Container condition monitoring and life prediction system and method

Abstract

The application discloses a container state monitoring and life prediction system and method. The system comprises: a container state acquisition module acquires structural response data and environmental load data of a first container; a digital twin model management module extracts features from the structural response data and the environmental load data to obtain multi-source feature data; a first digital twin model of the first container is constructed according to the multi-source feature data, damage evolution analysis of the first container is performed to obtain damage evolution analysis results; life prediction is performed according to the damage evolution analysis results to obtain life prediction results of the first container; and an operation and maintenance management module generates operation and maintenance tasks of the first container according to the life prediction results. The technical scheme of the application solves the problem that the reliability of current life prediction needs to be improved, can improve the accuracy and reliability of container life prediction through fusion processing of multi-source data, and improves the timeliness of operation and maintenance task generation.

Description

Technical Field

[0001] The present invention relates to the field of container technology, and in particular to a container condition monitoring and life prediction system and method. Background Technology

[0002] Currently, existing life prediction models cannot effectively integrate multi-source heterogeneous data, resulting in significant uncertainty in the prediction results, which in turn affects the safe use and timely maintenance of containers. Summary of the Invention

[0003] This invention provides a container condition monitoring and life prediction system and method to address the current problem of insufficient reliability in life prediction. By fusing multi-source data, the accuracy and reliability of container life prediction can be improved, as well as the timeliness of maintenance task generation.

[0004] In a first aspect, embodiments of the present invention provide a container condition monitoring and life prediction system, the system comprising: The container status acquisition module includes at least one sensor group, which is installed on the first container and is used to acquire the structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data and vibration data, and the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data and impact load data. The container status acquisition module also includes an edge computing unit, which is used to preprocess structural response data and environmental load data, and send the preprocessed structural response data and environmental load data to the digital twin model management module. The digital twin model management module is used to extract features from structural response data and environmental load data to obtain multi-source feature data; A first digital twin model of the first container is constructed based on multi-source feature data. Damage evolution analysis of the first container is then performed based on the first digital twin model to obtain the damage evolution analysis results. Based on the damage evolution analysis results, the life prediction is performed to obtain the life prediction result of the first container, and the life prediction result is sent to the operation and maintenance management module. The operation and maintenance management module is used to generate operation and maintenance tasks for the first container based on the life prediction results.

[0005] Secondly, embodiments of the present invention also provide a method for container condition monitoring and life prediction, the method comprising: The container status acquisition module collects structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data and vibration data, and the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data and impact load data. The container status acquisition module includes at least one sensor group, which is installed on the first container. The edge computing unit of the container status acquisition module preprocesses the structural response data and environmental load data, and sends the preprocessed structural response data and environmental load data to the digital twin model management module. The first feature vector is obtained by extracting features from the structural response data and environmental load data through the digital twin model management module. A first digital twin model of the first container is constructed based on the first feature vector. Damage evolution analysis is performed on the first container based on the first digital twin model to obtain the damage evolution analysis results. Based on the damage evolution analysis results, the life prediction is performed to obtain the life prediction result of the first container, and the life prediction result is sent to the operation and maintenance management module. The operation and maintenance management module generates the operation and maintenance task for the first container based on the life prediction results.

[0006] In this embodiment of the invention, a container status acquisition module includes at least one sensor group installed on a first container. The sensor group is used to acquire structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data, and vibration data. The environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data, and impact load data. The container status acquisition module also includes an edge computing unit, which preprocesses the structural response data and environmental load data and sends the preprocessed data to a digital twin model management module. The digital twin model management module extracts features from the structural response data and environmental load data to obtain multi-source feature data. Based on the multi-source feature data, a first digital twin model of the first container is constructed. Damage evolution analysis is performed on the first container based on the first digital twin model to obtain damage evolution analysis results. Lifetime prediction is performed based on the damage evolution analysis results to obtain the lifetime prediction results of the first container, and these results are sent to the operation and maintenance management module. The operation and maintenance management module generates operation and maintenance tasks for the first container based on the lifetime prediction results. The technical solution of this invention addresses the problem of insufficient reliability in current lifespan prediction. It can improve the accuracy and reliability of container lifespan prediction and enhance the timeliness of maintenance task generation through the fusion processing of multi-source data. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the structure of a container condition monitoring and life prediction system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of another container condition monitoring and life prediction system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of another container condition monitoring and life prediction system provided in an embodiment of the present invention; Figure 4 A flowchart of a container condition monitoring and life prediction method provided in an embodiment of the present invention. Detailed Implementation

[0009] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0010] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. The acquisition, storage, use, and processing of data in the technical solutions of this application all comply with relevant laws and regulations.

[0011] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the relevant content of the solution.

[0012] Figure 1 This is a schematic diagram of a container condition monitoring and life prediction system provided in an embodiment of the present invention. This embodiment can be applied to scenarios involving container condition monitoring and life prediction.

[0013] like Figure 1As shown, the container status monitoring and life prediction system includes: a container status acquisition module 110, a digital twin model management module 120, and an operation and maintenance management module 130.

[0014] The container status acquisition module 110 includes at least one sensor group, which is installed on the first container and is used to acquire the structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data and vibration data, and the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data and impact load data.

[0015] The container status acquisition module 110 also includes an edge computing unit, which is used to preprocess the structural response data and environmental load data, and send the preprocessed structural response data and environmental load data to the digital twin model management module.

[0016] The digital twin model management module 120 is used to extract features from structural response data and environmental load data to obtain multi-source feature data; construct a first digital twin model of the first container based on the multi-source feature data; perform damage evolution analysis on the first container based on the first digital twin model to obtain damage evolution analysis results; perform life prediction based on the damage evolution analysis results to obtain the life prediction results of the first container, and send the life prediction results to the operation and maintenance management module.

