A vehicle urea water moisture content analysis system

Abstract

The application relates to the technical field of chemical component analysis and measurement, and particularly discloses a vehicle-used urea water moisture content analysis system, which comprises a multi-physical field coupling sensing module, a dynamic physical property characteristic extraction module, a nonlinear temperature compensation model library, a self-adaptive model selection and fusion decision module and a moisture content calculation output module. By synchronously collecting conductivity, multi-point temperature and flow rate data, a dynamic physical property characteristic vector is extracted, and a preset special compensation model is adaptively selected or fused based on the characteristic vector, so that an accurate moisture content value is finally calculated. The scheme significantly improves the measurement accuracy and robustness of vehicle-used urea water moisture content analysis under complex temperature working conditions.

Description

Technical Field

[0001] This invention belongs to the field of chemical substance composition analysis and measurement technology, and specifically relates to a moisture content analysis system for automotive urea. Background Technology

[0002] In the field of diesel engine exhaust aftertreatment, selective catalytic reduction (SCR) technology is the core means of reducing nitrogen oxide emissions, which relies on the precise injection and decomposition of automotive urea aqueous solution. The water content of the urea solution is a key physical property parameter affecting its freezing point, storage stability, and final reduction efficiency. Rapid and accurate online analysis of this water content is of great significance for ensuring the reliable operation of the SCR system.

[0003] Detection methods based on the correlation between solution conductivity and moisture content have been widely studied and applied due to their fast response speed and ease of integration. The basic principle of this method is that the conductivity of urea solution changes regularly with its moisture content, and the moisture content can be indirectly calculated by measuring the conductivity value.

[0004] A fixed temperature compensation coefficient is used to correct the conductivity measurements to eliminate the influence of temperature on conductivity. However, automotive urea solutions undergo complex phase transitions over a wide temperature range, and the conductivity changes with temperature in a non-linear manner. Drastic fluctuations in ambient temperature can cause nonlinear abrupt changes in solution properties. In such cases, traditional temperature compensation models calibrated at a single point or within a narrow temperature range will produce significant adaptation errors, failing to accurately reflect the true correlation between conductivity and moisture content under sudden temperature changes, resulting in distorted moisture content analysis results. This sudden increase in measurement nonlinearity error under complex temperature variations limits the accuracy and reliability of the conductivity method in real-world automotive environments. Summary of the Invention

[0005] The purpose of this invention is to provide a moisture content analysis system for automotive urea, in order to solve the problem that existing technology based on conductivity measurement fails under wide temperature ranges, especially under conditions of drastic temperature fluctuations, due to the nonlinear abrupt change in the physical properties of urea solution causing the traditional temperature compensation model to fail, resulting in a serious decrease in the accuracy of moisture content analysis.

[0006] To achieve the above objectives, the present invention provides a vehicle urea moisture content analysis system, including a multi-physics field coupled sensing module, a dynamic physical property feature extraction module, a nonlinear temperature compensation model library, an adaptive model selection and fusion decision module, and a moisture content calculation and output module. The multi-physics coupling sensing module integrates a high-precision conductivity sensor, a multi-point temperature sensor array distributed along the flow direction and radial direction, and a flow velocity sensor. It simultaneously collects the conductivity, multi-point temperature, and flow velocity data of the solution flowing through the measurement cavity and marks them with a unified timestamp. The dynamic physical property feature extraction module is connected to the multi-physics field coupled sensing module, performs synchronous alignment and filtering on the sensing data, calculates the instantaneous temperature value, temperature change rate, and temperature drop amplitude, extracts the steady-state conductivity value, conductivity fluctuation variance, and conductivity temperature response slope, and generates a 5-dimensional dynamic physical property feature vector. The nonlinear temperature compensation model library has multiple pre-set dedicated compensation models for different temperature ranges. Each model is divided into ranges based on the phase change behavior of the entire temperature range. It has a built-in data table of the correspondence between conductivity, temperature and moisture content and a cubic spline interpolation algorithm, and is associated with applicable boundary conditions. The adaptive model selection and fusion decision module is connected to the dynamic physical property feature extraction module and the nonlinear temperature compensation model library. Based on the feature vector matching model applicability conditions, a single compensation model is selected in steady-state conditions, and multiple models are fused through initial weights and neural network confidence weighting in transient conditions to output the compensated conductivity value. The moisture content calculation output module is connected to the adaptive model selection and fusion decision module. Based on the compensated conductivity value and the benchmark lookup table, it obtains a preliminary moisture estimate and outputs the final moisture content value after adjustment by a minor correction factor.

