Automatic Measurement and Analysis Method for Bedside Intake and Exit Volumes Based on Multi-Source Sensor Fusion

Abstract

This invention discloses a multi-source sensor fusion-based method for automatic bedside intake and output measurement and analysis, belonging to the field of intelligent clinical monitoring technology. It simultaneously acquires cumulative weight increment data and fluid level increment data within the same time slice; processes the two sets of data to calculate the temporal deviation and dynamic proportionality coefficient between the rate of gravity increment and the rate of fluid level rise; extracts delayed features and jointly maps them to a clinical fluid characteristic database for fluid characteristic classification and identification; based on the hit features, it obtains the corresponding dynamic density base and wall adhesion compensation coefficient, performs nonlinear compensation conversion, and compares with historical characteristic classification results to trigger an early warning. This invention can adaptively and accurately measure various body fluid characteristics without frequent manual calibration, effectively reducing errors and simultaneously providing clinical alerts.

Description

Technical Field

[0001] This invention relates to the field of intelligent clinical monitoring technology, specifically to a method for automatic measurement and analysis of bedside intake and output using multi-source sensor fusion. Background Technology

[0002] In intensive care and postoperative rehabilitation, accurate measurement and dynamic monitoring of bedside fluid intake and output are core indicators for clinicians to assess organ function, fluid balance, infection control progress, and postoperative bleeding risk. Traditional fluid intake and output measurement mainly relies on manual timed observation with measuring cups or automated equipment using single sensors. However, existing automated measurement equipment reveals many serious technical shortcomings when faced with complex clinical scenarios.

[0003] First, traditional measurement algorithms generally employ idealized physical models based on "fixed parameter conversion." Existing systems assume fluid density and flow characteristics are constants and rely solely on independent conversions of weight or liquid level to derive the total volume. However, actual postoperative drainage, hematuria, or discharge from critically ill patients often contain blood clots, necrotic tissue, purulent secretions, or exhibit high viscosity. The physical properties of such fluids differ significantly from those of pure urine. When the physical properties of the discharge undergo dynamic abrupt changes, the algebraic relationship used to convert weight to volume using fixed constants will result in a severe mismatch between the algebraic relationship and the actual liquid level reading within the drainage container, leading to substantial cumulative volume measurement errors.

[0004] Secondly, existing technologies fail to address the rheological interference issues associated with complex fluids. High-viscosity drainage fluids or purulent secretions exhibit significant non-Newtonian fluid characteristics, resulting in severe "wall adhesion" and hysteresis effects on liquid level fluctuations when flowing through drainage pipes or dripping into drainage containers. This portion of the fluid contributes to the system's total weight, but due to the viscous trapping at the pipe walls, it has not yet reached the bottom to cause a true rise in the liquid level. Traditional systems rely solely on direct linear division for volume conversion, completely ignoring the implicit loss volume trapped by the pipe walls. This leads to significantly underestimated calculation results, making them unsuitable for complex drainage scenarios involving strong viscosity and wall adhesion.

[0005] Furthermore, traditional equipment suffers from extremely limited sensing dimensions and very low information utilization. Existing systems merely use sensor data for simple "volume accumulation," completely ignoring the response time and proportional deviations exhibited by sensors when capturing different fluids. In fact, these deviations themselves contain valuable microscopic physical information about the fluid. Because these dynamic deviations cannot be used to reverse-engineer the evolution of drainage fluid properties, current technology misses a valuable opportunity to provide early warnings of clinical crisis events through property mutations. For example, when a patient experiences occult postoperative hemorrhage or secondary severe infection, the drainage fluid will exhibit a sharp increase in density or a significant increase in viscosity, respectively. Existing equipment, unable to sense these evolutions in microscopic physical characteristics, naturally cannot provide alarms for trend mutations. Medical staff still need to frequently patrol and assess the drainage fluid properties by visual observation, which not only consumes a large amount of nursing manpower but also results in significant diagnostic and treatment delays. Summary of the Invention

[0006] The purpose of this invention is to provide a method for automatic measurement and analysis of bedside intake and output based on multi-source sensor fusion, so as to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for automatic measurement and analysis of bedside intake and output volume based on multi-source sensor fusion, comprising the following steps:

[0008] S1. Obtain the cumulative weight increment data of the discharged material collected by the weight sensor within the same time slice, and the liquid level height increment data collected by the liquid level intelligent sensor.

[0009] S2. Process the cumulative weight increment data and the liquid level increment data, calculate the temporal deviation between the gravity increment rate and the liquid level rise rate, and the dynamic proportional coefficient between the gravity increment rate and the liquid level rise rate.

[0010] S3. Determine whether the dynamic proportional coefficient deviates from the benchmark pure water density threshold, extract the delay feature of the time deviation, and jointly map the dynamic proportional coefficient and the delay feature to a preset clinical liquid feature library for fluid property classification and identification to obtain property classification results.