[0017] The operation and maintenance management module 130 is used to generate operation and maintenance tasks for the first container based on the life prediction results.

[0018] Specifically, the sensor group includes a structural response sensor group and an environmental load sensor group, which are used to collect structural response data and environmental load data, respectively.

[0019] The structural response sensor array is installed at key structural locations of the first container, specifically at critical load-bearing areas as specified in ISO 668, including the four corners of the container, the mid-span of the floor, the middle of the side walls, and the top longitudinal beams. This ensures comprehensive data collection of the container's structural response under various operating conditions. The array is used to collect strain, acceleration, and vibration data of the first container during loading, unloading, transportation, and storage. Specifically, the structural response sensor array includes multiple strain gauge sensors, accelerometers, and vibration sensors. The strain gauge sensors utilize the principle of resistance strain gauges, measuring resistance changes at key structural locations to reflect strain magnitude. The accelerometers employ MEMS triaxial accelerometers to measure the container's acceleration response in three orthogonal directions. The vibration sensors use piezoelectric sensors to measure the container's vibration frequency and amplitude.

[0020] An environmental load sensor array is installed on the surface and inside the first container to collect data on ambient temperature, ambient humidity, corrosive medium concentration, and impact load. Specifically, the environmental load sensor array includes temperature sensors (such as thermocouples or thermistors), humidity sensors (such as capacitive humidity sensors), corrosive gas sensors (such as electrochemical sensors) to measure the concentration of corrosive gases such as sulfur dioxide and hydrogen chloride, and impact sensors (such as barometric or photoelectric sensors) to measure the magnitude and duration of the impact load. The environmental load sensor array is evenly distributed on the surface and inside the first container to ensure accurate reflection of the load conditions of the environment in which the first container is located.

[0021] The edge computing unit is located inside the first container and is used to perform preliminary filtering, noise reduction and feature extraction on the raw data collected by the structural response sensor group and the environmental load sensor group, namely the structural response data and the environmental load data, to obtain preprocessed sensing data. The preprocessed sensing data is then sent to the digital twin platform, namely the digital twin model management module 120, through the wireless communication network.

[0022] Specifically, the edge computing module includes a signal conditioning circuit, an A / D converter, a microprocessor, and a wireless communication module. The signal conditioning circuit amplifies, filters, and linearizes the analog signals output by the sensors. The A / D converter converts the analog signals into digital signals. The microprocessor runs lightweight filtering algorithms such as Kalman filtering and feature extraction algorithms to extract time-domain features such as peak value, mean, and variance, and frequency-domain features such as main frequency and spectral energy to process the digital signals, achieving preliminary data compression and dimensionality reduction. The edge computing module uses an embedded processor with a power consumption of less than 100mW, enabling it to operate independently inside the container for extended periods.

[0023] The raw data is a time series data set formed by continuous acquisition according to a preset time window; specifically, the structural response sensor group and the environmental load sensor group acquire continuous time series data according to a preset sampling frequency, and divide it into segments according to a fixed time window, such as 1 minute, 10 minutes or 1 hour as a data window. Each data window contains multi-dimensional data such as strain, acceleration, vibration, temperature and humidity and corrosion environment parameters within that time period.

[0024] The digital twin model management module 120 performs data cleaning, missing value imputation, and feature extraction on the structural response data and environmental load data sent by the edge computing unit to obtain multi-source feature data. Based on the multi-source feature data, a first digital twin model of the first container is constructed. Damage evolution analysis is performed on the first container based on the first digital twin model and damage principles to obtain damage evolution analysis results for the associated areas of each sensor group. Based on the damage evolution analysis results and the relationship between damage and remaining lifespan, lifespan prediction is performed to obtain lifespan prediction results for the associated areas of each sensor group of the first container, and the lifespan prediction results are sent to the operation and maintenance management module 130.

[0025] The operation and maintenance management module 130 can calculate the confidence level of the life prediction results based on historical prediction data, and generate operation and maintenance tasks based on the confidence level and life prediction results and push them to the operation and maintenance personnel.

[0026] The technical solution of this embodiment includes a container status acquisition module comprising at least one sensor group mounted on a first container. The sensor group is used to acquire structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data, and vibration data, while the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data, and impact load data. The container status acquisition module also includes an edge computing unit, which preprocesses the structural response data and environmental load data and sends the preprocessed data to a digital twin model management module. The digital twin model management module extracts features from the structural response data and environmental load data to obtain multi-source feature data. Based on the multi-source feature data, it constructs a first digital twin model of the first container, performs damage evolution analysis on the first container based on the first digital twin model, and obtains damage evolution analysis results. Based on the damage evolution analysis results, it performs life prediction to obtain the life prediction results of the first container and sends the life prediction results to the operation and maintenance management module. The operation and maintenance management module generates operation and maintenance tasks for the first container based on the life prediction results. The technical solution of this invention addresses the problem of insufficient reliability in current lifespan prediction. It can improve the accuracy and reliability of container lifespan prediction and enhance the timeliness of maintenance task generation through the fusion processing of multi-source data.

[0027] Figure 2 This is a schematic diagram of a container condition monitoring and life prediction system provided in an embodiment of the present invention. This embodiment can be applied to scenarios involving container condition monitoring and life prediction.

[0028] like Figure 2As shown, the container condition monitoring and life prediction system includes: a container condition acquisition module 210, a digital twin model management module 220, and an operation and maintenance management module 230; wherein, the digital twin model management module 220 includes: The data cleaning unit 221 is used to remove outliers, fill in missing values, and perform timestamp calibration and synchronization processing on the structural response data and environmental load data to obtain data cleaning results; the feature extraction unit 222 is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the data cleaning results to obtain structural response features, environmental load features, and historical operation features; the data alignment unit 223 is used to align the structural response features, environmental load features, and historical operation features in terms of time dimension and feature dimension to obtain multi-source feature data. The container health status prediction unit 224 is used to input multi-source feature data into the container health status prediction sub-model of the first digital twin model to obtain the health status of the first container; wherein, the container health status prediction sub-model is a random forest model, which is trained by multi-source feature sample data and the real health status labels corresponding to the multi-source feature sample data.

