; the health assessment layer is divided into three levels: healthy, alert, and warning, based on the health status index; and the prediction layer is divided into three levels: long-term availability, medium-term maintenance, and immediate repair, based on the remaining service life prediction value. The first confidence level is used as the weighting coefficient for the decision results at each level in the three-level decision matrix; Based on the weighted three-level decision matrix output, a predefined maintenance strategy rule base is matched to generate a comprehensive monitoring conclusion that includes maintenance level, recommended measures, and execution time. Specifically, when the binary determination result is locking failure, the highest level real-time alarm is immediately generated; when the health status index is at the warning level and the remaining service life prediction value is at the immediate maintenance level, a maintenance instruction for time-limited maintenance is generated. The comprehensive monitoring conclusions, along with the corresponding health status index and the predicted remaining service life, are packaged into a standardized maintenance work order and output to the maintenance management system.

[0013] As a preferred embodiment of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car according to the present invention, the analysis of data sensitivity information includes: based on the unified virtual model, performing parameter sensitivity analysis in the advanced simulation process, and calculating the influence factors of the monitoring parameters corresponding to each data source on the health status index and the predicted value of the remaining service life; Based on the magnitude of the impact factor, the data sources are ranked according to their sensitivity to generate data sensitivity information. The data sensitivity information is fed back to the multi-source information fusion decision model in real time; In the subsequent fusion inference process, the weight allocation of each data source in the fusion calculation is dynamically adjusted based on data sensitivity information.

[0014] Secondly, embodiments of the present invention provide an intelligent monitoring system for the locking status of the bottom door of a coal hopper car, comprising: Data acquisition and preprocessing module: Collects multi-source monitoring data from the bottom door of the coal hopper car and preprocesses it to obtain a standardized data stream; Multi-source information fusion decision module: Inputs the standardized multi-source data stream into the preset multi-source information fusion decision model for fusion reasoning, outputs a binary judgment result, and simultaneously outputs a first confidence level characterizing the reliability of the binary judgment result; Digital twin consistency module: Construct a digital twin virtual model corresponding to the physical locking mechanism, take the binary judgment result and the first confidence level as the primary boundary conditions, and take the feature parameters in the standardized data stream as auxiliary driving data, and input them into the digital twin virtual model to perform online consistency processing to obtain a consistent virtual model; Advanced simulation and health assessment module: Utilizes the standardized virtual model to perform advanced simulation, calculates and outputs the health status index and the predicted value of remaining useful life; Integrated Decision and Maintenance Output Module: By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated. Sensitivity analysis and feedback module: Based on the health status index and the predicted remaining useful life, data sensitivity information is obtained and fed back to the multi-source information fusion decision model.

[0015] The beneficial effects of this invention are as follows: By combining multi-source information fusion and digital twin technology, it achieves integrated accurate perception, reliability assessment and life prediction of the locking status of the bottom door of the coal hopper car, effectively overcoming the shortcomings of insufficient reliability of single sensor monitoring. At the same time, it uses advanced simulation and sensitivity feedback to construct a closed-loop system of "monitoring-prediction-optimization", which significantly improves the accuracy of status judgment and the foresight of operation and maintenance decisions, providing a complete technical solution for predictive maintenance of railway freight equipment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 The flowchart shows the overall process of the intelligent monitoring method for the locking status of the bottom door of the coal hopper car provided in the first embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0018] Example 1, referring to Figure 1 As an embodiment of the present invention, an intelligent monitoring method for the locking status of the bottom door of a coal hopper car is provided, comprising: S1: Collect multi-source monitoring data from the bottom door of the coal hopper car and preprocess it to obtain a standardized data stream.

[0019] Multi-source monitoring data includes force and strain data reflecting the locked state; Displacement, clearance, and angle data reflecting the relative state of the door; Magnetic field, magnetic flux, and inductance data reflecting the state of the locked magnetic circuit; Actuation process data characterizing the operating conditions of the actuator include: pressure, stroke and valve position, flow rate, drive current and voltage; Thermal data reflecting the effects of sealing and icing include: temperature, heat flux, and heat distribution information; Environmental and operating parameters include: ambient temperature, humidity, icing and snow accumulation indicators, dust concentration, vehicle speed, and load information.

[0020] It should be noted that the selection of multi-source monitoring data is based on an in-depth analysis of the failure mechanism of the locking mechanism. Among them, force and strain data directly reflect the stress state of mechanical components during the locking process, providing direct mechanical evidence for determining whether the locking is in place; displacement, gap, and angle data quantify the spatial relationship between the door and the locking components, providing a geometric basis for determining whether the locking action is completed; magnetic field, magnetic flux, and inductance data are specifically used to monitor the electromagnetic locking device, indirectly determining whether the iron core is properly engaged through changes in magnetic circuit characteristics, solving the problem that purely mechanical sensors cannot detect the internal state of the electromagnetic mechanism; actuation process data such as pressure, flow, current, and voltage reflect the working state of the actuator (such as cylinders, hydraulic cylinders, or motors), and its abnormal modes (such as increased current accompanied by insufficient stroke) can provide early warning of actuator jamming or internal leakage faults; thermal distribution information, by analyzing the temperature field differences in the door gap area, is innovatively used to assess the contact state of the sealing strip and detect icing phenomena, which is a unique non-contact method for monitoring the effectiveness of the seal; environmental and operating parameters construct the boundary conditions for system operation, used to distinguish between state changes caused by external environment (such as material shrinkage due to low temperature, and increased friction due to dust accumulation) and the performance degradation of the mechanism itself, significantly improving the contextual perception ability of state judgment. This multi-physical-quantity, multi-perspective perception system forms a reliable data foundation for subsequent intelligent diagnosis and prediction.