[0007] Preferably, the high-precision conductivity sensor is used to measure the raw conductivity signal of the solution; the multi-point temperature sensor array monitors the solution inlet temperature, the core measurement area temperature, and the solution outlet temperature respectively; and the solution flow rate sensor is used to monitor the volumetric flow rate of the solution flowing through the measurement cavity.

[0008] Preferably, the dynamic property feature extraction module performs a moving average filter on the sensor data; the temperature change rate is calculated based on the most recent 2-second time window; the conductivity fluctuation variance is calculated based on the most recent 1-second time window; and the 5-dimensional dynamic property feature vector is composed of the instantaneous temperature value, temperature change rate, temperature drop amplitude, steady-state conductivity value, conductivity fluctuation variance, and conductivity temperature response slope.

[0009] Preferably, the method for calculating the conductivity temperature response slope is as follows: within one sampling period, the difference value of the conductivity signal and the difference value of the core measurement area temperature are acquired simultaneously; the conductivity difference value is divided by the temperature difference value, and the resulting quotient is smoothed by a first-order low-pass filter with a time constant of 0.2 seconds. After filtering out high-frequency noise, it is used as the conductivity temperature response slope.

[0010] Preferably, the nonlinear temperature compensation model library has a full temperature range of -40℃ to 85℃, and the temperature range is divided into at least 5 characteristic sub-ranges based on the inflection point of the conductivity-temperature curve of the solution, the crystallization phase transition point, and the supercooled state; the dedicated compensation model is a combination of database and interpolation algorithm; the applicability boundary conditions include temperature range, temperature change rate range, and conductivity signal stability threshold.

[0011] Preferably, the adaptive model selection and fusion decision module internally runs a decision engine based on rules and confidence evaluation; the decision engine matches the instantaneous temperature value and temperature change rate in the feature vector with the applicability boundary conditions of each dedicated compensation model one by one; When the feature vector falls completely within the applicable conditions of a certain dedicated compensation model, and the variance of the conductivity signal fluctuation is lower than the preset stability threshold, the dedicated compensation model is selected as the current effective model. When the eigenvector indicates that the operating condition is in a stage of drastic temperature change or in the boundary region where the applicable ranges of two dedicated compensation models are located, the model fusion mechanism is activated: first, the initial weights of each candidate model are calculated based on the distance between the eigenvector and the applicable conditions of each candidate model; then, the lightweight neural network evaluator is called to output the confidence scores of each candidate model; the initial weights and confidence scores are weighted together to obtain the fusion weights; and finally, the weighted sum of the compensation results of each candidate model is output as the compensated conductivity value.

[0012] Preferably, 1. the lightweight neural network evaluator is a fully connected network comprising an input layer, two hidden layers, and an output layer; the input layer has 5 nodes, corresponding to a 5-dimensional dynamic physical property feature vector; the two hidden layers have 8 and 4 nodes respectively; the output layer has the same number of nodes as the number of candidate models, and the output value of each node, after being normalized by the Softmax function, is the confidence score of the corresponding model; the neural network is trained offline using historical experimental datasets.

[0013] Preferably, the moisture content calculation output module stores a moisture content-conductivity reference lookup table that has been rigorously calibrated at a standard reference temperature; the trace correction factor is used to eliminate residual errors caused by temperature reference differences.

[0014] Preferably, the trace correction factor is determined through the following process: based on the temperature sub-interval corresponding to the finally selected effective compensation model, the difference between the center temperature of the interval and the standard reference temperature is obtained; based on the difference and the statistical average value of the conductivity temperature coefficient within the interval, a proportional correction coefficient is calculated; this proportional correction coefficient is applied to the preliminary moisture content estimate to complete the fine adjustment.

[0015] Preferably, the temperature sub-range specifically includes: The first temperature range is the critical region between supercooled liquid and crystallization, which is below -11 degrees Celsius. The second temperature range is the crystallization phase transformation influence zone from -11 degrees Celsius to -5 degrees Celsius; The third temperature range is a non-linear variation zone from -5 degrees Celsius to 20 degrees Celsius; The fourth temperature range is an approximately linear region from 20 degrees Celsius to 60 degrees Celsius; The fifth temperature range is the evaporation-affected zone above 60 degrees Celsius; The data table calibration density within each interval should be no less than one data point per degree Celsius.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention, by constructing a multi-physics coupled sensing module and a dynamic property feature extraction module, introduces transient features such as temperature change rate, temperature gradient, and conductivity dynamic response slope into the field of moisture analysis for automotive urea for the first time. These feature vectors can accurately characterize the nonlinear abrupt changes in solution properties over a wide temperature range, especially when temperatures fluctuate drastically, providing a comprehensive and accurate basis for subsequent intelligent compensation decisions, fundamentally changing the limitations of traditional methods that rely solely on single-point temperature for compensation.

