A leakage monitoring system for hazardous chemical transportation

Abstract

The application discloses a leakage monitoring system in the transportation of dangerous chemicals, and relates to the technical field of sensor detection, which comprises a tank state sensing array arranged along the outer wall and / or inner wall of a transportation tank and comprising a plurality of heterogeneous sensor nodes; a transportation dynamic state acquisition module installed on a transportation vehicle; an interference signal prediction model pre-trained based on historical transportation data and used to predict interference signal components that should be generated in the multi-dimensional state parameters under the current transportation dynamics according to the motion state parameters; and a residual extraction and leakage judgment module used to calculate the residual between the actual acquisition value of the multi-dimensional state parameters and the predicted interference signal components output by the interference signal prediction model, and to perform leakage judgment based on the residual. By constructing the interference signal prediction model and the adaptive detection threshold, the dynamic interference generated by the vehicle motion is effectively stripped, and the detection reliability under complex road conditions is improved.

Description

Technical Field

[0001] This invention relates to the field of sensor detection technology, and more specifically to a leak monitoring system for hazardous chemicals during transportation. Background Technology

[0002] With the rapid development of the modern chemical industry, the types of hazardous chemicals are becoming increasingly diverse, and the demand for their cross-regional transportation continues to grow. Hazardous chemicals typically possess characteristics such as flammability, explosiveness, toxicity, or strong corrosivity. Leaks during transportation can easily lead to major accidents involving environmental pollution, fires, explosions, or mass casualties. Therefore, safety monitoring technologies for hazardous chemical transportation have always been a key focus of industry research. Currently, the mainstream monitoring methods in the industry mainly rely on various physical sensors installed on the transport tanks. Specifically, common monitoring methods include placing point temperature sensors at tank welds and valves, installing pressure transmitters at the bottom of the tank to monitor hydraulic pressure changes, and setting up electrochemical or infrared gas sensors in the gas phase space to detect the concentration of volatile organic compounds (VOCs). In addition, to assist in assessing the vehicle's safety status, transport vehicles are usually equipped with GPS-based driving recorders and basic vehicle attitude sensors. These technologies, to a certain extent, achieve passive monitoring of the tank's condition, forming the foundation of the current hazardous chemical logistics safety control system.

[0003] However, the transportation of hazardous chemicals is characterized by high dynamism and complexity. During operation, transport vehicles inevitably experience various conditions such as bumps, sudden braking, cornering, and incline driving, causing severe sloshing of the liquid inside the tank. Simultaneously, road surface excitation is transmitted to the tank through the tires and suspension system, inducing random vibrations in the vehicle structure. These complex mechanical environments significantly couple with the pressure distribution, temperature field, and structural vibration modes of the tank wall. For example, severe sloshing of the liquid causes large fluctuations in tank wall pressure, which are spectrally very similar to the pressure drop caused by minor leaks. Similarly, prolonged mechanical vibration of the vehicle generates frictional heat on the tank's metal surface, interfering with temperature sensor readings. This challenge of detecting weak signals under such strong interference makes traditional monitoring systems ineffective in complex road conditions.

[0004] To address the aforementioned interference issues, existing technologies typically employ fixed threshold alarm strategies or simple digital filtering algorithms. For example, a fixed lower pressure limit or upper temperature limit is set, and an alarm is triggered once the sensor reading exceeds this limit. However, this approach has serious limitations: because it doesn't consider the dynamic modulation effect of vehicle motion on sensor readings, the system is highly susceptible to misinterpreting normal physical disturbances as leaks when the vehicle passes over speed bumps, makes sharp turns, or travels on unpaved roads, resulting in a very high false alarm rate. Conversely, raising the alarm threshold to reduce the false alarm rate can cause minor early leaks to be masked by environmental noise and go undetected, leading to missed alarms.

[0005] It should be noted that the information disclosed in this background section is only for understanding the background technology of the present invention, and therefore may include information that does not constitute prior art. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of high false alarm rate and high false alarm rate in leakage monitoring caused by vehicle vibration and liquid sloshing interference during the transportation of hazardous chemicals.

[0007] To address the aforementioned technical problems, this invention proposes a leakage monitoring system for hazardous chemicals during transportation, comprising:

[0008] A tank state sensing array is arranged along the outer and / or inner wall of the transport tank, including multiple heterogeneous sensor nodes, for real-time acquisition of multi-dimensional state parameters of the tank, including tank wall pressure distribution, tank wall temperature distribution, gas concentration in the gas phase space and tank structure vibration response.