[0021] For the time-series data in the multi-source monitoring data, perform filtering, noise reduction, and timestamp synchronization and alignment operations; Image enhancement and distortion correction operations are performed on the heat distribution information in the multi-source monitoring data; A quality assessment operation is performed on the multi-source monitoring data, and a data quality flag is generated for each data source according to predefined rules; Based on the data quality flag, invalid data is removed, and dimensional normalization is performed on valid numerical data to form the standardized data stream.

[0022] It should also be noted that the preprocessing process is tailored to the heterogeneous, noisy, and asynchronous characteristics of industrial field data. Filtering and noise reduction primarily target easily interfered signals such as vibration and current, employing digital filters (such as Butterworth low-pass filters) to suppress high-frequency electromagnetic noise and random pulses, while retaining characteristic frequency bands reflecting the essence of mechanism movements, providing clean signals for subsequent feature extraction. Timestamp synchronization and alignment, through hardware triggering or software interpolation algorithms, unifies data distributed across different acquisition modules to the same time base, eliminating causal logic errors caused by sampling delays in multi-source data—a prerequisite for accurate fusion inference. Image enhancement and distortion correction, targeting infrared thermal imager data, enhance the thermally subtle contrast of the sealing contact area through histogram equalization and correct lens distortion using camera calibration parameters, ensuring the spatial geometric authenticity of the thermal distribution image and making the sealing condition assessment results based on temperature field analysis more reliable. Quality assessment and flag generation are key innovative steps in improving system robustness. These steps are based on predefined rules, including the reasonableness of signal amplitude (whether it is within the sensor's range), the continuity of the rate of change (whether there is abrupt noise), and logical consistency with associated signals (e.g., whether pressure changes synchronously when displacement increases). This step effectively identifies anomalies such as sensor failure, signal loss, and communication interruption, and provides a basis for data reliability for subsequent fusion algorithms. Dimensional normalization maps values ​​with different physical meanings and magnitudes (e.g., pressure in MPa, displacement in mm, current in A) to a dimensionless specific interval (e.g., [0,1]), eliminating the negative impact of data scale differences on model convergence and stability. The entire preprocessing process works synergistically to transform raw, coarse field data into a high-quality, highly consistent "standardized data stream," providing stable and reliable data input for subsequent complex intelligent analysis algorithms, thus ensuring the accuracy and reliability of the entire monitoring system from the source.

[0023] S2: Input the standardized multi-source data stream into the preset multi-source information fusion decision model for fusion reasoning, output a binary judgment result, and simultaneously output a first confidence level characterizing the reliability of the binary judgment result.

[0024] The pre-defined multi-source information fusion decision model is a probabilistic reasoning model based on DS evidence theory; Fusion reasoning includes: transforming each data source in the standardized data stream into an independent source of evidence, and constructing a basic probability assignment for each source of evidence for the {locked, unlocked} proposition; Calculate the dynamic weight of the evidence source corresponding to each data source based on the data quality flag bit. The basic probability allocation of the corresponding evidence sources is corrected using the dynamic weights, and the Dempster combination rule is used to synthesize all the corrected evidence sources to obtain the reliability and similarity of the synthesized proposition. The confidence level is used as an intermediate probability value and compared with a preset judgment threshold to generate the binary judgment result; The first confidence level is obtained by weighted calculation based on the interval width formed by confidence and similarity, and the relative distance between the intermediate probability value and the judgment threshold.

[0025] It should be noted that the preset multi-source information fusion decision model is a probabilistic reasoning model built on DS evidence theory. The construction of this model begins with defining a clear identification framework. In this invention, this framework includes two mutually exclusive basic propositions: "locked" and "unlocked." The core task of the model is to uniformly transform standardized data from different physical sensors, with different dimensions and characteristics, into quantitative supporting evidence for the above two propositions, and to make a scientific and comprehensive decision.

[0026] The fusion reasoning process specifically includes the following steps: First, each data source in the standardized data stream is transformed into an independent source of evidence. This transformation is manifested in constructing a unique basic probability assignment for each data source for the {locked, unlocked} proposition. The basic probability assignment is a mathematical function that assigns support probabilities to the three propositions: "locked," "unlocked," and "uncertain" (i.e., it could be either locked or unlocked). The key is that the probability value assigned to the "uncertain" proposition subtly characterizes the cognitive uncertainty inherent in the data source itself. For example, when a displacement sensor detects that the piston rod has reached a predetermined position, its basic probability assignment might strongly support the "locked" proposition, but it would also assign a small portion of the probability to the "uncertain" proposition to reflect factors such as measurement errors or mechanical clearances inherent in the sensor.

[0027] Secondly, based on the data quality flags corresponding to each data source, the dynamic weight of the corresponding evidence source is calculated. This step is a key innovation in improving the robustness of the system. The data quality flags originate from the preprocessing stage and quantify the instantaneous reliability of each data source. According to predefined mapping rules, the system converts high-quality flags into high dynamic weights and low-quality flags into low dynamic weights, or even zero weights. This means that the influence of evidence from a sensor experiencing signal distortion due to strong electromagnetic interference will be automatically suppressed in subsequent decision-making, thus preventing low-quality evidence from contaminating the final fusion result.