[0011] S4. Based on the feature entries hit by the trait classification results, obtain the corresponding dynamic density base and wall-hanging compensation coefficient, perform nonlinear compensation conversion on the weight-volume dimension, and calculate the actual volume result of a single discharge after adaptive calibration and the cumulative amount of classification within a specified time.

[0012] S5. Compare the current trait classification result with the historical trait classification result over time. When it is determined that the trait classification result has undergone a cross-level mutation, trigger a trait mutation warning.

[0013] Furthermore, S1 also includes: aligning the time series of the weight sensor and the liquid level smart sensor with timestamps to extract data within the same time slice; the time series deviation is the time difference between the liquid level response and the weight change; and the dynamic proportional coefficient is the real-time equivalent density.

[0014] Furthermore, in S2, the calculation steps for the timing deviation and dynamic scaling factor include:

[0015] The rate of gravity increment is obtained by calculating the first derivative of the cumulative weight increment data.

[0016] The rate of liquid level rise is obtained by calculating the first derivative of the liquid level height increment data.

[0017] The ratio of the gravity increment rate to the liquid level rise rate is used as the dynamic proportionality coefficient.

[0018] Furthermore, in step S3, when determining whether the deviation from the baseline pure water density threshold is met, the pure water density needs to be dynamically calibrated in conjunction with the ambient temperature. When the absolute value of the deviation difference between the dynamic proportional coefficient and the baseline pure water density threshold is greater than the preset allowable deviation range, the fluid property classification and identification is initiated. When the absolute value of the deviation difference is less than or equal to the preset allowable deviation range, the fluid property classification and identification is skipped, and the property classification result is directly determined as pure urine.

[0019] Furthermore, the clinical fluid characteristic database includes: pure urine, serous drainage fluid, high-viscosity drainage fluid, bloody drainage fluid, and purulent drainage fluid, along with their corresponding density ranges and wall adhesion compensation coefficients; the bloody drainage fluid is graded according to bleeding concentration.

[0020] Furthermore, the delay feature characterizes the response hysteresis caused by fluid adhesion to the wall; the mapping and identification process in S3 is as follows: the dynamic scaling factor and the delay feature are used as multidimensional sensing response features, and similarity matching is performed with the features in the clinical fluid feature library to match the corresponding fluid properties.

[0021] Furthermore, the formula for the nonlinear compensation conversion is:

[0022]

[0023] in, For the actual volume result, The cumulative weight increment data, The dynamic density base number is... The wall-mounted compensation coefficient is denoted as .

[0024] Furthermore, the calculation of the cumulative classification amount within a specified time period includes: establishing independent classification and measurement ledgers according to the classification results of different traits; and accumulating the actual volume results into the corresponding classification and measurement ledgers to generate a 24-hour cumulative classification amount.

[0025] Furthermore, the time-series comparison in S5 includes: establishing the time-series evolution trajectory of the effluent traits, and calculating the difference in trait grade between the current trait classification result and the historical trait classification result within a preset time window.

[0026] Furthermore, the triggering conditions for the trait mutation warning include: a sudden bleeding warning caused by a sharp increase in density from serous mutation to severe bloody isodense, or an infection aggravation warning caused by an increase in viscosity due to the increase in the delayed feature.

[0027] Compared with existing technologies, the multi-source sensor fusion-based automatic bedside intake and output measurement and analysis method provided by this invention has the following advantages:

[0028] First, this approach breaks through the limitations of traditional fixed-parameter models, achieving adaptive and accurate measurement in complex fluid scenarios. This scheme extracts the dynamic proportionality coefficient and corresponding temporal deviation of cumulative weight increment data and liquid level increment data, then reverses this to deduce the fluid's dynamic density baseline and wall-attachment compensation coefficient, and applies a nonlinear compensation conversion formula to adaptively calibrate the volume of a single discharge. This logic completely solves the hidden retention and calculation loss errors caused by the adhesion of purulent secretions, high-viscosity drainage fluid, and bloody drainage fluid as they flow through pipelines, breaking the measurement blind spot of traditional single-sensor measurement when dealing with sudden changes in physical properties. It is expected to significantly reduce the volume measurement error of traditional fixed-parameter methods from 15%-30% to less than 5%.

[0029] Secondly, by employing a reverse morphology identification technique based on multidimensional sensing response bias, fluid property extraction is achieved with "zero hardware addition." The system cleverly utilizes the physical response time difference between fluids of different viscosities exhibiting increased gravity and rising liquid surface speed, combining this with real-time equivalent density and jointly mapping it to a clinical fluid feature database for pattern recognition and matching. This eliminates the need for additional expensive microfluidic viscometers or precision densitometers; the system accurately distinguishes between pure urine, serous, purulent, and bloody drainage fluids of varying concentrations using only a basic dual-channel signal cross-comparison algorithm based on weight and liquid level sensors. This significantly enhances the diagnostic value and intelligence of the equipment while effectively controlling manufacturing costs.