[0009] The transportation dynamic status acquisition module is installed on the transport vehicle to collect the vehicle's motion status parameters in real time. The motion status parameters include the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate, pitch rate, and travel speed.

[0010] An interference signal prediction model, pre-trained based on historical transportation data, is used to predict the interference signal components that should be generated in the multi-dimensional state parameters under the current transportation dynamics, according to the motion state parameters.

[0011] The residual extraction and leakage determination module is used to calculate the residual between the actual acquired value of the multidimensional state parameter and the predicted interference signal component output by the interference signal prediction model, and to determine leakage based on the residual.

[0012] Furthermore, the interference signal prediction model includes a vibration interference prediction sub-model and a sway interference prediction sub-model;

[0013] The vibration interference prediction sub-model is used to predict the interference signal components generated by the vibration of the transport vehicle in the vibration response of the tank structure and the temperature distribution of the tank wall based on the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate and pitch rate.

[0014] The sloshing interference prediction sub-model is used to predict the interference signal components generated by the sloshing of liquid inside the tank in the pressure distribution of the tank wall, based on the vehicle's longitudinal acceleration, lateral acceleration, and driving speed, combined with the tank's geometric parameters and the current loaded liquid level.

[0015] Furthermore, the swaying disturbance prediction sub-model is constructed based on a hybrid training method combining computational fluid dynamics simulations and measured data;

[0016] The hybrid training method includes: establishing a computational fluid dynamics simulation model based on the geometric parameters of the tank and a preset liquid level range, generating simulation data of liquid sloshing pressure distribution under various transportation conditions; collecting actual motion state parameters and measured data of tank wall pressure distribution during actual transportation, and using the measured data to calibrate the parameters and correct the output of the computational fluid dynamics simulation model to obtain the sloshing interference prediction sub-model.

[0017] Furthermore, the determination process of the residual extraction and leakage determination module includes:

[0018] The magnitude, duration, and spatial distribution pattern of the residual are used as features input to a pre-trained leakage determination classifier;

[0019] The leakage determination classifier is trained based on residual data under normal transportation conditions and residual data under simulated leakage conditions, and is used to distinguish between residuals caused by sensor noise and model prediction errors and residuals caused by actual leakage.

[0020] When the magnitude of the residual exceeds the adaptive detection threshold determined by the model prediction error distribution, and exhibits a spatial correlation pattern that conforms to the leakage diffusion law on multiple adjacent sensor nodes, it is determined to be a leakage event.

[0021] Furthermore, the adaptive detection threshold is dynamically determined by the prediction error statistical characteristics of the interference signal prediction model;

[0022] The interference signal prediction model continuously records the deviation between the predicted value and the actual interference signal during operation and updates the probability distribution parameters of the prediction error.

[0023] The adaptive detection threshold is set as the preset confidence quantile of the prediction error probability distribution, so that the false alarm rate of leakage detection is maintained below the preset target false alarm rate under the condition of no leakage.

[0024] Furthermore, the system also includes an online adaptive update module for the interference signal prediction model;

[0025] The online adaptive update module is used to perform online incremental updates to the interference signal prediction model during transportation when it is determined that the current operating condition is a leak-free steady state. This is done using real-time collected motion state parameters and multi-dimensional state parameters to adapt to changes in tank loading status, tank structure aging, and differences in road surface characteristics of different transportation routes.

[0026] Furthermore, the multiple heterogeneous sensor nodes in the tank state sensing array are divided into multiple sensor clusters according to the structural regions of the tank, and each sensor cluster covers a local area of ​​the tank.

[0027] The residual extraction and leakage determination module calculates the local residual for each sensor cluster and locates the leakage source based on the spatial correlation of the residuals between adjacent sensor clusters.

[0028] When the residuals of multiple adjacent sensor clusters simultaneously exceed their respective adaptive detection thresholds, and the residual amplitudes exhibit a spatial gradient decreasing outward from a certain sensor cluster, the tank area corresponding to that sensor cluster is determined to be the location of the leak source.

[0029] Furthermore, the system also includes a loading status identification module;

[0030] The loading status identification module is used to identify the current loading liquid level and liquid centroid position of the tank based on the vehicle dynamic response characteristics output by the transportation dynamic status acquisition module.

[0031] The interference signal prediction model uses the identified loading liquid level and liquid centroid position as input parameters to output accurate interference signal predictions under different loading conditions.