[0017] 2. This invention innovatively establishes a nonlinear temperature compensation model library and an adaptive model selection and fusion decision mechanism. The system abandons a single global compensation formula and adopts a partitioned modeling approach to accurately describe the complex relationship between conductivity and temperature within different phase ranges. Through a rule-based and confidence-based decision engine, the system can dynamically select the optimal compensation model based on real-time operating conditions or intelligently fuse models under transient conditions. This ensures that high-precision temperature compensation is continuously provided under any temperature conditions, including the temperature transition region where errors in traditional methods suddenly increase, significantly improving the robustness and reliability of analysis across the entire temperature range.

[0018] 3. This invention eliminates systematic errors caused by differences in benchmarks between model intervals through a trace correction factor in the moisture content calculation output module for final calibration. The entire system forms a closed-loop, adaptive analysis chain, from dynamic sensing, feature extraction, intelligent model matching and fusion to accurate calculation. This solution effectively solves the core contradiction in the background technology where temperature fluctuations cause a sudden increase in measurement nonlinearity errors, enabling the conductivity-based automotive urea moisture content analysis system to truly adapt to complex and harsh in-vehicle environments, providing crucial technical support for the efficient and reliable operation of the SCR system. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical architecture of the vehicle urea moisture content analysis system proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the adaptive model selection and fusion decision-making mechanism in this invention; Figure 3 This is a flowchart illustrating the logical flow of multi-physics field coupled sensing and dynamic physical property feature extraction in this invention. Figure 4This is a schematic diagram illustrating the construction and partitioning principle of the nonlinear temperature compensation model library in this invention; Figure 5 This is a flowchart illustrating the logical process of moisture content calculation and trace correction in this invention. Detailed Implementation

[0020] Example 1: This invention provides a urea moisture content analysis system for vehicles. This system aims to solve the problem that existing conductivity-based technologies fail under wide temperature ranges, especially in conditions of drastic temperature fluctuations, due to the nonlinear abrupt changes in the physical properties of the urea solution causing the traditional temperature compensation model to fail, resulting in a significant decrease in the accuracy of moisture content analysis. Please refer to the appendix. Figures 1 to 5 This system constitutes a complete closed-loop analysis chain. Its core lies in achieving high-precision and robust analysis of the moisture content of automotive urea solutions through multi-physics field coupled sensing, dynamic feature extraction, intelligent model decision-making and fusion, and precise solution output. The following, with reference to the accompanying diagrams, provides a detailed explanation of the various components of the system and their collaborative operation.

[0021] Please refer to the attached document. Figure 1 With appendix Figure 3 The first core component of the system is the multiphysics coupling sensing module. This module is responsible for synchronously and in real-time acquiring raw sensing data of the automotive urea solution flowing through the measurement chamber. The measurement chamber is a fluid dynamics-designed pipe structure to ensure smooth flow of the solution and full contact with the sensors. This module integrates three types of sensors, forming a collaborative sensing array.

[0022] The first type is a high-precision conductivity sensor. Its electrodes are made of a special alloy material that is resistant to urea corrosion. It continuously measures the original conductivity signal of the solution flowing between the electrodes at a sampling frequency of 100 times per second. This signal is a basic physical quantity that reflects the concentration of ions in the solution and thus indirectly relates to the water content.

[0023] The second type is an embedded multi-point temperature sensor array. This array is distributed along the fluid flow direction and radial direction of the measurement cavity, containing at least three independent temperature measurement points. Specifically, the first temperature measurement point is located at the solution inlet of the measurement cavity, used to monitor the initial temperature of the solution entering the measurement area; the second temperature measurement point is located in the core measurement area of ​​the measurement cavity, which coincides with the electrode area of ​​the conductivity sensor, used to directly monitor the solution temperature at the location where the conductivity measurement occurs; the third temperature measurement point is located at the solution outlet of the measurement cavity, used to monitor the temperature of the solution flowing out of the measurement area. All temperature sensors operate at a sampling frequency of 10 times per second, and their distributed layout is designed to accurately capture the spatial temperature gradient and instantaneous temporal changes of the solution as it flows through the entire measurement cavity, providing multi-dimensional temperature information for subsequent analysis.

[0024] The third type is the solution flow rate sensor, which typically uses a turbine or ultrasonic flow meter. It is installed on the inlet or outlet pipe of the measuring chamber to monitor the volumetric flow rate of the solution flowing through the chamber in real time, measured in liters per minute. Flow rate data is not only used to monitor the system's operating status, but more importantly, it provides a time reference for subsequent signal synchronization processing and dynamic effect correction, such as compensating for sensor response hysteresis caused by changes in flow rate.