[0029] The data cleaning unit 221 is used to identify and remove abnormal data in the received structural response data and environmental load data, interpolate and complete missing data, and calibrate and synchronize data timestamps.

[0030] Specifically, the data cleaning unit 221 uses a statistical anomaly detection algorithm to calculate the mean and standard deviation of each sensor data, and judges data exceeding three times the standard deviation as abnormal data and removes them; for missing data, linear interpolation or spline interpolation methods are used to complete it; and a time synchronization algorithm is used to calibrate the timestamps of each sensor to unify the sampling time of different sensors to a standard time reference.

[0031] The feature extraction unit 222 is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the cleaned data to obtain structural response features, environmental load features, and historical operation features. Specifically, the feature extraction unit is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the cleaned data to obtain structural response features, environmental load features, and historical operation features. Specifically, the time-domain features may include: Peak value: Mean effective value Peak factor ; Pulse factor Margin factor: .

[0032] Frequency domain features include: obtaining the spectrum through FFT transform, taking the dominant frequency as the frequency point corresponding to the maximum value of the spectrum, and the centroid of the spectrum according to... Calculation, where The spectral amplitude is the spectral energy distribution, which is the ratio of the energy of each frequency band to the total energy.

[0033] The time-frequency domain features are decomposed into three levels using Daubechies-4 wavelets, and the approximation coefficients and detail coefficients of each level are extracted as feature vectors.

[0034] Data alignment unit 223 is used to align structural response features, environmental load features, and historical operation features with different sampling rates and data formats in terms of both time dimension and feature dimension, resulting in aligned multi-source heterogeneous fused data, i.e., multi-source feature data. Specifically, time dimension alignment uses cubic spline interpolation to interpolate high-frequency sampling data, such as acceleration at 1000Hz and vibration at 2000Hz, to low-frequency sampling time points, such as temperature and humidity at 1Hz. The interpolation formula is as follows: Interpolation nodes are obtained by solving a system of tridiagonal matrix equations; before interpolation, a Butterworth low-pass filter with a cutoff frequency of 0.8 times the Nyquist frequency is applied to the high-frequency data for anti-aliasing preprocessing; feature dimension alignment is achieved using... Standardization, the formula is ,in The characteristic mean, The characteristic standard deviation is denoted as .

[0035] The digital twin model management module 220 constructs a digital twin model of the first container based on multi-source feature data through adaptive dynamic modeling, namely the fused structural feature vector, environmental feature vector, and temporal feature vector, and updates the model parameters of the digital twin model in real time, so that the digital twin model keeps synchronized with the state of the first container.

[0036] The health status predicted by the container health status prediction sub-model can be used in subsequent damage evolution analysis to correct the cumulative damage index.

[0037] The initial training process of the container health status prediction sub-model can be as follows: train an initial random forest model based on a historical dataset of no less than 10,000 labeled samples. The random forest model is used to establish a nonlinear mapping relationship between multi-source environmental load and structural response data and container health status. It contains T = 100 decision trees, with a maximum depth of 20 for each tree and a minimum number of leaf node samples of 5.

[0038] In an optional implementation, the container health status prediction unit 224 further includes: The incremental update subunit is used to select historical multi-source feature sample data and new multi-source feature sample data as incremental training set according to a preset ratio. For each decision tree in the container health status prediction sub-model, it traverses each first sample in the incremental training set, starting from the root node of the tree, and assigns the first sample to the left child node or the right child node according to the node splitting feature and threshold. When the sample assignment reaches the leaf node, the category count or regression value of the leaf node is updated. When the number of first samples assigned to the leaf node exceeds the preset sample number threshold, the node split is triggered, and the feature with the largest decrease in Gini coefficient is selected as the splitting feature to complete the incremental update of the container health status prediction sub-model.

[0039] For example, whenever M=100 new multi-source feature samples are received, an incremental update is triggered; the new data samples are mixed with the random sampled data in the historical data buffer at a ratio of 7:3 to form the incremental training set, and the historical data buffer has a capacity of B = 1000 records, which are updated using a first-in-first-out strategy.

[0040] In this embodiment, each sample corresponds to a comprehensive description of the container status within a time window; for example, taking a 10-minute sampling window as an example, the sample contains feature vectors obtained by cleaning, synchronizing and extracting features from all data collected by sensors within the 10 minutes, as well as structural health status labels at the corresponding time points.

[0041] For each decision tree in the random forest, traverse each sample in the incremental training set, starting from the root node of the tree, and assign the sample to the left or right child node according to the node splitting feature and threshold; when reaching a leaf node, update the class count or regression value of that leaf node using the incremental average update formula. When the number of samples in a leaf node exceeds a preset threshold When a node split is triggered, the feature with the largest decrease in the Gini coefficient is selected as the splitting feature.

[0042] In incremental updates, an elastic weight fixation method can also be used to apply a large regularization constraint to the parameters in the model that are related to historical data, with the regularization coefficient... Apply a small regularization constraint to the parameters related to the new data; the regularization coefficient... To achieve a balance between new and old knowledge.

[0043] In one alternative implementation, the digital twin model management module 220 includes: The model parameter update unit is used to estimate the model parameters of the digital twin model based on multi-source feature data using the recursive least squares method or the extended Kalman filter algorithm, and obtain updated model parameters; wherein the model parameters include at least structural state parameters and degradation parameters.