[0028] Subsequently, the basic probability allocations of the corresponding evidence sources are corrected using the aforementioned dynamic weights. The purpose of this correction is to weaken the influence of low-weight evidence and enhance the influence of high-weight evidence. Next, the Dempster combination rule is used to synthesize all the corrected evidence sources. The Dempster combination rule is a mathematical orthogonal sum operation that can orderly and pairwise merge the basic probability allocations of all evidence sources. The core mechanism of this rule is that it calculates the degree of conflict between different evidence sources (i.e., one piece of evidence strongly supports "locked" while another piece of evidence strongly supports "unlocked") and uses a normalization factor to handle this conflict, ultimately obtaining a synthesized basic probability allocation that comprehensively reflects the joint opinion of all evidence sources. From this, we obtain the reliability and plausibility of the synthesized proposition. Reliability is the sum of the strengths of all evidence that directly and explicitly support the "locked" proposition, representing the minimum level of credibility supporting the proposition; plausibility is the sum of the strengths of all evidence that explicitly oppose the "locked" proposition (i.e., support "unlocked"), representing the maximum level of credibility that the proposition may hold. Reliability and similarity together constitute a probability interval that precisely quantifies the overall uncertainty caused by incomplete, imprecise, or conflicting evidence.

[0029] Then, the confidence level is used as an intermediate probability value and compared with a preset decision threshold to generate the binary decision result. This is a conservative and reliable decision-making strategy. Only when the lower bound of evidence supporting "locking" (i.e., the confidence level) is strong enough to exceed the decision threshold will the system finally output a binary decision result of "locking"; otherwise, it outputs "not locked". This effectively reduces the risk of misjudging a safe state when the evidence is insufficient.

[0030] Finally, the first confidence level is calculated by weighting the interval width formed by confidence and similarity, and the relative distance between the intermediate probability value and the judgment threshold. The interval width reflects the consistency of the evidence; the narrower the interval, the less contradictory the evidence, the higher the consistency, and thus the higher the confidence level. The relative distance between the intermediate probability value and the judgment threshold reflects the clarity of the decision; the larger the distance, the less ambiguous the decision, and thus the higher the confidence level. By weighting and fusing the information from these two dimensions, the final output first confidence level is a quantitative comprehensive evaluation value of the reliability of this binary judgment result. This value provides crucial decision-making assistance information for downstream systems or operators. For example, when the binary judgment result is "locked" but the first confidence level is low, a review mechanism or warning can be triggered, thereby achieving a leap from simple state judgment to integrated intelligent perception of "state-confidence".

[0031] S3: Construct a digital twin virtual model corresponding to the physical locking mechanism, take the binary judgment result and the first confidence level as the primary boundary conditions, and take the feature parameters in the standardized data stream as auxiliary driving data, and input them into the digital twin virtual model to perform online consistency processing to obtain a consistent virtual model.

[0032] Construct a digital twin virtual model that includes the geometry, material physical properties, and multibody dynamics characteristics of the physical locking mechanism; The binary determination result is used as the initial state constraint condition of the digital twin virtual model; The first confidence level is used as a control factor for the intensity of model parameter calibration, wherein the lower the first confidence level, the greater the magnitude of the model parameter calibration triggered. The displacement, gap, and angle data in the standardized data stream are used as the geometric motion driving input for the digital twin virtual model; Force and strain data, as well as pressure, flow rate, driving current and voltage data from the standardized data stream, are used as dynamic state observation inputs for the digital twin virtual model; The dynamic response of the calculation model is verified online with the actual state of the physical mechanism. When the response deviation exceeds the preset threshold, the friction coefficient and connection stiffness parameters in the model are adaptively corrected based on the control factor of the first confidence level. When the response deviation drops to within the preset threshold, the digital twin virtual model is determined to have completed online consistency processing, and the model with the corrected parameters at this time is output as the consistent virtual model.

[0033] It should be noted that the construction of the aforementioned digital twin virtual model is a high-fidelity digital mapping process. It is not simply a three-dimensional geometric model, but a virtual entity that deeply integrates geometric structure, material physical properties, and multibody dynamics characteristics. The geometric structure accurately reproduces the shape, assembly relationship, and kinematic pair types (such as revolute joints and prismatic joints) of each component in the locking mechanism through 3D CAD modeling. Material physical properties infuse the model with a physical essence, including but not limited to the elastic modulus, density, and Poisson's ratio of key components; these parameters determine the deformation behavior of the components under stress. Multibody dynamics characteristics describe the motion law of the mechanism under the combined action of driving force, friction, and inertial force by establishing the system's dynamic differential equations. Such a multi-level model can respond to input and simulate dynamic behavior that highly approximates the physical entity.

[0034] The purpose of online consistency processing is to ensure that the virtual model and the physical entity remain synchronized in the current moment and in the short future time domain. First, the binary determination result is used as the initial state constraint condition for the digital twin virtual model. This means that at the beginning of each simulation cycle, the digital twin model is not calculated from zero, but is forcibly initialized to the macroscopic state (locked or unlocked) determined by step S2. This provides a correct starting point consistent with the physical world for subsequent dynamic simulations, avoiding a fundamental deviation between the model state and the entity state.

[0035] Secondly, the first confidence level is used as a control factor for the intensity of model parameter calibration. This is an adaptive mechanism. The first confidence level quantifies the uncertainty of the front-end perception and fusion results. When the first confidence level is low, it indicates that there is significant noise or conflict in the real-world observation data. In this case, the parameters of the digital twin model itself (such as friction coefficient and connection stiffness) may deviate significantly from the true values. Therefore, the system will trigger a larger-scale self-correction of model parameters, significantly adjusting the internal parameters of the model to try to match the model output with those observation data that, although highly uncertain, may be true. Conversely, when the first confidence level is high, fine calibration is performed to prevent overfitting. This enables the digital twin model to have the ability to learn and adapt in uncertain environments.

[0036] It should also be noted that during the simulation, displacement, gap, and angle data from the standardized data stream are used as the geometric motion driving inputs for the digital twin virtual model. These data directly define the time-varying position and attitude of the model's moving parts, forming the basis for driving the simulation. Simultaneously, force and strain data, as well as pressure, flow rate, driving current, and voltage data from the standardized data stream, are used as the dynamic state observation inputs for the digital twin virtual model. These data reflect the system's force state and energy conversion process, and are used for comparison with the theoretical dynamic responses calculated by the model (such as theoretical contact force and theoretical driving pressure).