[0025] The multiphysics coupled sensing module integrates a high-speed synchronous sampling circuit and a microcontroller unit. This microcontroller unit coordinates the sampling clocks of the three sensor subsystems, ensuring that conductivity data, multi-point temperature data, and flow velocity data are synchronously acquired and tagged with a unified timestamp at every millisecond-level timestamp, forming a set of time-aligned raw data packets. These packets are then transmitted in real-time to the downstream dynamic property feature extraction module via an internal high-speed serial communication interface in a fixed data frame format. The data frame format includes fields such as a frame header, timestamp, raw conductivity value, raw values ​​from temperature sensors 1 to 3, raw flow velocity value, and frame check sequence, ensuring the integrity and reliability of data transmission.

[0026] Please continue to refer to the appendix. Figure 3 The second core component of the system is the dynamic physical property feature extraction module. This module connects to and receives raw data packets from the multiphysics coupled sensing module via a high-speed communication interface. Its core task is to preprocess, calculate features, and extract higher-order features from these raw, time-noise-laden sensing data, ultimately generating a feature vector that comprehensively characterizes the dynamic physical property state of the solution under the current operating conditions. The operation of this module consists of several sequentially executed sub-stages.

[0027] The first stage is the data synchronization, alignment, and preprocessing sub-stage. This sub-stage first performs precise time alignment between the conductivity signal and the signals from the three temperature sensors based on the unified timestamp in the data packet. Since the conductivity sampling frequency is 100 times per second, while the temperature sampling frequency is 10 times per second, this sub-stage uses a linear interpolation algorithm to upsample the temperature data to 100 times per second, maintaining strict point-to-point synchronization with the conductivity data. Next, a moving average filter is applied to both the synchronized conductivity sequence and each temperature sequence. The length of the moving window is configured according to the expected noise characteristics of the signal; for the conductivity signal, the window length is typically set to 10 sampling points (0.1 seconds); for the temperature signal, the window length is typically set to 5 sampling points (0.5 seconds). This filtering operation effectively suppresses high-frequency random interference introduced by sensor circuit thermal noise, fluid turbulence noise, etc., and extracts the trend components in the signal.

[0028] The second stage is the temperature characteristic calculation sub-stage. This sub-stage performs calculations based on the filtered temperature data sequence. First, the filtered value from the temperature sensor in the core measurement area is used as the instantaneous temperature value at the current moment. This is a scalar. Next, calculate the rate of change of this instantaneous temperature value over the most recent 2-second time window. The specific algorithm is as follows: Take the historical temperature value corresponding to the 200th sampling point before the current time. Calculate the current instantaneous temperature value and The difference, divided by the 2-second time interval, yields the rate of temperature change in degrees Celsius per second. This parameter directly quantifies the severity of the temperature change. Simultaneously, this sub-stage calculates the temperature drop as the solution flows through the measurement chamber. Its value is the inlet temperature sensor filter value minus the outlet temperature sensor filter value. This parameter reflects the degree of heat exchange between the solution and the environment during the measurement process.

[0029] The third stage is the dynamic feature extraction sub-stage of conductivity. This sub-stage is based on the filtered conductivity signal sequence. Calculations were performed to extract three key features. The first feature is the steady-state conductivity value at the current moment. The last value of the filtered conductivity sequence is directly used. The second characteristic is the variance of the conductivity signal's fluctuation within the most recent one-second time window. The specific algorithm is: take the most recent 100... The value is used to calculate its variance. This variance reflects the short-term stability of the conductivity measurement; a high variance may indicate uneven fluid mixing, bubble interference, or poor sensor contact. The third characteristic is the instantaneous response slope of the conductivity signal relative to the rate of temperature change in the core measurement area. This feature is used to capture the dynamic sensitivity of conductivity to temperature changes. Specifically, the calculation method involves synchronously calculating the difference in the conductivity signal within a sampling period of 0.01 seconds. The difference between the temperature and the core measurement area .Will The original instantaneous slope quotient is obtained. Since the original quotient is susceptible to noise, it is input into a first-order low-pass digital filter with a time constant of 0.2 seconds for smoothing. The stable output value after filtering out high-frequency noise is used as the final instantaneous response slope characteristic. This parameter is particularly important when temperature changes rapidly, as it reveals the dynamic characteristics of changes in physical properties. Finally, the dynamic physical property feature extraction module extracts the five parameters calculated above—instantaneous temperature value—from these parameters. Temperature change rate Temperature drop range steady-state conductivity value conductivity fluctuation variance Instantaneous response slope — Encapsulated as a 5-dimensional dynamic physical property feature vector The feature vector is updated every 0.01 seconds and pushed to the adaptive model selection and fusion decision module in real time via the internal data bus.