[0044] The model parameter update unit employs recursive least squares or extended Kalman filtering algorithms to estimate and update the model parameters of the digital twin model online. These parameters include structural state parameters and degradation parameters, such as material elastic modulus correction coefficients, structural stiffness attenuation coefficients, damage growth rate parameters, and sensor observation error parameters. The material elastic modulus correction coefficient can be used to correct the elastic modulus E of the container body material in damage evolution analysis, thus improving the calculated stress. The values ​​are more accurate; the structural stiffness attenuation coefficient is used to correct the stiffness matrix of the digital twin model / finite element model; the damage growth rate parameter corresponds to dD / dt in the life prediction formula; the sensor observation error parameter corresponds to the observation noise V(k) or observation noise covariance R in the extended Kalman filter.

[0045] Because containers are subjected to fatigue loads, impact loads, and environmental corrosion during long-term transportation, their actual structural state changes over service time. Therefore, the initially established digital twin model cannot maintain consistency with the actual state in the long term. By using recursive least squares or extended Kalman filtering algorithms, the model parameters are corrected online based on real-time collected data, enabling the digital twin model to dynamically reflect the actual container state. The digital twin model itself describes the physical entity's state, while the life prediction model infers future degradation trends based on the current state; their functions are different. The digital twin model improves the accuracy of state description through online parameter updates, and the updated state parameters are then input into the life prediction module to achieve a more accurate remaining life prediction. Specifically, when the digital twin model adopts the recursive least squares form, the model update unit updates the model parameters according to the following recursive formula: ; ; ; in, This represents the estimated model parameters at time k. This represents the observation value at time k. This represents the eigenvector at time k. Represents the gain matrix. Represents the covariance matrix. Forgetting factor, range of values Typical value 0.98; I is the identity matrix; initial parameters Take the offline calibration results and the initial covariance matrix. Set as a diagonal matrix, with diagonal elements set to 1.0. When using extended Kalman filtering, the state-space expression of the digital twin model needs to be established: State Equation Observation equation ,in It is a nonlinear state transition function. Let W be the nonlinear observation function, and let V be the process noise and observation noise, respectively. We assume that the noise is zero-mean Gaussian white noise, and the covariance matrices are Q and R, respectively.

[0046] The digital twin model may also include a state synchronization unit, which is used to calculate the state deviation between the digital twin model and the first container, and to correct the digital twin model according to the state deviation, so that the state of the digital twin model and the first container are kept synchronized.

[0047] Define the state deviation vector ,in For stress prediction deviation, To account for strain prediction bias, For acceleration prediction bias; These are the stress, strain, and acceleration values ​​actually measured for the first container at time t. These are the predicted values ​​of the digital twin model at the corresponding time points.

[0048] When the Euclidean norm of the state deviation Time-triggered model correction, where the threshold The typical value is determined based on three times the standard deviation of the sensor's measurement accuracy. .

[0049] Correction is performed using a Kalman filter framework: the state deviation is used as the observed value, and the Kalman gain is applied... Update the model state with the optimal estimate: ; in Here, H represents the model state estimate, H is the observation matrix, and Kalman gain is... P is the estimation error covariance matrix, and R is the observation noise covariance matrix.

[0050] In one alternative implementation, the digital twin model management module 220 includes: The damage evolution analysis unit is used to calculate the structural damage index of the first container for each load cycle based on the strain data, material ultimate strength, load cycle number and material constant in the structural response data of the first digital twin model. The load cycle number is determined by the rainflow counting method. The cumulative damage index is obtained by summing all structural damage indices.

[0051] The structural damage index is calculated using the following formula: ; Where D represents the structural damage index, This represents the stress value during the i-th load cycle. The ultimate strength of the material is represented by n, the number of load cycles is represented by C, and the material constants are represented by m. The determination is achieved through the following steps: Step 1: Conduct accelerated fatigue-corrosion coupling test according to GB / T 15248-2017 "Metallic Materials Axial Constant Amplitude Low Cyclic Fatigue Test Method".

[0052] Step 2: Measure different stress amplitudes The corresponding fatigue life N is plotted in logarithmic coordinates. curve.

[0053] Step 3: Fit the curve using the least squares linear regression method to obtain the slope (1 / m) and intercept log(C).

[0054] Step four, for structural steel, , For aluminum alloys, , .

[0055] The damage evolution model submodule also considers the impact of environmental factors on damage, introducing an environmental correction coefficient. This reflects the accelerating effect of temperature, humidity, and corrosive media on the damage rate.

[0056] stress value The strain data was determined using the following method: According to the constitutive relation of materials Calculate the dynamic stress time history, where E is the elastic modulus of the container body material (for structural steel, take...). Aluminum alloy The dynamic stress time history was analyzed using the rainflow counting method to perform cyclic counting, and the stress amplitude of the i-th load cycle was extracted. And its corresponding number of cycles n, where the number of load cycles n is the total number of cumulative cycles obtained by the rainflow counting method.

[0057] The specific steps of the rainflow counting method are as follows: Step 1: Calculate the stress time history Peak and valley values ​​are extracted to obtain alternating peak and valley sequences.

[0058] Step 2: Rearrange the peak-valley sequence, starting with the maximum peak or minimum valley value.

[0059] Step 3: Traverse the peak-valley value sequence. When three consecutive points satisfy... At that time, extract a complete loop with a loop amplitude of .

[0060] Step four: Repeat step three until all peak and valley values ​​have been extracted, and then calculate the number of cycles n corresponding to each stress amplitude.

[0061] In this embodiment, the structural damage index is calculated to achieve a quantitative assessment of container structural damage. Based on structural feature vectors and environmental feature vectors, and taking into account both load cycle effects and environmental degradation effects, the structural damage index is calculated, and the cumulative effect of the damage index is used to predict the trend of structural strength degradation. This damage evolution model can identify the early stages of microscopic damage, providing a scientific basis for preventive maintenance and eliminating the drawbacks of the passive maintenance mode in the prior art, which can only be maintained after visible damage occurs.