[0037] Next, an online consistency check is performed. The system calculates the model's dynamic response in real time and performs an online consistency check with the actual state of the physical mechanism. Specifically, it compares the dynamic observations calculated by the model simulation (such as the theoretical force on the connecting rod) with the corresponding dynamic state observations actually measured from the physical mechanism (such as the force converted from the measured strain value) and calculates the response deviation.

[0038] When the response deviation exceeds a preset threshold, the friction coefficient and connection stiffness parameters in the model are adaptively corrected based on a control factor with a first confidence level. The friction coefficient and connection stiffness are key parameters in the model that are difficult to measure directly but have a significant impact on the dynamic response. The system iteratively adjusts these parameters according to the magnitude and direction of the deviation, combined with the calibration intensity determined by the first confidence level, using optimization algorithms (such as gradient descent), so that the theoretical output of the model continuously approximates the measured data.

[0039] When the response deviation drops below a preset threshold, the digital twin virtual model is deemed to have completed online consistency processing. At this point, the virtual model not only matches the physical entity in its macroscopic state, but its internal parameters and dynamic response characteristics have also been calibrated, enabling it to represent the true health status of the physical entity at the current moment with high fidelity. This calibrated and highly synchronized virtual model with the physical entity lays a reliable foundation for accurate advanced simulation and health prediction in the next step.

[0040] S4: Utilize the standardized virtual model to perform advanced simulation, calculate and output the health status index and the predicted value of remaining useful life.

[0041] Set preset future operating conditions that include load spectrum, operating frequency and environmental conditions; The uniform virtual model is driven to perform advanced numerical simulation under the preset future operating conditions to simulate the performance evolution of the locking mechanism within a preset time range; The key performance parameters representing the health status of the mechanism are extracted from the simulation process. These key performance parameters include the peak locking contact force, locking time, and mechanism transmission efficiency. Based on the degradation trajectory of the key performance parameters, the percentage of performance degradation at the current moment relative to the initial healthy state is calculated to obtain the health state index. The simulation data of key performance parameters are compared with the preset failure threshold. The simulation time point when the performance parameter first exceeds the failure threshold is taken as the end point of the lifespan. The time from the current moment to the end point of the lifespan is calculated to obtain the predicted value of the remaining lifespan.

[0042] It should be noted that the aforementioned advanced simulation is a predictive analysis based on a high-fidelity virtual model that has undergone online consistency processing in step S3. Its primary step is to set a preset future operating condition that includes load spectrum, operating frequency, and environmental conditions. This future operating condition is not a single scenario, but rather a typical operating condition spectrum constructed based on statistical analysis of historical operating data and task planning. For example, it can simulate continuous operation scenarios of the vehicle under heavy load, frequent start-stop, and low-temperature icy / snowy environments over the next month. This preset operating condition provides the simulation with external stimuli and boundary conditions that conform to actual expectations.

[0043] Furthermore, the uniform virtual model is driven to perform advanced numerical simulations under the preset future operating conditions. This simulation process runs in virtual space at a speed faster than the real-time clock, simulating the performance evolution of the locking mechanism within a preset time range. The simulation engine calculates the stress and strain, wear accumulation, and performance degradation of each component in the mechanism under continuous action cycles and loads by solving multibody dynamics equations.

[0044] During simulation, the system extracts time-series data of key performance parameters characterizing the health status of the mechanism. The selection of these parameters is crucial; they must be direct or indirect sensitive indicators of the reliability of the locking function. The peak locking contact force defined in this invention reflects the effective engagement strength between the locking hook and the seat, and its attenuation indicates increased wear. The locking time characterizes the smoothness of the mechanism's movement; a prolonged time may indicate increased friction or jamming. The mechanism's transmission efficiency comprehensively reflects the energy transfer loss from the power source to the actuator, serving as a macroscopic representation of the system's internal health status. These parameters collectively constitute a multi-dimensional indicator system for evaluating the health of the mechanism.

[0045] Then, based on the degradation trajectory of the key performance parameters, the percentage of performance degradation at the current moment relative to the initial healthy state is calculated to obtain the health status index. Specifically, the system sets an initial health baseline value for each key performance parameter (usually taken from data after the equipment is brand new or has undergone major repairs). At each time point in the simulation, the currently simulated parameter value is compared with the initial baseline value to calculate the proportion of performance loss. Finally, by weighting and fusing the degradation percentages of these key parameters, a comprehensive and quantitative health status index is formed. This index intuitively reflects the current health level of the organization; for example, a health status index of 85% indicates that its overall performance has degraded by 15% relative to its brand new state.

[0046] Finally, the simulation data of key performance parameters are compared with preset failure thresholds. The failure threshold is a pre-set limit value based on design specifications, safety guidelines, and engineering experience. When the performance parameter exceeds this threshold, the mechanism is considered to have failed. The system continuously monitors the simulation data stream, using the simulation time point when the performance parameter first exceeds the failure threshold as the end-of-life point. The duration from the current moment to the end-of-life point is calculated to obtain the predicted remaining service life. This predicted value is not a precise point in time, but an expected lifespan based on a given future operating condition spectrum. It provides crucial decision-making basis for predictive maintenance, enabling users to plan maintenance activities in advance and avoid sudden failures.

[0047] S5: By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated.