[0030] Please refer to the attached document. Figure 1 With appendix Figure 4 The third core component of the system is a nonlinear temperature compensation model library. This module is not a real-time computing unit, but rather a pre-built, static database and algorithm collection stored in the system's non-volatile memory. The model library is built based on a systematic experimental study of the phase transition behavior of automotive urea solution across the entire temperature range of -40°C to 85°C. The research focuses on analyzing the conductivity-temperature relationship curve of the solution, identifying nonlinear inflection points, abrupt change points, and characteristic temperature points related to phase transitions.

[0031] Based on these characteristics, the entire temperature range is divided into at least five sub-ranges with different physical properties. The specific division criteria include: the first temperature range is below -11°C, the critical region for supercooled liquid and crystallization, where the solution is in a metastable state and its conductivity is extremely sensitive to temperature and disturbances; the second temperature range is from -11°C to -5°C, the region affected by crystallization phase transitions, where crystallization or dissolution processes may occur, and the conductivity exhibits significant non-monotonicity; the third temperature range is from -5°C to 20°C, a nonlinear variation region, where conductivity increases with increasing temperature, but the rate of increase is nonlinear; the fourth temperature range is from 20°C to 60°C, an approximately linear region, where the relationship between conductivity and temperature is relatively stable and close to linear; and the fifth temperature range is above 60°C, the evaporation-affected region, where the water evaporation effect begins to appear, affecting solution concentration and conductivity.

[0032] For each predefined temperature sub-interval, a precise correspondence table is established between conductivity values, measured temperature, and moisture content within that interval through high-precision laboratory calibration. The calibration process is conducted in a constant-temperature bath, controlling the temperature variation in increments of no more than 0.1 degrees Celsius. At each stable temperature point, the conductivity of a series of standard urea solution samples with known precise moisture content is measured. The calibration density of the data table within each sub-interval is no less than one data point per degree Celsius, with a higher density near key inflection points. Based on this data table, a dedicated compensation model is defined for each sub-interval. Each dedicated compensation model is essentially an algorithm encapsulation, with cubic spline interpolation as its core.

[0033] When inputting the measured conductivity value and measured temperature value At that time, the model finds several data rows in the data table corresponding to the temperature sub-interval that are closest to T_meas. Then, based on the conductivity-moisture content relationship in these data rows, it uses a cubic spline interpolation algorithm to calculate the corresponding... The moisture content value. Each dedicated compensation model is associated with a model applicability description, stored in a data structure, which defines the boundary conditions under which the model can operate optimally. These conditions typically include: effective temperature range. , This is the temperature sub-interval corresponding to the model; the tolerance range for the rate of temperature change. , For example, linear region models can tolerate larger rates of change, while phase transition region models are limited to a smaller range; conductivity signal stability threshold. This refers to the upper limit of the allowable variance in conductivity fluctuation. The nonlinear temperature compensation model library provides a model query and invocation interface for the upstream decision-making module.

[0034] Please refer to the attached document. Figure 1 With appendix Figure 2 The fourth core component of the system is the adaptive model selection and fusion decision module. This module, connected to the dynamic physical property feature extraction module and the nonlinear temperature compensation model library via a high-speed data path, serves as the system's intelligent decision-making center. Its core function is to intelligently select or fuse the most suitable compensation model based on real-time acquired dynamic physical property feature vectors, and generate the final conductivity value for water decomposition calculations after adaptive nonlinear temperature compensation. Internally, this module operates a decision engine based on rules and confidence assessment, and its workflow includes multiple judgment and calculation stages.