[0062] In one alternative implementation, the digital twin model management module 220 includes: The life prediction module is used to calculate the remaining life prediction result of the first container based on the cumulative damage index, critical damage index, preset safety factor and damage rate; wherein, the damage rate is obtained by linear regression fitting of the structural damage index at the most recent preset number of time points to time.

[0063] The current cumulative damage index is calculated based on the damage evolution model. Finally, the remaining useful life is predicted based on the following formula: ; RUL represents the remaining useful life; This represents the safety factor, and its value range is... The settings are determined by the operation and maintenance management personnel based on the container's service environment and importance; This indicates the current cumulative damage index. Indicates the critical damage index; Indicates damage rate; critical damage index The determination method is as follows: Accelerated fatigue testing is conducted according to GB / T 15248-2017 "Metallic Materials - Axial Constant Amplitude Low Cyclic Fatigue Test Method". The cumulative damage index corresponding to the reduction of the material's yield strength to 80% of its initial value is the critical damage index. For structural steel, The fatigue damage threshold corresponds to a 20% reduction in yield strength; for aluminum alloys, ; The damage index at the most recent N time points was obtained by performing a linear regression fit on time, where N takes a value of 1. .

[0064] Rolling predictions can be performed according to preset time windows, such as 30 days, 90 days, and 180 days, to obtain the remaining service life under different time windows. The time window is used for the data update cycle of the prediction model, rather than limiting the remaining service life calculation cycle. The service life prediction module predicts the damage growth at different time scales in the future based on the current cumulative damage status and damage development trend, and calculates the remaining time required to reach the critical damage state, i.e., the remaining service life.

[0065] The predicted remaining useful life and the corresponding confidence interval are sent to the operation and maintenance management terminal. Specifically, the prediction output unit sends the predicted value of the remaining useful life and the confidence interval, for example, the confidence interval at 95% confidence level is the predicted value ±15%, to the operation and maintenance management terminal through the data interface. The confidence interval is calculated using the Monte Carlo simulation method, based on the statistical distribution of historical prediction errors, to determine the confidence interval at different confidence levels.

[0066] Based on historical prediction error statistics and model uncertainty analysis, the confidence level of the remaining useful life prediction result is calculated and sent to the operation and maintenance management terminal. Specifically, the confidence level calculation unit collects historical prediction data, calculates the error statistical distribution between the predicted value and the actual value, such as the mean and standard deviation of the error, and calculates the confidence level of the current prediction result based on the error statistical distribution. At the same time, the model uncertainty analysis method is used to analyze the parameter uncertainty and structural uncertainty of the digital twin model and comprehensively evaluate the reliability of the prediction result. The confidence level is expressed as a percentage and sent to the operation and maintenance management module 230 through the data interface.

[0067] The damage rate used in lifetime prediction can also be corrected by calculating environmental correction factors, which are calculated according to... Calculate, where: This indicates temperature correction, where T is the ambient temperature in °C. This indicates humidity correction; RH represents relative humidity (%). This indicates correction for corrosive media. The concentration is ppm (chloride ions).

[0068] In one alternative implementation, the digital twin model management module 220 includes: The visualization module is used to calculate the stress distribution of the first container based on strain and acceleration data, combined with the three-dimensional finite element model of the first container, and to generate and display a stress distribution cloud map based on the stress distribution.

[0069] In one alternative implementation, the visualization module is further used for: Based on the location of the source sensor group for calculating the cumulative damage index, the mapping region of the cumulative damage index and the remaining life prediction results on the three-dimensional finite element model is determined. The mapping area displays the deformation state corresponding to the cumulative damage index and the life state corresponding to the remaining life prediction result.

[0070] A digital twin visualization model of the first container is generated. Based on this model, a visualization module generates a three-dimensional structural visualization model of the first container and overlays a structural stress distribution cloud map, damage area annotations, and a predicted life heatmap onto the three-dimensional model. Specifically, the visualization module uses three-dimensional modeling software to construct a three-dimensional geometric model of the container, mapping the structural stress distribution calculated by the digital twin model to the corresponding positions in the three-dimensional model to generate a stress distribution cloud map, where red represents high-stress areas and green represents low-stress areas. The visualization module also annotates the damage areas calculated by the damage evolution model, using different colors and symbols to mark areas of different degrees of damage. Based on the prediction results from the life prediction module, the visualization module also generates a predicted life heatmap, where red represents short-life areas and green represents long-life areas. Since sensors can only collect discrete response data from key locations of the container, this invention establishes a mapping relationship between sensor measurement points and the overall structural state through a digital twin model. Specifically, a three-dimensional finite element model of the container is first established, with the sensor installation locations serving as observation nodes. Then, the strain and acceleration data collected by the sensors are used to correct the boundary conditions and material parameters of the finite element model. Finally, the stress distribution and deformation state of the areas where no sensors are installed are obtained through finite element calculations. Since different areas may have different damage levels and remaining lifespans due to different stress states, the damage index is calculated on a per-regional-element basis, and the overall health status of the container is determined based on the maximum damage state of different regions.

[0071] The visualization module is also used to present the 3D structural visualization model and overlay display information to the operation and maintenance management personnel through the 3D display terminal. Specifically, the visualization module encapsulates the generated 3D visualization model and overlay display information into a standard 3D rendering format and displays it through the 3D display terminal. The operation and maintenance management personnel can view the internal structure and status distribution of the container through interactive operation and intuitively judge the health status of the container.