[0032] Furthermore, the transportation dynamic status acquisition module also includes a positioning unit and a road surface information acquisition unit;

[0033] The positioning unit is used to obtain the real-time location of the transport vehicle;

[0034] The road surface information acquisition unit is used to acquire road surface feature information of a preset road section in front of the vehicle based on the real-time location and a pre-built road surface smoothness database.

[0035] The interference signal prediction model uses the road surface feature information as prior information and adjusts the model parameters for interference signal prediction in advance before the vehicle enters a road section with abrupt changes in road surface features.

[0036] Furthermore, the system also includes an onboard emergency response module;

[0037] The vehicle-mounted emergency linkage module is connected to the residual extraction and leakage determination module. After a leakage event is determined, the severity of the leakage is determined based on the spatial distribution characteristics and amplitude level of the residual, and corresponding emergency response actions are executed.

[0038] The emergency response actions include: closing the tank pipeline valves corresponding to the leak source location, activating the local spray device corresponding to the leak source location, reporting the leak location and severity information to the monitoring center, and controlling the vehicle to stop safely when the leak severity exceeds a preset threshold.

[0039] The beneficial effects of this invention are: by constructing an interference signal prediction model and an adaptive detection threshold, dynamic interference generated by vehicle movement is effectively eliminated, improving detection reliability under complex road conditions. By utilizing the spatial distribution characteristics of a multi-dimensional sensor array and residual gradient analysis, the shortcomings of traditional single-point sensors in terms of traceability are overcome.

[0040] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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 scope of protection of the present invention.

[0044] like Figure 1 As shown, this embodiment provides a leak monitoring system for hazardous chemicals during transportation, including:

[0045] A tank state sensing array is arranged along the outer and / or inner wall of the transport tank, including multiple heterogeneous sensor nodes, for real-time acquisition of multi-dimensional state parameters of the tank, including tank wall pressure distribution, tank wall temperature distribution, gas concentration in the gas phase space and tank structure vibration response.

[0046] The transportation dynamic status acquisition module is installed on the transport vehicle to collect the vehicle's motion status parameters in real time. The motion status parameters include the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate, pitch rate, and travel speed.

[0047] An interference signal prediction model, pre-trained based on historical transportation data, is used to predict the interference signal components that should be generated in the multi-dimensional state parameters under the current transportation dynamics, according to the motion state parameters.

[0048] The residual extraction and leakage determination module is used to calculate the residual between the actual acquired value of the multidimensional state parameter and the predicted interference signal component output by the interference signal prediction model, and to determine leakage based on the residual.

[0049] In this implementation, the tank status sensing array adopts a distributed bus topology. Specifically, a flexible piezoresistive pressure sensor array based on polyimide is spirally wound around the inner wall of the tank to construct a full-circumferential pressure distribution map of the tank wall; thin-film platinum resistance temperature sensors are attached near the welding points on the outer wall of the tank to capture micron-level temperature change signals; a tunable diode laser absorption spectroscopy gas concentration sensor is installed in the gas phase space at the top of the tank for high-sensitivity detection of volatile media; and a high-frequency MEMS triaxial accelerometer is installed on the tank's reinforcing ring to collect structural vibration responses.

[0050] The transportation dynamic status acquisition module uses a fiber optic inertial navigation unit, which is fixed at the center of gravity of the vehicle chassis. It directly acquires the three-axis linear acceleration and angular velocity, and at the same time reads the pulse width modulation signal from the vehicle ECU through the CAN bus to calculate the driving speed.

[0051] The interference signal prediction model is deployed in an on-board industrial PC and adopts a time-series prediction architecture based on a long short-term memory network. It uses vehicle motion data over a period of time to predict the pressure, temperature, and vibration baseline values ​​of the tank affected by mechanical vibration and liquid sloshing in the future.

[0052] The residual extraction and leakage determination module is implemented through a high-speed FPGA. It performs subtraction operations to extract the residual and uses digital signal processing algorithms to filter out high-frequency white noise, outputting the residual as a leakage feature vector.

[0053] In this embodiment, the multidimensional state parameters collected by the tank state sensing array are essentially composed of three superimposed parts: the first is the change in real physical parameters caused by leakage, the second is the interference component caused by transportation dynamics, and the third is the random noise of the sensor.

[0054] Among them, the dynamic interference component of transportation mainly originates from the impact of the tank wall caused by road excitation, the liquid phase sloshing and slapping caused by vehicle acceleration, deceleration and steering, the structural vibration transmitted from the powertrain to the tank, and the heat conduction through the tank wall due to environmental temperature changes. During transportation, the amplitude of this interference component is usually much higher than the parameter changes caused by minor leaks. If the threshold is set directly based on the original collected values ​​for leak judgment, it is very easy to cause false alarms due to the normal driving conditions of the vehicle.