[0035] The decision engine first receives the latest 5-dimensional dynamic physical property feature vector. The first step is model screening and rule matching. The decision engine reads the applicability descriptions of all dedicated compensation models in the nonlinear temperature compensation model library. It then uses the instantaneous temperature values ​​from the feature vectors. and rate of temperature change Each model's effective temperature range and tolerance range for temperature change rate are compared individually. Simultaneously, the variance of conductivity fluctuations is examined. Is it below the stability threshold set by the model? When a specific compensation model exists, such that... , It falls completely within its scope of application, and When the conductivity is below its threshold, the decision engine determines that the current operating condition is within a stable and typical physical property range. At this point, the decision engine directly selects the dedicated compensation model as the current effective model and outputs the model's response to the current measured conductivity. The median moisture content obtained after compensation calculation is the equivalent conductivity value corresponding to this median value, which is the compensated conductivity value. The second step is model fusion decision-making under transient operating conditions. When the eigenvectors indicate that the operating condition is unstable, such as the rate of temperature change... The absolute value exceeds the preset transient threshold of 0.5 degrees Celsius per second, or the instantaneous temperature value. When the decision engine is in the overlapping region where the applicability ranges of two dedicated compensation models overlap, it will initiate a complex model fusion mechanism. The goal of this mechanism is to integrate the advantages of multiple models, smoothly transition, and avoid output jumps caused by hard model switching. The fusion mechanism first calculates initial weights based on the "distance" between the feature vectors and the applicability conditions of each candidate model. Taking temperature as an example, the calculation... The normalized distance to the center point of the temperature range of each candidate model is used; the closer the distance, the higher the initial weight. Simultaneously, the following is considered: The degree of matching with the model's tolerance range. Assuming there are N candidate models, this step generates a set of initial weights. , The weights range from 1 to N, with a sum of 1. Next, the decision engine calls a pre-built lightweight neural network evaluator. This evaluator is a pre-trained, fully connected neural network with the following structure: the input layer has 5 nodes, corresponding to a 5-dimensional dynamic property feature vector; the first hidden layer contains 8 nodes, using the ReLU activation function; the second hidden layer contains 4 nodes, also using the ReLU activation function; the output layer has the same number of nodes as the number of candidate models N in the model library, and each node outputs a raw score. The raw scores of the output layer are normalized using the Softmax function, transforming them into a set of probability distributions, i.e., the confidence score of each model's compensation accuracy under the current transient condition. , From 1 to N, and all The sum is 1. The training data for this neural network comes from historical experimental datasets, which contain dynamic physical property feature vectors collected under various complex temperature variations, as well as real moisture content values ​​obtained through high-precision benchmark methods. By calculating the compensation error of each dedicated compensation model under this feature vector and using the true error value as a supervision signal, the neural network is trained to predict the error magnitude of each model based on the feature vector, and then converts it into confidence. Finally, the decision engine uses the initial weights calculated by the rules. Confidence score of neural network output A weighted fusion is performed. The fusion formula uses a weighted average to calculate the final fusion weight for each candidate model. A specific weighted aggregation formula is expressed as follows:

[0036] in, It is an adjustable parameter, ranging from 0 to 1, used to balance the influence of rule-based prior knowledge and data-driven confidence. When A value of 0.5 indicates that both are equally important. Ultimately, the decision engine will combine the candidate models according to their fusion weights. For the current measured conductivity The intermediate results obtained after compensation are weighted and summed. Assume the equivalent conductivity corresponding to the i-th model after compensation is... The final compensated conductivity value Calculated by the following formula:

[0037] this The value is the "pure" conductivity signal after adaptive nonlinear temperature compensation and intelligent fusion processing, eliminating temperature nonlinear interference. The decision engine will... The final model identification information, along with the output of the model, is then sent to the moisture content calculation output module.

[0038] Please refer to the attached document. Figure 1 With appendix Figure 5 The fifth core component of the system is the moisture content calculation and output module. This module receives two key inputs from the adaptive model selection and fusion decision module: the final conductivity value after adaptive compensation. And the instantaneous temperature value of the core measurement area extracted from the dynamic physical property feature vector. This module stores a table at a standard reference temperature. Below is a rigorously calibrated moisture content-conductivity reference table. Standard reference temperature. 25 degrees Celsius is typically chosen as the widely accepted calibration benchmark. The benchmark lookup table was obtained through extensive experiments at a constant temperature of 25 degrees Celsius, establishing a high-precision one-to-one mapping between conductivity values ​​and moisture content percentages. The data points are dense enough to support high-precision linear interpolation. The calculation output module operates in two sub-steps. The first step is benchmark lookup and preliminary estimation. The module uses the received data... The value is looked up and interpolated in the baseline lookup table. Because... After temperature compensation, theoretically its corresponding moisture content should be directly equal to that at... The moisture content corresponding to the same conductivity value measured at the reference temperature. Therefore, by using an interpolation algorithm, a preliminary estimate of the moisture content at the reference temperature can be quickly obtained. The second step is minor correction and final output. Despite adaptive compensation, the initial estimate still has some minor bias due to model interval partitioning, residual errors in the interpolation algorithm, and small deviations introduced by model fusion. There may still be very minor systematic errors.