[0048] Establish a three-level decision matrix comprising a real-time status layer, a health assessment layer, and a prediction layer; The real-time status layer is divided into two states: successful locking and failed locking, based on the binary judgment result; the health assessment layer is divided into three levels: healthy, alert, and warning, based on the health status index; and the prediction layer is divided into three levels: long-term availability, medium-term maintenance, and immediate repair, based on the remaining service life prediction value. The first confidence level is used as the weighting coefficient of the decision results of each level in the three-level decision matrix, wherein the lower the first confidence level, the higher the weight ratio of the real-time state layer. Based on the weighted three-level decision matrix output, a predefined maintenance strategy rule base is matched to generate a comprehensive monitoring conclusion that includes maintenance level, recommended measures, and execution time. Specifically, when the binary determination result is locking failure, the highest level real-time alarm is immediately generated; when the health status index is at the warning level and the remaining service life prediction value is at the immediate maintenance level, a maintenance instruction for time-limited maintenance is generated. The comprehensive monitoring conclusions, along with the corresponding health status index and the predicted remaining service life, are packaged into a standardized maintenance work order and output to the maintenance management system.

[0049] It should be noted that the core of this step lies in establishing a multi-layered intelligent decision-making mechanism capable of comprehensively processing real-time status, medium-term health trends, and long-term lifespan prediction. First, a three-level decision matrix is ​​established, comprising a real-time status layer, a health assessment layer, and a prediction layer. This matrix structure reflects a comprehensive consideration from instantaneous safety to long-term planning: the real-time status layer directly reflects whether the equipment is currently in a safe and usable basic state based on binary judgment results, representing the bottom line for safe operation; the health assessment layer classifies the current performance status of the equipment based on a health status index, revealing the potential degree of performance degradation; and the prediction layer forecasts the future reliability of the equipment based on remaining service life predictions, providing a basis for maintenance resource planning.

[0050] The first confidence level is used as the weighting coefficient for the judgment results of each level in the three-level decision matrix. The lower the first confidence level, the higher the weight of the real-time state layer. The engineering logic is as follows: when the uncertainty of the sensing system is high (low first confidence level), it indicates a decrease in the reliability of long-term health assessment and lifespan prediction based on multi-source data fusion. At this time, the decision system will rely more on the most direct and critical real-time state judgment (i.e., the fundamental safety issue of whether the bottom door is successfully locked) to make decisions, which is a conservative and safe strategy. For example, even if the health status index shows "caution," if the first confidence level is very low, the system will reduce the weight of this assessment to avoid making inappropriate maintenance arrangements based on unreliable predictions, while prioritizing the monitoring and alarm of the real-time locking status.

[0051] Subsequently, based on the weighted three-level decision matrix output, a predefined maintenance strategy rule base is matched. This rule base defines specific action guidelines corresponding to different state combinations. The rule base contains clear priority logic: when the binary judgment result is locking failure, the highest level real-time alarm is immediately generated because this state directly threatens driving safety and requires immediate intervention; while for non-urgent but planned maintenance situations, such as when the health status index is at the warning level and the remaining service life prediction value is at the immediate repair level, indicating that the equipment is not only in poor current condition but also has a very short expected lifespan, the system will generate a maintenance instruction with a time limit, requiring the repair to be completed within the specified time window.

[0052] Finally, the comprehensive monitoring conclusions, along with the corresponding health status index and remaining service life prediction, are packaged into a standardized maintenance work order. This work order not only includes the conclusion of "what to do" but also provides data support for "why to do it," offering an integrated decision-making information package for the maintenance management system. This achieves closed-loop management from status monitoring to maintenance execution, significantly improving the accuracy, planning, and efficiency of maintenance work.

[0053] S6: Based on the analysis of the health status index and the predicted remaining useful life, data sensitivity information is obtained and fed back to the multi-source information fusion decision model.

[0054] Based on the unified virtual model, parameter sensitivity analysis is performed during the advanced simulation process to calculate the influence factors of the monitoring parameters corresponding to each data source on the health status index and the predicted value of the remaining service life. The data sources are ranked according to the magnitude of the influencing factors to generate data sensitivity information. The data sources with the greater impact on health status and lifespan prediction are assigned a higher sensitivity level. The data sensitivity information is fed back to the multi-source information fusion decision model in real time; In the subsequent fusion inference process, the weight allocation of each data source in the fusion calculation is dynamically adjusted according to the data sensitivity information, with data sources with high sensitivity levels being given higher fusion weights.

[0055] It should be noted that this step is the core closed-loop link for the self-optimization and continuous improvement of the system in this invention. Its function is to feed back the profound understanding of the back-end prediction model to the front-end perception fusion system, thereby forming a continuously evolving intelligent agent.

[0056] Specifically, the system performs parameter sensitivity analysis during advanced simulation based on the standardized virtual model. This analysis is conducted in a virtual space by introducing small, controllable numerical disturbances into each monitoring parameter (which directly originates from and corresponds to the aforementioned data sources, such as displacement sensor readings and strain gauge micro-strain values) within the standardized virtual model, while keeping all other conditions constant. The influence of these small disturbances on the two key prediction indicators—the final output health status index and the predicted remaining service life—is then observed and recorded. This influence is quantified as an influence factor, calculated as the ratio of the change in the prediction indicator to the disturbance magnitude of the monitoring parameter. A larger influence factor for a monitoring parameter indicates that even a slight change in that parameter can lead to a significant change in the health status or remaining service life, signifying that the parameter is a highly sensitive indicator of system performance degradation.

[0057] Subsequently, the system ranks the data sources based on the magnitude of the influencing factors. Since each monitoring parameter belongs to a specific data source (e.g., multiple strain gauge readings all belong to the "force / strain data" source), the system integrates the influencing factors of all monitoring parameters under the same data source (e.g., taking the average or maximum value) to obtain the overall influencing factor of that data source. Based on this, data sensitivity information is generated, with the core principle being that data sources with a greater impact on health status and lifespan prediction are assigned higher sensitivity levels. For example, analysis might reveal that the "locking contact force," a monitoring parameter derived from strain data, has a decisive impact on remaining lifespan prediction, thus giving the "force / strain data" data source the highest sensitivity level.