[0055] To address the aforementioned technical issues, this embodiment introduces an interference signal prediction model. This model takes as input the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate, pitch rate, and speed, continuously collected by the transportation dynamics acquisition module over a period of time, and outputs multi-dimensional state parameter prediction values ​​that match the current transportation dynamics. Since this model is trained offline based on historical leak-free transportation data, it only learns the nonlinear mapping relationship between vehicle motion posture and tank pressure, temperature, gas phase concentration, and vibration. Therefore, it can simulate baseline interference generated by vehicle motion without including leakage characteristics.

[0056] The residual extraction and leakage determination module performs a subtraction operation, subtracting the actual acquired values ​​of the multidimensional state parameters from the predicted interference signal components output by the interference signal prediction model to obtain the residual signal. The physical mechanism of this process is as follows: under leak-free conditions, the effective components in the actual acquired values ​​can be completely explained by the predicted interference components, and the residual after subtraction approaches the sensor's random noise, exhibiting a zero-mean distribution after digital filtering. However, when a leak occurs, the minute changes in physical parameters caused by the leak cannot be explained by the interference signal prediction model trained based on historical data, and therefore are completely reflected in the residual signal. Thus, this residual signal is the pure anomalous component after stripping away transportation dynamic interference and sensor white noise, and its physical meaning directly corresponds to the residual response of the leakage event in various modes of tank wall pressure, tank wall temperature, gas phase concentration, and structural vibration.

[0057] Furthermore, the aforementioned residual signals are organized according to the dimensions of the sensor array to form a leakage feature vector. Specifically, this feature vector includes the pressure residuals of each node of the spiral pressure array, the temperature residuals of the thin-film platinum resistance thermometers, the gas concentration residuals in the gas phase space, and the vibration residuals of the triaxial accelerometers. The judgment module can execute judgment logic based on this leakage feature vector: triggering an alarm when the residual of a single dimension exceeds a preset threshold, or confirming the leakage level when the residuals of multiple dimensions show specific correlated changes.

[0058] Furthermore, the interference signal prediction model includes a vibration interference prediction sub-model and a sway interference prediction sub-model;

[0059] The vibration interference prediction sub-model is used to predict the interference signal components generated by the vibration of the transport vehicle in the vibration response of the tank structure and the temperature distribution of the tank wall based on the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate and pitch rate.

[0060] The sloshing interference prediction sub-model is used to predict the interference signal components generated by the sloshing of liquid inside the tank in the pressure distribution of the tank wall, based on the vehicle's longitudinal acceleration, lateral acceleration, and driving speed, combined with the tank's geometric parameters and the current loaded liquid level.

[0061] In this embodiment, the interference signal prediction model adopts a dual-channel parallel processing mechanism. The vibration interference prediction sub-model is specifically designed to handle high-frequency mechanical noise. Its input is connected to the high-frequency vibration channel of the IMU. The model extracts the road bump features through wavelet transform and outputs two signals: one is the background vibration waveform in the tank structure vibration response, which is used to separate it from the accelerometer data; the other is the low-frequency drift of the tank wall temperature distribution, which is caused by mechanical friction heat generation. The model predicts this temperature rise component through the thermodynamic transfer function to eliminate its masking of the low-temperature leakage characteristics.

[0062] The sloshing interference prediction sub-model focuses on low-frequency hydrodynamic effects. In addition to IMU acceleration data, it also incorporates GPS velocity signals as inputs, and combines pre-stored tank CAD geometric parameters and real-time liquid level height measured by radar level gauges. This sub-model uses potential flow theory or deep learning surrogate models to calculate the dynamic pressure field distribution generated by the sloshing of liquid inside the tank on the tank wall under a given inertial force, thereby separating the pseudo-leakage pressure signal from the total pressure signal.

[0063] Furthermore, to address the issues of strong nonlinearity in liquid sloshing pressure and poor physical interpretability and weak generalization ability in purely data-driven models, this embodiment employs a hybrid training method combining computational fluid dynamics simulation and measured data to construct the sloshing disturbance prediction sub-model. This hybrid training method specifically includes two stages: offline simulation pre-training and online measured calibration. Through the dual constraints of physical mechanisms and data characteristics, it ensures that the model conforms to fluid dynamics laws and is adaptable to actual transportation conditions.