[0039] To achieve the highest accuracy, the module introduces a minor correction factor based on the characteristics of the final selected or fused model. The process for determining this factor is as follows: First, based on the model identification information transmitted from the decision module, the temperature sub-interval corresponding to the final effective compensation model is determined. The center temperature of this interval is then obtained. Calculate the difference between the center temperature and the standard reference temperature. Then, from the calibration data of this temperature sub-range, the average value of the temperature coefficient of conductivity within this range is statistically calculated. Minor correction factor Then according to and The calculation, its function is to make proportional adjustments to the preliminary valuation; the calculation formula is logically equivalent to... ,in This is an experimentally determined attenuation coefficient, less than 1, used to characterize the proportional relationship between the compensated residual error and the theoretical temperature difference error. Ultimately, this yields a precise percentage value for the moisture content. Determined by the following formula: This step completely eliminates the step error or minute drift that might be caused by benchmark inconsistencies resulting from model range switching or fusion. Ultimately, Through the system's standard onboard communication interface, such as the controller area network bus, data is uploaded in real time to the vehicle's selective catalytic reduction system controller in a specified message format and transmission cycle for precise control of urea injection quantity. The entire system begins with synchronous acquisition of multi-physics field data, undergoing dynamic feature extraction, intelligent model decision-making and fusion, precise calculation and minor correction, forming a complete, closed-loop, and adaptive analysis chain to ensure stable, reliable, and high-precision moisture content data output under various complex temperature environments.

[0040] Example 2: This example provides another specific implementation scheme for the vehicle urea moisture content analysis system. Its core architecture is consistent with that of Example 1, but the feature calculation strategy of the dynamic physical property feature extraction module and the fusion weight generation mechanism of the adaptive model selection and fusion decision module have been optimized and expanded, aiming to further improve the system's response speed and fusion smoothness under extreme rapid temperature change conditions.

[0041] In the dynamic physical property feature extraction module, in addition to the 5-dimensional basic feature vectors described in Example 1, this example adds statistical features based on a short-term historical window to more richly characterize the trend of physical property changes. Specifically, the module maintains a 3-second first-in-first-out historical buffer to store the 5-dimensional basic feature vectors calculated at each moment within the past 3 seconds. After a new feature vector is generated, the module additionally calculates two sets of statistics. The first set is trend statistics: for the instantaneous temperature sequence within the buffer, its linear regression slope is calculated as the average temperature change trend within the past 3 seconds. For the steady-state conductivity sequence within the buffer zone, the slope of its linear regression is also calculated as the average trend of conductivity change over the past 3 seconds. .

[0042] These two trend quantities can distinguish whether the temperature is continuously rising, falling, or fluctuating, and reveal the overall direction of conductivity changes. The second set is extreme value statistics: extracting the temperature change rate sequence within the buffer zone and finding its absolute maximum value. This value represents the most dramatic temperature change shock that has occurred in the last 3 seconds; extract the conductivity fluctuation variance sequence within the buffer and find its maximum value. This value represents the most unstable measurement period that occurred within the last 3 seconds. These newly added statistical features, together with real-time features, constitute an expanded 9-dimensional feature vector, providing a more comprehensive basis for the decision engine.

Claims

1. A system for analyzing the moisture content of automotive urea, characterized in that, It includes a multi-physics coupling sensing module, a dynamic physical property feature extraction module, a nonlinear temperature compensation model library, an adaptive model selection and fusion decision module, and a moisture content calculation and output module; The multi-physics coupling sensing module integrates a high-precision conductivity sensor, a multi-point temperature sensor array distributed along the flow direction and radial direction, and a flow velocity sensor. It simultaneously collects the conductivity, multi-point temperature, and flow velocity data of the solution flowing through the measurement cavity and marks them with a unified timestamp. The dynamic physical property feature extraction module is connected to the multi-physics field coupled sensing module, performs synchronous alignment and filtering on the sensing data, calculates the instantaneous temperature value, temperature change rate, and temperature drop amplitude, extracts the steady-state conductivity value, conductivity fluctuation variance, and conductivity temperature response slope, and generates a 5-dimensional dynamic physical property feature vector. The nonlinear temperature compensation model library has multiple pre-set dedicated compensation models for different temperature ranges. Each model is divided into ranges based on the phase change behavior of the entire temperature range. It has a built-in data table of the correspondence between conductivity, temperature and moisture content and a cubic spline interpolation algorithm, and is associated with applicable boundary conditions. The adaptive model selection and fusion decision module is connected to the dynamic physical property feature extraction module and the nonlinear temperature compensation model library. Based on the feature vector matching model applicability conditions, a single compensation model is selected in steady-state conditions, and multiple models are fused through initial weights and neural network confidence weighting in transient conditions to output the compensated conductivity value. The moisture content calculation output module is connected to the adaptive model selection and fusion decision module. Based on the compensated conductivity value and the benchmark lookup table, it obtains a preliminary moisture estimate and outputs the final moisture content value after adjustment by a minor correction factor.