[0058] Subsequently, the system feeds back data sensitivity information to the multi-source information fusion decision model in real time. This feedback establishes an information bridge between the long-term prediction module and the real-time status diagnosis module.

[0059] Ultimately, in the subsequent fusion inference process, the weight allocation of each data source in the fusion calculation is dynamically adjusted based on data sensitivity information. The rule is: data sources with higher sensitivity levels are assigned higher fusion weights. This means that data that has been proven by digital twin simulation to be crucial to the long-term health of equipment will have their "voice" in real-time status determination systematically enhanced. For example, even if the instantaneous data quality flag of a high-sensitivity data source (such as strain data reflecting contact force) is not optimal, the system may still give it a higher weight than low-sensitivity data sources due to its long-term importance. This dynamic weight adjustment mechanism allows the entire monitoring system to no longer rely solely on the instantaneous quality of the data, but to intelligently and proactively allocate attention based on a deep understanding of the physical degradation mechanism of the system, thereby continuously optimizing its status perception and early fault identification capabilities, achieving a qualitative leap from "static fusion" to "cognitive-driven fusion".

[0060] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0062] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0063] Example 3, an embodiment of the present invention, provides an intelligent monitoring system for the locking status of the bottom door of a coal hopper car, including a data acquisition and preprocessing module, a multi-source information fusion decision-making module, a digital twin consistency module, an advanced simulation and health assessment module, a comprehensive decision-making and maintenance output module, and a sensitivity analysis and feedback module.

[0064] Data acquisition and preprocessing module: Collects multi-source monitoring data from the bottom door of the coal hopper car and preprocesses it to obtain a standardized data stream; Multi-source information fusion decision module: Inputs the standardized multi-source data stream into the preset multi-source information fusion decision model for fusion reasoning, outputs a binary judgment result, and simultaneously outputs a first confidence level characterizing the reliability of the binary judgment result; Digital twin consistency module: Construct a digital twin virtual model corresponding to the physical locking mechanism, take the binary judgment result and the first confidence level as the primary boundary conditions, and take the feature parameters in the standardized data stream as auxiliary driving data, and input them into the digital twin virtual model to perform online consistency processing to obtain a consistent virtual model; Advanced simulation and health assessment module: Utilizes the standardized virtual model to perform advanced simulation, calculates and outputs the health status index and the predicted value of remaining useful life; Integrated Decision and Maintenance Output Module: By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated. Sensitivity analysis and feedback module: Based on the health status index and the predicted remaining useful life, data sensitivity information is obtained and fed back to the multi-source information fusion decision model.

[0065] Example 4 is an embodiment of the present invention, which provides an intelligent monitoring method for the locking status of the bottom door of a coal hopper car. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0066] This embodiment tested the same model of funnel car bottom door locking mechanism, selecting three types of controls: traditional limit switch detection scheme (A), single Hall sensor scheme (B), and rule threshold fusion scheme (G); and four application scenarios of the present invention scheme: standard scenario (C), low temperature (D, around -22℃), high dust (E, 3.5–5.0 mg / m³), and 60 days after service wear (F); another set of the present invention scheme (H) with "sensitivity feedback" function enabled after 30 days of operation. All tests were completed on the same line segment, with vehicle speed of 46–53 km / h and load of approximately 62 t, and environmental temperature, humidity, and dust were recorded according to natural working conditions. On the hardware side, displacement / angle, force / strain, magnetic flux / inductance, pneumatic pressure / flow / valve position, motor current / voltage, point temperature, and heat flux were sampled simultaneously, with an action time window of 200–500 Hz and a static time window of 5–10 Hz; S1 preprocessing unified the time base and generated quality flag bits, and distortion and saturation segments did not participate in subsequent fusion. In stage S2, a fusion decision based on DS evidence theory is executed in the edge computing unit: using "locked / unlocked" as the identification framework, basic probability assignments are established for features from different sources, and dynamic weights are mapped by quality flag bits; after Dempster combination, the confidence level is taken as a conservative intermediate probability, compared with the dynamic threshold to produce a binary decision, and the first confidence level is calculated by "confidence-likelihood interval width" and "threshold distance". In stage S3, a digital twin model containing geometry, material properties and multibody dynamics is constructed, using the binary decision as the initial state and boundary constraints, and the first confidence level modulates the parameters to calibrate the intensity; displacement / gap / angle are used as geometric drivers, and force / strain and pressure / flow / current / voltage are used as dynamic observations. After the online consistency index meets the standards (RMSE, correlation coefficient, etc.), a consistent virtual model is output. In stage S4, the load spectrum and environmental sequence of the next month are injected into the consistent model, and the trajectory of "contact force peak, arrival time, and transmission efficiency" is extracted by advanced simulation, a health status index is synthesized and the RUL is calculated according to the failure criteria. S5 incorporates real-time judgment, health index, RUL, and first confidence level into a three-level decision matrix, combining it with a rule base to output maintenance level, recommended measures, and timeliness, encapsulating them into a work order; S6 utilizes advanced simulation sensitivity analysis to generate data sensitivity information and feeds it back into the fusion model, adjusting weights with a smoothing strategy. Data from each group is statistically analyzed within the same batch of 10 loading / unloading cycles to ensure comparability; the table presents the measurement / calculation results for representative sections.

[0067] The experimental data recording table is shown in Table 1.