[0064] The hybrid training method includes: firstly, establishing a computational fluid dynamics simulation model based on the actual geometric parameters of the transport tank (including diameter, length, position of reinforcing ring, and preset liquid level range). In this embodiment, the VOF multiphase flow model is preferably used for modeling, and an appropriate grid resolution is set according to the computational accuracy requirements to accurately capture the dynamic changes of the gas-liquid interface.

[0065] In setting the boundary conditions, the liquid level change range and vehicle acceleration excitation spectrum covering typical road transport conditions are selected. The excitation spectrum includes dynamic parameters corresponding to various typical driving behaviors such as braking, steering and bumping. By solving the unsteady flow field equations, a large-scale liquid sloshing pressure field simulation dataset is generated in batches as the initial training base for the sloshing disturbance prediction sub-model, so that the model can learn the basic flow law of liquid in the tank under stress.

[0066] During actual vehicle operation, motion state parameters under real road conditions are collected using a transportation dynamic state acquisition module, while corresponding measured data of tank wall pressure distribution are collected using a tank state sensor array. This measured data is used as observations, and a Bayesian calibration algorithm is employed to correct the parameters of the simulation model generated in the first stage. Specifically, the turbulent viscosity coefficient and wall shear stress parameters in the computational fluid dynamics model are corrected through inversion, eliminating systematic deviations between the simulation environment and the actual physical environment caused by factors such as manufacturing tolerances and surface roughness, resulting in the final sway interference prediction sub-model deployed on the vehicle-mounted industrial PC.

[0067] Compared to a simple deep learning model, this embodiment introduces computational fluid dynamics simulation as a physical prior constraint, which makes the prediction results have clear fluid dynamics physical interpretation. Compared to a pure simulation model, the model's matching degree to complex and ever-changing real road conditions is improved through Bayesian calibration of measured data.

[0068] Furthermore, the determination process of the residual extraction and leakage determination module includes:

[0069] The magnitude, duration, and spatial distribution pattern of the residual are used as features input to a pre-trained leakage determination classifier;

[0070] The leakage determination classifier is trained based on residual data under normal transportation conditions and residual data under simulated leakage conditions, and is used to distinguish between residuals caused by sensor noise and model prediction errors and residuals caused by actual leakage.

[0071] When the magnitude of the residual exceeds the adaptive detection threshold determined by the model prediction error distribution, and exhibits a spatial correlation pattern that conforms to the leakage diffusion law on multiple adjacent sensor nodes, it is determined to be a leakage event.

[0072] In this embodiment, the residual extraction and leakage determination module runs on an ARM architecture processor. The module first performs feature engineering on the residual signal: calculating the root mean square value of the residual as the amplitude feature; calculating the duration exceeding the 3σ threshold continuously as the duration feature; and calculating the cross-correlation coefficient of the residuals of adjacent sensor nodes as the spatial distribution feature. These features are concatenated into a feature vector and input to a pre-trained XGBoost classifier. The training set of this classifier consists of two parts: negative samples are the residuals predicted by the model under normal driving conditions, and positive samples are residual data collected by artificially drilling holes in the tank to simulate minor leaks. Through learning, the classifier discovers that the residuals of real leaks often exhibit extremely high amplitudes, long durations, and significant spatial gradients in specific regions, while residuals caused by model errors are usually global and random.

[0073] The decision logic uses an AND gate mechanism. The leakage flag is triggered only when the amplitude exceeds the threshold and the spatial correlation conforms to the leakage diffusion model (such as the analogy of the Gaussian plume model on the solid tank wall), thereby greatly reducing the false alarm rate.

[0074] The adaptive detection threshold is dynamically determined by the statistical characteristics of the prediction error of the interference signal prediction model;

[0075] The interference signal prediction model continuously records the deviation between the predicted value and the actual interference signal during operation and updates the probability distribution parameters of the prediction error.

[0076] The adaptive detection threshold is set as the preset confidence quantile of the prediction error probability distribution, so that the false alarm rate of leakage detection is maintained below the preset target false alarm rate under the condition of no leakage.

[0077] To further balance detection sensitivity and false alarm rate, this embodiment introduces the concept of statistical process control and constructs an adaptive detection threshold mechanism based on residual statistical characteristics.

[0078] Under non-leakage conditions, the difference between the predicted value of the interference signal prediction model and the actual acquired value is continuously monitored and calculated; this is the aforementioned residual signal. Theoretically, when no leakage occurs, this residual signal only contains sensor random noise and model fitting residuals, and its probability distribution follows a normal distribution with a mean of 0. Based on this assumption, the variance parameter of this residual distribution, denoted as σ², is calculated and updated in real time using a recursive estimation algorithm.