[0072] The technical solution of this embodiment utilizes a container status acquisition module, including at least one sensor group mounted on a first container, to acquire structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data, and vibration data, while the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data, and impact load data. The container status acquisition module also includes an edge computing unit, which preprocesses the structural response data and environmental load data and sends the preprocessed data to a digital twin model management module. The digital twin model management module extracts features from the structural response data and environmental load data to obtain multi-source feature data. Based on the multi-source feature data, a first digital twin model of the first container is constructed. Damage evolution analysis is performed on the first container based on the first digital twin model to obtain damage evolution analysis results. Based on the damage evolution analysis results, a lifespan prediction is performed to obtain the lifespan prediction result of the first container, and the lifespan prediction is then displayed. The results are sent to the operation and maintenance management module; the operation and maintenance management module is used to generate the operation and maintenance tasks for the first container based on the life prediction results; the digital twin model management module includes: a data cleaning unit, used to remove outliers, fill in missing values, and perform timestamp calibration and synchronization processing on structural response data and environmental load data to obtain data cleaning results; a feature extraction unit, used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the data cleaning results to obtain structural response features, environmental load features, and historical operation features; a data alignment unit, used to align the structural response features, environmental load features, and historical operation features in terms of time dimension and feature dimension to obtain multi-source feature data; the digital twin model management module also includes: a container health status prediction unit, used to input the multi-source feature data into the container health status prediction sub-model of the first digital twin model to obtain the health status of the first container; wherein, the container health status prediction sub-model is a random forest model, which is trained by multi-source feature sample data and the real health status labels corresponding to the multi-source feature sample data. The technical solution of this invention addresses the problem of insufficient reliability in current lifespan prediction. It achieves standardized preprocessing of multi-source sensing data through data cleaning, feature extraction, and multi-dimensional alignment. It constructs a health prediction sub-model based on random forest to determine the nonlinear correlation between load, environment, and container health. It achieves real-time output of container health status based on unified feature input. Data preprocessing ensures the quality of model input, effectively reduces noise interference, and improves the accuracy of health status prediction, providing a reliable basis for structural damage assessment and remaining lifespan prediction.

[0073] In a preferred embodiment, a container condition monitoring and life prediction system such as Figure 3 As shown.

[0074] The physical container terminal uses a multimodal sensing unit to collect structural response data, environmental load data, and historical operation data of the container in real time during loading, unloading, transportation, and storage. The edge computing module performs preliminary filtering, noise reduction, and feature extraction on the raw data to obtain preprocessed sensing data, which is then sent to the digital twin platform terminal via a wireless communication network.

[0075] Secondly, the multi-source heterogeneous data fusion module on the digital twin platform performs data cleaning, feature extraction, and data alignment on the received raw sensing data stream to obtain fused structural feature vectors, environmental feature vectors, and temporal feature vectors.

[0076] Then, the adaptive dynamic modeling module constructs a digital twin model based on the fused data, the incremental learning unit performs incremental updates based on the latest perceived data, the model update unit uses the recursive least squares method or the extended Kalman filter algorithm to estimate and update the model parameters online, and the state synchronization unit calculates and corrects the state deviation to ensure that the digital twin model is synchronized with the state of the physical container. At the same time, the damage evolution model submodule calculates the structural damage index and predicts the trend of structural strength degradation.

[0077] Next, the lifespan prediction module assesses the current health status, predicts the remaining lifespan, and calculates the confidence level and confidence interval of the prediction results based on the updated digital twin model and damage evolution model, and sends the prediction results to the operation and maintenance management terminal.

[0078] Finally, the maintenance decision module of the operation and maintenance management terminal automatically generates maintenance suggestions and maintenance tasks according to the preset maintenance strategy rules. The task scheduling module assigns the tasks to the corresponding maintenance personnel and equipment and tracks the execution status. At the same time, the visualization module generates a three-dimensional visualization model, overlays and displays stress distribution cloud map, damage area marking and predicted life heat map, and presents it to the operation and maintenance management personnel through the three-dimensional display terminal to assist them in making intuitive judgments and decisions.

[0079] The maintenance decision module of the operation and maintenance management terminal receives the prediction results and confidence levels sent by the digital twin platform, and automatically generates maintenance suggestions and maintenance tasks according to preset maintenance strategy rules. Specifically, the maintenance decision module receives the predicted value and confidence level of the remaining service life, and makes judgments according to the following maintenance strategy rules: when the remaining service life is lower than the first preset threshold, such as 30 days, a priority maintenance task is generated, and it is recommended to perform maintenance checks and repairs immediately; when the remaining service life is between the first and second preset thresholds, such as 30 days to 90 days, a regular maintenance task is generated and included in the maintenance plan; when the remaining service life is higher than the second preset threshold, such as 90 days, it is included in the normal monitoring range and no special maintenance is required; the maintenance decision module also adjusts the maintenance strategy according to the confidence level. When the confidence level is lower than 70%, the priority of the maintenance task is reduced and a review and confirmation step is added.

[0080] The task scheduling module is used to assign generated maintenance tasks to the corresponding maintenance personnel and equipment, and to track the execution status of maintenance tasks. Specifically, the task scheduling module's database of maintenance personnel and equipment automatically assigns appropriate maintenance personnel and equipment based on the type and urgency of the maintenance task. The task scheduling module also tracks the execution status of maintenance tasks through a workflow engine, including the status of task creation, task assignment, task execution, and task acceptance, and collects maintenance records and updates the historical maintenance data of the container after the task is completed.