[0068] Table 1: Experimental Data Recording Table

[0069] Comparing the three control schemes (A, B, and G) with the five schemes of this invention (C, D, E, F, and H), three core advantages can be observed: judgment reliability, health / lifespan prediction quality, and the specificity of strategy output. First, at the real-time judgment level, the "first confidence level" of the standard scenario (C) of this invention reaches 0.93, which is about 35% higher than that of a single Hall effect (B, 0.69) and about 26% higher than that of rule threshold fusion (G, 0.74). Even under low temperature (D, -21.7℃) and high dust (E, 4.8 mg / m³) conditions, the confidence level remains at 0.88 and 0.90, respectively, without significant degradation, indicating that the dynamic weighting and conservative decision strategy driven by the quality flag of S2 effectively suppresses the influence of noise in complex working conditions. The misclassification rate also confirms this: C is 0.7%, and D / E are 1.1% and 0.9% respectively, both significantly lower than A (4.8%) and G (2.7%), reflecting the robustness of multi-source fusion and the stability benefits brought by DS conflict handling.

[0070] Secondly, in terms of health and lifespan, C's health status index is 93.6%, and its relative ulterior radii (RUL) is 712 h; H (after enabling sensitivity feedback) further improves to 94.2% and 736 h, respectively, an improvement of approximately 10.1 and 188 hours compared to G (84.1%, 548 h). This indicates that S3's online uniformization, using binary judgment and confidence level as boundary and calibration strength control quantities, effectively suppresses parameter divergence; S4's advanced simulation outputs a degradation trajectory closer to reality on the uniformized model, significantly extending the RUL estimate and reducing fluctuations. In the scenario of 60 days after wear (F), the health index is 88.3%, and the RUL is 592 h. Although slightly lower than C, it still has a significant advantage over B (82.7%, 506 h), demonstrating the invention's sensitive capture capability of degradation in the later stages of lifespan. This capability stems from the discriminative power of multi-domain features (peak contact force, arrival time, transmission efficiency) on the degradation mode, as well as the physical consistency of the model parameters after uniformized calibration.

[0071] Thirdly, in terms of maintenance strategy and operational value, the early warning time for faults is particularly crucial for predictive maintenance. The warning time for C is 24.7 min, and for H it is 27.4 min, significantly better than A (6.5 min) and G (10.2 min). This longer lead time is attributed to the closed loop of "index / RUL → rule base" in S4-S5: when RUL approaches the threshold and the health index declines, the rule base can still provide an "on-demand maintenance" strategy even if the real-time layer is "successful," avoiding unplanned downtime; and when the combination of health = "warning" and RUL = "immediate repair" occurs (more common in scenario F), the system generates "planned / time-limited maintenance," fundamentally different from the traditional "on-time is qualified" criterion. Furthermore, D and E maintained a confidence level of 0.88–0.90 and an early warning time of 21.9–22.8 min even under extreme conditions, indicating that after environmental factors were involved in threshold contextualization, the consistency between the decision of S2 and S3 was not significantly disturbed by low temperature or dust, thus ensuring the engineering adaptability of the solution.

[0072] In summary, the data shows that the innovation of this invention mainly lies in the closed-loop design of "perception-fusion-consistency-prediction-decision-feedback". On the one hand, it reduces misjudgments under complex operating conditions through DS fusion driven by quality flags; on the other hand, it makes the predicted health index and RUL closer to the actual degradation by consistent modeling with "binary judgment + confidence" as the boundary and calibration intensity; finally, it achieves on-demand maintenance in the rule base using a three-level matrix with confidence weighting. Compared with the single-sensor or static threshold fusion of existing technologies, this embodiment significantly outperforms the control in four key indicators: reliability, misjudgment rate, RUL, and warning time, demonstrating the inventiveness and engineering advantages of the invention.

[0073] It should be noted that 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent monitoring of the locking status of the bottom door of a coal hopper car, characterized in that, include: Multi-source monitoring data from the bottom door of the coal hopper car were collected and preprocessed to obtain a standardized data stream; The standardized multi-source data stream is input into a preset multi-source information fusion decision model for fusion inference, and a binary judgment result is output. At the same time, a first confidence level characterizing the reliability of the binary judgment result is also output. A digital twin virtual model corresponding to the physical locking mechanism is constructed. The binary judgment result and the first confidence level are used as the primary boundary conditions, and the feature parameters in the standardized data stream are used as auxiliary driving data. They are all input into the digital twin virtual model, and online consistency processing is performed to obtain a consistent virtual model. The standardized virtual model is used to perform advanced simulation, calculate and output the health status index and the predicted value of remaining useful life; By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated. Data sensitivity information is obtained based on the analysis of the health status index and the predicted remaining useful life, and then fed back to the multi-source information fusion decision model.

2. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 1, characterized in that, The multi-source monitoring data includes force and strain data reflecting the locking stress state; Displacement, clearance, and angle data reflecting the relative state of the door; Magnetic field, magnetic flux, and inductance data reflecting the state of the locked magnetic circuit; Actuation process data characterizing the operating conditions of the actuator include: pressure, stroke and valve position, flow rate, drive current and voltage; Thermal data reflecting the effects of sealing and icing include: temperature, heat flux, and heat distribution information; Environmental and operating parameters include: ambient temperature, humidity, icing and snow accumulation indicators, dust concentration, vehicle speed, and load information.

3. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 2, characterized in that, The preprocessing includes performing filtering and noise reduction and timestamp synchronization and alignment operations on the time-series data in the multi-source monitoring data. Image enhancement and distortion correction operations are performed on the heat distribution information in the multi-source monitoring data; A quality assessment operation is performed on the multi-source monitoring data, and a data quality flag is generated for each data source according to predefined rules; Based on the data quality flag, invalid data is removed, and dimensional normalization is performed on valid numerical data to form the standardized data stream.

4. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 3, characterized in that, The preset multi-source information fusion decision model is a probabilistic reasoning model built on DS evidence theory; The fusion reasoning includes: transforming each data source in the standardized data stream into an independent evidence source, and constructing a basic probability assignment for each evidence source for the {locked, unlocked} proposition; Calculate the dynamic weight of the evidence source corresponding to each data source based on the data quality flag bit. The basic probability allocation of the corresponding evidence sources is corrected using the dynamic weights, and the Dempster combination rule is used to synthesize all the corrected evidence sources to obtain the reliability and similarity of the synthesized proposition. The confidence level is used as an intermediate probability value and compared with a preset judgment threshold to generate the binary judgment result; The first confidence level is obtained by weighted calculation based on the interval width formed by confidence and similarity, and the relative distance between the intermediate probability value and the judgment threshold.

5. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 4, characterized in that, The online consistency processing includes constructing a digital twin virtual model that includes the geometry, material physical properties, and multibody dynamics characteristics of the physical locking mechanism; The binary determination result is used as the initial state constraint condition of the digital twin virtual model; The first confidence level is used as a control factor for the calibration intensity of model parameters; The displacement, gap, and angle data in the standardized data stream are used as the geometric motion driving input for the digital twin virtual model; Force and strain data, as well as pressure, flow rate, driving current and voltage data from the standardized data stream, are used as dynamic state observation inputs for the digital twin virtual model; The dynamic response of the calculation model is verified online with the actual state of the physical mechanism. When the response deviation exceeds the preset threshold, the friction coefficient and connection stiffness parameters in the model are adaptively corrected based on the control factor of the first confidence level. When the response deviation drops to within the preset threshold, the digital twin virtual model is determined to have completed online consistency processing, and the model with the corrected parameters at this time is output as the consistent virtual model.

6. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 5, characterized in that, The calculation and output of the health status index and remaining service life prediction value includes setting a preset future operating condition that includes load spectrum, operating frequency and environmental conditions; The uniform virtual model is driven to perform advanced numerical simulation under the preset future operating conditions to simulate the performance evolution of the locking mechanism within a preset time range; The key performance parameters representing the health status of the mechanism are extracted from the simulation process. These key performance parameters include the peak locking contact force, locking time, and mechanism transmission efficiency. Based on the degradation trajectory of the key performance parameters, the percentage of performance degradation at the current moment relative to the initial healthy state is calculated to obtain the health state index. The simulation data of key performance parameters are compared with the preset failure threshold. The simulation time point when the performance parameter first exceeds the failure threshold is taken as the end point of the lifespan. The time from the current moment to the end point of the lifespan is calculated to obtain the predicted value of the remaining lifespan.

7. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 6, characterized in that, The generation of comprehensive monitoring conclusions that include maintenance strategies includes establishing a three-level decision matrix that includes a real-time status layer, a health assessment layer, and a prediction layer. The real-time status layer is divided into two states: successful locking and failed locking, based on the binary judgment result; the health assessment layer is divided into three levels: healthy, alert, and warning, based on the health status index; and the prediction layer is divided into three levels: long-term availability, medium-term maintenance, and immediate repair, based on the remaining service life prediction value. The first confidence level is used as the weighting coefficient for the decision results at each level in the three-level decision matrix; Based on the weighted three-level decision matrix output, a predefined maintenance strategy rule base is matched to generate a comprehensive monitoring conclusion that includes maintenance level, recommended measures, and execution time. Specifically, when the binary determination result is locking failure, the highest level real-time alarm is immediately generated; when the health status index is at the warning level and the remaining service life prediction value is at the immediate maintenance level, a maintenance instruction for time-limited maintenance is generated. The comprehensive monitoring conclusions, along with the corresponding health status index and the predicted remaining service life, are packaged into a standardized maintenance work order and output to the maintenance management system.

8. The intelligent monitoring method for the locking status of the bottom door of the coal hopper car as described in claim 7, characterized in that, The data sensitivity information obtained from the analysis includes, based on the consistent virtual model, performing parameter sensitivity analysis during the advanced simulation process, and calculating the influence factors of the monitoring parameters corresponding to each data source on the health status index and the predicted value of the remaining service life; Based on the magnitude of the impact factor, the data sources are ranked according to their sensitivity to generate data sensitivity information. The data sensitivity information is fed back to the multi-source information fusion decision model in real time; In the subsequent fusion inference process, the weight allocation of each data source in the fusion calculation is dynamically adjusted based on data sensitivity information.

9. An intelligent monitoring system for the locking status of the bottom door of a coal hopper car, used to implement the intelligent monitoring method for the locking status of the bottom door of a coal hopper car as described in any one of claims 1 to 8, characterized in that, include: Data acquisition and preprocessing module: Collects multi-source monitoring data from the bottom door of the coal hopper car and preprocesses it to obtain a standardized data stream; Multi-source information fusion decision module: Inputs the standardized multi-source data stream into the preset multi-source information fusion decision model for fusion reasoning, outputs a binary judgment result, and simultaneously outputs a first confidence level characterizing the reliability of the binary judgment result; Digital twin consistency module: Construct a digital twin virtual model corresponding to the physical locking mechanism, take the binary judgment result and the first confidence level as the primary boundary conditions, and take the feature parameters in the standardized data stream as auxiliary driving data, and input them into the digital twin virtual model to perform online consistency processing to obtain a consistent virtual model; Advanced simulation and health assessment module: Utilizes the standardized virtual model to perform advanced simulation, calculates and outputs the health status index and the predicted value of remaining useful life; Integrated Decision and Maintenance Output Module: By combining the binary judgment result, the first confidence level, the health status index, and the remaining service life prediction value, a comprehensive monitoring conclusion including maintenance strategies is generated. Sensitivity analysis and feedback module: Based on the health status index and the predicted remaining useful life, data sensitivity information is obtained and fed back to the multi-source information fusion decision model.

Images