2. The vehicle urea moisture content analysis system according to claim 1, characterized in that, The high-precision conductivity sensor is used to measure the raw conductivity signal of the solution; the multi-point temperature sensor array monitors the solution inlet temperature, the core measurement area temperature and the solution outlet temperature respectively; the solution flow rate sensor is used to monitor the volumetric flow rate of the solution flowing through the measurement cavity.

3. The vehicle urea moisture content analysis system according to claim 1, characterized in that, The dynamic physical property feature extraction module performs a moving average filter on the sensor data; the temperature change rate is calculated based on the most recent 2-second time window; the conductivity fluctuation variance is calculated based on the most recent 1-second time window; the instantaneous temperature value, temperature change rate, temperature drop amplitude, steady-state conductivity value, conductivity fluctuation variance, and conductivity temperature response slope together constitute a 5-dimensional dynamic physical property feature vector.

4. The vehicle urea moisture content analysis system according to claim 3, characterized in that, The method for calculating the conductivity temperature response slope is as follows: within one sampling period, the difference value of the conductivity signal and the difference value of the temperature in the core measurement area are acquired simultaneously; the conductivity difference value is divided by the temperature difference value, and the resulting quotient is smoothed by a first-order low-pass filter with a time constant of 0.2 seconds. After filtering out high-frequency noise, it is used as the conductivity temperature response slope.

5. The vehicle urea moisture content analysis system according to claim 1, characterized in that, The nonlinear temperature compensation model library has a full temperature range of -40℃ to 85℃. The temperature range is divided into at least 5 characteristic sub-ranges based on the inflection point of the conductivity-temperature curve of the solution, the crystallization phase transition point, and the supercooled state. The dedicated compensation model is a collection of database and interpolation algorithms. The applicability boundary conditions include the temperature range, the temperature change rate range, and the conductivity signal stability threshold.

6. The vehicle urea moisture content analysis system according to claim 1, characterized in that, The adaptive model selection and fusion decision module internally runs a decision engine based on rules and confidence assessment; the decision engine matches the instantaneous temperature value and temperature change rate in the feature vector with the applicability boundary conditions of each dedicated compensation model one by one; When the feature vector falls completely within the applicable conditions of a certain dedicated compensation model, and the variance of the conductivity signal fluctuation is lower than the preset stability threshold, the dedicated compensation model is selected as the current effective model. When the eigenvector indicates that the operating condition is in a stage of drastic temperature change or in the boundary region where the applicable ranges of two dedicated compensation models are located, the model fusion mechanism is activated: first, the initial weights of each candidate model are calculated based on the distance between the eigenvector and the applicable conditions of each candidate model; then, the lightweight neural network evaluator is called to output the confidence scores of each candidate model; the initial weights and confidence scores are weighted together to obtain the fusion weights; and finally, the weighted sum of the compensation results of each candidate model is output as the compensated conductivity value.

7. The vehicle urea moisture content analysis system according to claim 6, characterized in that, 1. The lightweight neural network evaluator is a fully connected network comprising an input layer, two hidden layers, and an output layer; the input layer has 5 nodes, corresponding to a 5-dimensional dynamic physical property feature vector; the two hidden layers have 8 and 4 nodes, respectively; the output layer has the same number of nodes as the number of candidate models, and the output value of each node, after being normalized by the Softmax function, becomes the confidence score of the corresponding model; the neural network is trained offline using historical experimental datasets.

8. The vehicle urea moisture content analysis system according to claim 1, characterized in that, The moisture content calculation output module stores a moisture content-conductivity reference lookup table calibrated at a standard reference temperature; the trace correction factor is used to eliminate residual errors caused by temperature reference differences.

9. The vehicle urea moisture content analysis system according to claim 8, characterized in that, The trace correction factor is determined through the following process: based on the temperature sub-interval corresponding to the finally selected effective compensation model, the difference between the center temperature of the interval and the standard reference temperature is obtained; based on the difference and the statistical average value of the conductivity temperature coefficient within the temperature sub-interval, a proportional correction coefficient is calculated; this proportional correction coefficient is applied to the preliminary moisture content estimate to complete the fine adjustment.

10. The vehicle urea moisture content analysis system according to claim 9, characterized in that, The temperature sub-range specifically includes: The first temperature range is the critical region between supercooled liquid and crystallization, which is below -11 degrees Celsius. The second temperature range is the crystallization phase transformation influence zone from -11 degrees Celsius to -5 degrees Celsius; The third temperature range is a non-linear variation zone from -5 degrees Celsius to 20 degrees Celsius; The fourth temperature range is an approximately linear region from 20 degrees Celsius to 60 degrees Celsius; The fifth temperature range is the evaporation-affected zone above 60 degrees Celsius; The data table calibration density within each interval should be no less than one data point per degree Celsius.

Images