[0081] This embodiment achieves real-time data acquisition of container structural response and environmental loads by setting up a multimodal sensing unit, overcoming the time lag bottleneck of offline detection data acquisition methods in existing technologies. This allows the digital twin model to keep pace with the actual container status, ensuring real-time status monitoring. By setting up an adaptive dynamic modeling module with incremental learning and online parameter update technology, the digital twin model achieves dynamic evolution, tracking changes in container status in real time and significantly improving model prediction accuracy. Furthermore, by setting up a multi-source heterogeneous data fusion module, effective fusion of structural response data, environmental load data, and historical operational data is achieved, reducing the uncertainty of prediction results and improving the reliability of lifespan prediction. By setting up a damage evolution model submodule, a damage evolution model for container structures was established, enabling identification and intervention in the early stages of microscopic damage. This eliminates the drawbacks of the passive maintenance mode in existing technologies, which can only be maintained after visible damage occurs, thus achieving preventative maintenance. Furthermore, by setting up a visualization module, a three-dimensional visualization model is generated and overlaid with stress distribution cloud maps, damage area annotations, and predicted life heat maps, allowing operation and maintenance personnel to intuitively judge the health status of the container and improve the efficiency and accuracy of maintenance decisions. By setting up maintenance decision-making and task scheduling modules on the operation and maintenance management end, the automatic generation and scheduling of maintenance tasks are realized, improving the level of intelligence in maintenance management and reducing the cost of manual intervention.

[0082] Figure 4 This is a flowchart of a container condition monitoring and life prediction method provided by an embodiment of the present invention. This embodiment can be applied to the scenario of container condition monitoring and life prediction.

[0083] like Figure 4 As shown, the container condition monitoring and life prediction method includes the following steps: S310. Collect structural response data and environmental load data of the first container through the container status acquisition module.

[0084] The structural response data includes at least strain data, acceleration data, and vibration data; the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data, and impact load data; and the container status acquisition module includes at least one sensor group, which is installed on the first container.

[0085] S320: The edge computing unit of the container status acquisition module preprocesses the structural response data and environmental load data, and sends the preprocessed structural response data and environmental load data to the digital twin model management module.

[0086] S330. The digital twin model management module extracts features from the structural response data and environmental load data to obtain the first feature vector.

[0087] S340. Construct a first digital twin model of the first container based on the first feature vector, and perform damage evolution analysis on the first container based on the first digital twin model to obtain the damage evolution analysis results.

[0088] S350. Based on the damage evolution analysis results, life prediction is performed to obtain the life prediction result of the first container, and the life prediction result is sent to the operation and maintenance management module.

[0089] S360 generates maintenance tasks for the first container based on the life prediction results through the maintenance management module.

[0090] The technical solution of this embodiment involves acquiring structural response data and environmental load data of a first container through a container status acquisition module. The structural response data includes at least strain data, acceleration data, and vibration data, while the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data, and impact load data. The container status acquisition module includes at least one sensor group installed on the first container. The edge computing unit of the container status acquisition module preprocesses the structural response data and environmental load data and sends the preprocessed data to a digital twin model management module. The digital twin model management module extracts features from the structural response data and environmental load data to obtain a first feature vector. A first digital twin model of the first container is constructed based on the first feature vector. Damage evolution analysis is performed on the first container based on the first digital twin model to obtain damage evolution analysis results. Lifetime prediction is performed based on the damage evolution analysis results to obtain the lifetime prediction results of the first container, which are then sent to the operation and maintenance management module. The operation and maintenance management module generates operation and maintenance tasks for the first container based on the lifetime prediction results. The technical solution of this invention addresses the problem of insufficient reliability in current lifespan prediction. It can improve the accuracy and reliability of container lifespan prediction and enhance the timeliness of maintenance task generation through the fusion processing of multi-source data.

[0091] In one optional implementation, the data cleaning unit of the digital twin model management module performs outlier removal, missing value imputation, timestamp calibration, and synchronization processing on the structural response data and environmental load data to obtain data cleaning results; the feature extraction unit of the digital twin model management module extracts time-domain features, frequency-domain features, and time-frequency-domain features from the data cleaning results to obtain structural response features, environmental load features, and historical operation features; the data alignment unit of the digital twin model management module aligns the structural response features, environmental load features, and historical operation features in terms of time dimension and feature dimension to obtain multi-source feature data.

[0092] In one optional implementation, the container health status prediction unit of the digital twin model management module inputs multi-source feature data into the container health status prediction sub-model of the first digital twin model to obtain the health status of the first container; wherein, the container health status prediction sub-model is a random forest model, which is trained by multi-source feature sample data and the real health status labels corresponding to the multi-source feature sample data.

[0093] In one optional implementation, the incremental update sub-unit of the container health status prediction unit selects historical multi-source feature sample data and new multi-source feature sample data as an incremental training set according to a preset ratio. For each decision tree in the container health status prediction sub-model, each first sample in the incremental training set is traversed. Starting from the root node of the tree, the first sample is assigned to the left child node or the right child node according to the node splitting feature and threshold. When the sample assignment reaches the leaf node, the category count or regression value of the leaf node is updated. When the number of first samples assigned to the leaf node exceeds the preset sample number threshold, node splitting is triggered, and the feature with the largest decrease in Gini coefficient is selected as the splitting feature, thus completing the incremental update of the container health status prediction sub-model.

[0094] In one optional implementation, the model parameter update unit of the digital twin model management module estimates the model parameters of the digital twin model based on multi-source feature data using recursive least squares or extended Kalman filtering algorithm to obtain updated model parameters; wherein, the model parameters include at least structural state parameters and degradation parameters.

[0095] In one optional implementation, the damage evolution analysis unit of the digital twin model management module calculates the structural damage index of the first container for each load cycle based on the strain data, material ultimate strength, load cycle number, and material constant in the structural response data of the first digital twin model, wherein the load cycle number is determined by the rainflow counting method; the cumulative damage index is obtained by summing all structural damage indices.

[0096] In one optional implementation, the life prediction module of the digital twin model management module calculates the remaining life prediction result of the first container based on the cumulative damage index, critical damage index, preset safety factor and damage rate; wherein, the damage rate is obtained by linear regression fitting of the structural damage index at the most recent preset number of time points to time.