[0079] The adaptive detection threshold T is dynamically set based on a preset confidence level, specifically expressed as follows: the detection threshold T is equal to the product of the scaling factor k and the variance σ at the current time. Here, the scaling factor k is the quantile corresponding to the standard normal distribution table based on the preset false alarm rate target (e.g., set to 0.5%).

[0080] When the vehicle runs smoothly and the model prediction accuracy is high, the fluctuation range of the residual is small, and the variance σ decreases accordingly, causing the detection threshold T to tighten automatically, thus improving the sensitivity to minute leaks.

[0081] When a vehicle travels on extremely rough roads or is in a state of violent motion, the violent sloshing of the liquid inevitably increases the model's prediction error, and the fluctuation of the residuals intensifies, resulting in a corresponding increase in the variance σ. The detection threshold T is automatically widened, thus allowing for a larger prediction bias and avoiding false alarms caused by the model's insufficient fitting ability at that moment.

[0082] Furthermore, the system also includes an online adaptive update module for the interference signal prediction model;

[0083] The online adaptive update module is used to perform online incremental updates to the interference signal prediction model during transportation when it is determined that the current operating condition is a leak-free steady state. This is done using real-time collected motion state parameters and multi-dimensional state parameters to adapt to changes in tank loading status, tank structure aging, and differences in road surface characteristics of different transportation routes.

[0084] In this embodiment, the steady-state condition is defined as follows: vehicle speed fluctuation is less than ±5 km / h, roll rate is less than 0.01 rad / s, and the residual amplitude output by the residual extraction module remains below the adaptive threshold for 300 seconds. Once the steady-state condition is reached, the current data is determined to be a clean, leak-free sample, and the incremental learning process is initiated. Using the currently collected vehicle motion parameters as input and the tank state parameters as labels, the interference signal prediction model is fine-tuned.

[0085] For example, as the number of transports increases, wear of the anti-corrosion coating inside the tank may cause changes in sloshing damping. Through online updates, the model can automatically track this change and always maintain the best fit to the current physical state of the tank, solving the problem of performance degradation of traditional models due to equipment aging.

[0086] Furthermore, the multiple heterogeneous sensor nodes in the tank state sensing array are divided into multiple sensor clusters according to the structural regions of the tank, and each sensor cluster covers a local area of ​​the tank.

[0087] The residual extraction and leakage determination module calculates the local residual for each sensor cluster and locates the leakage source based on the spatial correlation of the residuals between adjacent sensor clusters.

[0088] When the residuals of multiple adjacent sensor clusters simultaneously exceed their respective adaptive detection thresholds, and the residual amplitudes exhibit a spatial gradient decreasing outward from a certain sensor cluster, the tank area corresponding to that sensor cluster is determined to be the location of the leak source.

[0089] Specifically, in this embodiment, the tank surface is divided into 16 fan-shaped sensor cluster installation areas, and each cluster contains at least one pressure sensor, one temperature sensor, and one vibration sensor. The residual extraction module calculates the local residual energy of these 16 clusters in parallel. When determining a leak, it not only focuses on whether the residual exceeds the standard, but also analyzes the spatial topological relationship between the clusters. Then, it uses DS evidence theory to fuse the information of adjacent clusters and calculates the spatial gradient field of the residual amplitude. Since physical leak points usually form a pressure drop zone and a temperature anomaly zone centered on the leak outlet, when a cluster (e.g., cluster #7) is detected to have the highest residual amplitude, and the residual amplitudes of its adjacent clusters (#6, #8, #11, #12) decrease with the square of the distance, the tank circumferential weld area corresponding to cluster #7 is determined to be the leak source.

[0090] Furthermore, the system also includes a loading status identification module;

[0091] The loading status identification module is used to identify the current loading liquid level and liquid centroid position of the tank based on the vehicle dynamic response characteristics output by the transportation dynamic status acquisition module.

[0092] The interference signal prediction model uses the identified loading liquid level and liquid centroid position as input parameters to output accurate interference signal predictions under different loading conditions.

[0093] The loading status identification module in this embodiment utilizes a vehicle longitudinal dynamics model for status observation. When the vehicle starts or brakes, the liquid will tilt backward or forward due to inertia, causing changes in the vehicle suspension deformation. The module monitors the difference in vertical acceleration between different wheels and combines it with the vehicle's longitudinal acceleration to inversely solve for the equivalent center of mass position of the liquid inside the tank. Simultaneously, using a Kalman filter-based liquid level observer, the current loading liquid level is estimated in real time based on the mathematical relationship between sloshing frequency and liquid level. The identified liquid level and center of mass coordinates are encoded into high-dimensional vectors and input into the interference signal prediction model in real time, enabling the sloshing interference prediction sub-model to call corresponding fluid dynamic parameters for different loading states such as full tank, half tank, or empty tank.

[0094] Furthermore, the transportation dynamic status acquisition module also includes a positioning unit and a road surface information acquisition unit;

[0095] The positioning unit is used to obtain the real-time location of the transport vehicle;

[0096] The road surface information acquisition unit is used to acquire road surface feature information of a preset road section in front of the vehicle based on the real-time location and a pre-built road surface smoothness database.

[0097] The interference signal prediction model uses the road surface feature information as prior information and adjusts the model parameters for interference signal prediction in advance before the vehicle enters a road section with abrupt changes in road surface features.

[0098] In this embodiment, the positioning unit uses a BeiDou-3 RTK high-precision positioning terminal, achieving centimeter-level positioning accuracy. The road information acquisition unit connects to a cloud-based high-precision map server via a 4G / 5G communication module. This server stores data on road surface smoothness index, slope, and speed bump distribution on major transportation arteries nationwide. When a vehicle approaches a sudden change in road surface distance, the power spectral density characteristics of that section are acquired in advance. Upon receiving this prior information, the interference signal prediction model activates a pre-tensioning mode: adjusting the road surface spectrum input parameters in the vibration interference prediction sub-model in advance, increasing the model gain, and enabling the model to predict upcoming strong vibrations. This feedforward control strategy effectively solves the problem of instantaneous false alarms caused by model lag when traversing bumpy road sections.

[0099] Furthermore, the system also includes an onboard emergency response module;

[0100] The vehicle-mounted emergency linkage module is connected to the residual extraction and leakage determination module. After a leakage event is determined, the severity of the leakage is determined based on the spatial distribution characteristics and amplitude level of the residual, and corresponding emergency response actions are executed.

[0101] The emergency response actions include: closing the tank pipeline valves corresponding to the leak source location, activating the local spray device corresponding to the leak source location, reporting the leak location and severity information to the monitoring center, and controlling the vehicle to stop safely when the leak severity exceeds a preset threshold.

[0102] The vehicle-mounted emergency linkage module in this embodiment integrates the vehicle control protocol interface. Upon receiving a leak detection signal, the module first calculates the leak rate based on the integral area of ​​the residual amplitude and classifies the leak level. For a minor Level 1 leak, the module controls the solenoid valve via the CAN bus to close the branch pipeline corresponding to the leak area, cutting off the leak source, and activates the local spray solenoid valve in that area for dilution. Simultaneously, a report containing latitude and longitude, leak level, and estimated medium is sent to the monitoring center via the 4G network. For a severe Level 2 leak, in addition to the above actions, the module forcibly intervenes in the vehicle's power system, controls the engine ECU to limit torque output, controls the retarder for braking, and guides the vehicle automatically off the main road using a high-precision map to stop in a preset safe escape lane. The entire process requires no manual operation from the driver, achieving closed-loop safety control from perception to execution.

[0103] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A leak monitoring system for hazardous chemicals during transportation, characterized in that: include: A tank state sensing array is arranged along the outer and / or inner wall of the transport tank, including multiple heterogeneous sensor nodes, for real-time acquisition of multi-dimensional state parameters of the tank, including tank wall pressure distribution, tank wall temperature distribution, gas concentration in the gas phase space and tank structure vibration response. The transportation dynamic status acquisition module is installed on the transport vehicle to collect the vehicle's motion status parameters in real time. The motion status parameters include the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate, pitch rate, and travel speed. An interference signal prediction model, pre-trained based on historical transportation data, is used to predict the interference signal components that should be generated in the multi-dimensional state parameters under the current transportation dynamics, according to the motion state parameters. The residual extraction and leakage determination module is used to calculate the residual between the actual acquired value of the multidimensional state parameter and the predicted interference signal component output by the interference signal prediction model, and to determine leakage based on the residual.

2. The leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The interference signal prediction model includes a vibration interference prediction sub-model and a swaying interference prediction sub-model. The vibration interference prediction sub-model is used to predict the interference signal components generated by the vibration of the transport vehicle in the vibration response of the tank structure and the temperature distribution of the tank wall based on the vehicle's longitudinal acceleration, lateral acceleration, vertical acceleration, roll rate and pitch rate. The sloshing interference prediction sub-model is used to predict the interference signal components generated by the sloshing of liquid inside the tank in the pressure distribution of the tank wall, based on the vehicle's longitudinal acceleration, lateral acceleration, and driving speed, combined with the tank's geometric parameters and the current loaded liquid level.