[0097] In one optional implementation, the visualization module of the digital twin model management module calculates the stress distribution of the first container based on strain data and acceleration data, combined with the three-dimensional finite element model of the first container, and generates and displays a stress distribution cloud map based on the stress distribution.

[0098] In one optional implementation, the visualization module determines the mapping region of the cumulative damage index and the remaining life prediction result on the three-dimensional finite element model based on the location of the source sensor group of the strain data used to calculate the cumulative damage index; and displays the deformation state corresponding to the cumulative damage index and the life state corresponding to the remaining life prediction result on the mapping region.

[0099] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0100] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A container condition monitoring and life prediction system, characterized in that, include: The container status acquisition module includes at least one sensor group, which is installed on the first container and is used to acquire the structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data and vibration data, and the environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data and impact load data. The container status acquisition module also includes an edge computing unit, which is used to preprocess the structural response data and environmental load data, and send the preprocessed structural response data and environmental load data to the digital twin model management module. The digital twin model management module is used to extract features from the structural response data and the environmental load data to obtain multi-source feature data; A first digital twin model of the first container is constructed based on the multi-source feature data, and damage evolution analysis is performed on the first container based on the first digital twin model to obtain the damage evolution analysis results. Based on the damage evolution analysis results, a lifespan prediction is performed to obtain the lifespan prediction result of the first container, and the lifespan prediction result is sent to the operation and maintenance management module. The operation and maintenance management module is used to generate operation and maintenance tasks for the first container based on the life prediction results.

2. The system according to claim 1, characterized in that, The digital twin model management module includes: The data cleaning unit is used to perform outlier removal, missing value filling, timestamp calibration and synchronization processing on the structural response data and the environmental load data to obtain the data cleaning results. The feature extraction unit is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the data cleaning results to obtain structural response features, environmental load features, and historical operation features. The data alignment unit is used to perform time dimension alignment and feature dimension alignment on the structural response features, the environmental load features, and the historical operation features to obtain multi-source feature data.

3. The system according to claim 1, characterized in that, The digital twin model management module also includes: The container health status prediction unit is used to input the multi-source feature data into the container health status prediction sub-model of the first digital twin model to obtain the health status of the first container. The container health status prediction sub-model is a random forest model, which is trained using multi-source feature sample data and the corresponding real health status labels.

4. The system according to claim 3, characterized in that, The container health status prediction unit also includes: The incremental update sub-unit selects historical multi-source feature sample data and new multi-source feature sample data as the incremental training set according to a preset ratio. For each decision tree in the container health status prediction sub-model, it traverses each first sample in the incremental training set, starting from the root node of the tree, and assigns the first sample to the left child node or the right child node according to the node splitting feature and threshold. When sample allocation reaches a leaf node, the category count or regression value of the leaf node is updated. When the number of the first sample allocated to the leaf node exceeds a preset sample number threshold, node splitting is triggered, and the feature with the largest decrease in Gini coefficient is selected as the splitting feature to complete the incremental update of the container health status prediction sub-model.

5. The system according to claim 1, characterized in that, The digital twin model management module includes: The model parameter update unit is used to estimate the model parameters of the digital twin model based on the multi-source feature data by using the recursive least squares method or the extended Kalman filter algorithm to obtain updated model parameters. The model parameters include at least structural state parameters and degradation parameters.

6. The system according to claim 1, characterized in that, The digital twin model management module includes: The damage evolution analysis unit is used to calculate the structural damage index of the first container for each load cycle based on the strain data, material ultimate strength, load cycle number and material constant in the structural response data of the first digital twin model, wherein the load cycle number is determined by the rainflow counting method. The cumulative damage index is obtained by summing all the structural damage indices.

7. The system according to claim 6, characterized in that, The digital twin model management module includes: The life prediction module is used to calculate the remaining life prediction result of the first container based on the cumulative damage index, critical damage index, preset safety factor and damage rate. The damage rate is obtained by linearly regressing the structural damage index at the most recent preset number of time points onto time.

8. The system according to claim 7, characterized in that, The digital twin model management module includes: The visualization module is used to calculate the stress distribution of the first container based on the strain data and the acceleration data, combined with the three-dimensional finite element model of the first container, and to generate and display a stress distribution cloud map based on the stress distribution.

9. The system according to claim 8, characterized in that, The visualization module is also used for: Based on the location of the source sensor group from which the strain data for calculating the cumulative damage index is calculated, the mapping region of the cumulative damage index and the remaining life prediction result on the three-dimensional finite element model is determined. The deformation state corresponding to the cumulative damage index and the lifetime state corresponding to the remaining lifetime prediction result are displayed on the mapping area.

10. A method for container condition monitoring and life prediction, characterized in that, include: The container status acquisition module collects structural response data and environmental load data of the first container. The structural response data includes at least strain data, acceleration data and vibration data. The environmental load data includes at least ambient temperature data, ambient humidity data, corrosive medium concentration data and impact load data. The container status acquisition module includes at least one sensor group, which is installed on the first container. The edge computing unit of the container status acquisition module preprocesses the structural response data and environmental load data, and sends the preprocessed structural response data and environmental load data to the digital twin model management module. The digital twin model management module extracts features from the structural response data and the environmental load data to obtain a first feature vector; A first digital twin model of the first container is constructed based on the first feature vector, and damage evolution analysis is performed on the first container based on the first digital twin model to obtain the damage evolution analysis results. Based on the damage evolution analysis results, a lifespan prediction is performed to obtain the lifespan prediction result of the first container, and the lifespan prediction result is sent to the operation and maintenance management module. The operation and maintenance management module generates the operation and maintenance task for the first container based on the life prediction result.

Images