3. A leakage monitoring system for hazardous chemicals during transportation according to claim 2, characterized in that: The swaying disturbance prediction sub-model is constructed based on a hybrid training method using computational fluid dynamics simulation and measured data. The hybrid training method includes: establishing a computational fluid dynamics simulation model based on the geometric parameters of the tank and a preset liquid level range, and generating simulation data of liquid sloshing pressure distribution under various transportation conditions; The motion state parameters and measured data of tank wall pressure distribution during actual transportation are collected. The measured data are then used to calibrate the parameters and correct the output of the computational fluid dynamics simulation model to obtain the swaying disturbance prediction sub-model.

4. A leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The determination process of the residual extraction and leakage determination module includes: The magnitude, duration, and spatial distribution pattern of the residual are used as features input to a pre-trained leakage determination classifier; The leakage determination classifier is trained based on residual data under normal transportation conditions and residual data under simulated leakage conditions, and is used to distinguish between residuals caused by sensor noise and model prediction errors and residuals caused by actual leakage. When the magnitude of the residual exceeds the adaptive detection threshold determined by the model prediction error distribution, and exhibits a spatial correlation pattern that conforms to the leakage diffusion law on multiple adjacent sensor nodes, it is determined to be a leakage event.

5. A leakage monitoring system for hazardous chemicals during transportation according to claim 4, characterized in that: The adaptive detection threshold is dynamically determined by the statistical characteristics of the prediction error of the interference signal prediction model; The interference signal prediction model continuously records the deviation between the predicted value and the actual interference signal during operation and updates the probability distribution parameters of the prediction error. The adaptive detection threshold is set as the preset confidence quantile of the prediction error probability distribution, so that the false alarm rate of leakage detection is maintained below the preset target false alarm rate under the condition of no leakage.

6. A leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The system also includes an online adaptive update module for the interference signal prediction model; The online adaptive update module is used to perform online incremental updates to the interference signal prediction model during transportation when it is determined that the current operating condition is a leak-free steady state. This is done using real-time collected motion state parameters and multi-dimensional state parameters to adapt to changes in tank loading status, tank structure aging, and differences in road surface characteristics of different transportation routes.

7. A leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The multiple heterogeneous sensor nodes in the tank state sensing array are divided into multiple sensor clusters according to the structural regions of the tank, and each sensor cluster covers a local area of ​​the tank. The residual extraction and leakage determination module calculates the local residual for each sensor cluster and locates the leakage source based on the spatial correlation of the residuals between adjacent sensor clusters. When the residuals of multiple adjacent sensor clusters simultaneously exceed their respective adaptive detection thresholds, and the residual amplitudes exhibit a spatial gradient decreasing outward from a certain sensor cluster, the tank area corresponding to that sensor cluster is determined to be the location of the leak source.

8. A leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The system also includes a loading status identification module; The loading status identification module is used to identify the current loading liquid level and liquid centroid position of the tank based on the vehicle dynamic response characteristics output by the transportation dynamic status acquisition module. The interference signal prediction model uses the identified loading liquid level and liquid centroid position as input parameters to output accurate interference signal predictions under different loading conditions.

9. A leakage monitoring system for hazardous chemicals during transportation according to claim 1, characterized in that: The transportation dynamic status acquisition module also includes a positioning unit and a road surface information acquisition unit; The positioning unit is used to obtain the real-time location of the transport vehicle; The road surface information acquisition unit is used to acquire road surface feature information of a preset road section in front of the vehicle based on the real-time location and a pre-built road surface smoothness database. The interference signal prediction model uses the road surface feature information as prior information and adjusts the model parameters for interference signal prediction in advance before the vehicle enters a road section with abrupt changes in road surface features.

10. A leakage monitoring system for hazardous chemicals during transportation according to any one of claims 1 to 9, characterized in that: The system also includes an on-board emergency response module; The vehicle-mounted emergency linkage module is connected to the residual extraction and leakage determination module. After a leakage event is determined, the severity of the leakage is determined based on the spatial distribution characteristics and amplitude level of the residual, and corresponding emergency response actions are executed. The emergency response actions include: closing the tank pipeline valves corresponding to the leak source location, activating the local spray device corresponding to the leak source location, reporting the leak location and severity information to the monitoring center, and controlling the vehicle to stop safely when the leak severity exceeds a preset threshold.

Images