[0030] Third, a time-series evolution tracking model for the characteristics of excreted fluid was constructed, filling a clinical gap in early intelligent warning of drainage fluid status. Based on accurate volume measurement, this invention further establishes a time-series window comparison mechanism for characteristic classification results. When the system detects a rapid, cross-level mutation in the patient's drainage fluid, it can automatically trigger a clinical status mutation warning signal. This function upgrades the measuring instrument from a simple "volume recording tool" to a "bedside intelligent monitoring node," effectively assisting medical staff in early identification of potential hidden complications, significantly shortening the golden rescue detection period for critically ill patients, and significantly ensuring patient safety.

[0031] Fourth, it provides automated, structured classification and measurement ledgers, offering objective quantitative support for clinical evidence-based medicine decision-making. Unlike traditional equipment that only provides the total drainage volume in a mixed state, this system uses a fluid property classification and identification algorithm to independently accumulate the actual volume results within the same time slice into the corresponding classification and measurement ledger, automatically generating classification statistical reports based on different fluid properties. Medical staff can intuitively obtain the independent volume evolution trajectory of various fluids over 24 hours, providing clinicians with multi-dimensional and high-precision data quantification for objectively judging the effectiveness of internal drainage, accurately assessing wound bleeding, and monitoring infection control progress.

[0032] Fifth, the system incorporates a high-precision spatiotemporal alignment and environmental calibration mechanism at the system's underlying level to ensure absolute accuracy in feature extraction. During the underlying data processing, this solution pre-aligns the timestamps of the sensors to eliminate digital misalignment errors caused by asynchronous hardware sampling and transmission lag. Furthermore, when calculating threshold deviations, it dynamically calibrates the baseline pure water density in conjunction with ambient temperature. These underlying data cleaning and thermodynamic calibration mechanisms effectively filter out invalid disturbances caused by environmental temperature drift and equipment noise, ensuring the overall technical solution exhibits extremely high system robustness and anti-interference capabilities in the complex environment of a hospital ward. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0034] Figure 1 This is a flowchart illustrating the overall architecture of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0036] Please see Figure 1A method for automatic measurement and analysis of bedside intake and output based on multi-source sensor fusion includes the following steps:

[0037] S1. Obtain the cumulative weight increment data of the discharged material collected by the weight sensor within the same time slice, and the liquid level height increment data collected by the liquid level intelligent sensor.

[0038] S2. Process the cumulative weight increment data and liquid level increment data, calculate the temporal deviation between the gravity increment rate and the liquid level rise rate, and the dynamic proportionality coefficient between the gravity increment rate and the liquid level rise rate.

[0039] S3. Determine whether the dynamic proportional coefficient deviates from the baseline pure water density threshold, extract the delay feature of the time deviation, and jointly map the dynamic proportional coefficient and delay feature to the preset clinical liquid feature library for fluid property classification and identification to obtain the property classification result.

[0040] S4. Based on the feature entries hit by the trait classification results, obtain the corresponding dynamic density base and wall-hanging compensation coefficient, perform nonlinear compensation conversion of weight-volume dimensions, and calculate the actual volume result of a single discharge after adaptive calibration and the cumulative amount of classification within a specified time.

[0041] S5. Compare the current trait classification results with the historical trait classification results over time. When it is determined that a cross-level mutation has occurred in the trait classification results, trigger a trait mutation warning.

[0042] A weight sensor is installed at the suspension point of the drainage bag, and a liquid level intelligent sensor is installed on the side wall of the drainage container. The electrical signals of both sensors are read synchronously by a microprocessor, deviation features are extracted, a classification algorithm is used to infer the fluid type, and then matching parameters are called to calculate the actual volume and perform time-series monitoring.

[0043] This solution abandons the traditional single-dimensional static conversion logic and constructs a closed-loop analysis system from underlying multimodal signal processing to upper-level clinical sign diagnosis. The underlying principle is based on the cross-fusion technology of fluid mechanics and multi-sensor data: the weight sensor captures the absolute increase in fluid mass within the system, while the intelligent level sensor captures the apparent volume accumulation of fluid at the bottom of a container with a specific cross-sectional area. Due to the significant non-Newtonian viscous damping and pipe wall adhesion effects of high-viscosity or complex-composition drainage fluids as they flow through conduits and drip, the increase in total mass response will inevitably precede the actual rise in liquid level over time; furthermore, drainage fluids with different compositions have completely different physical densities. Therefore, the system extracts the "measurement bias" and "time lag" from the original single-sensor system through time-domain comparison of dual-channel data streams, transforming them into two core feature vectors characterizing fluid density and rheological viscosity. This fusion mechanism, which transforms sources of interference error into sources of data features, is the cornerstone for achieving subsequent accurate classification and adaptive compensation conversion.

[0044] It can adapt to fluid scenarios with different physical properties, significantly improve the accuracy of volume conversion of complex discharges, and reduce the measurement error of the traditional fixed parameter method from 15%-30% to less than 5%. At the same time, it can realize reverse inference of fluid properties and alarm of sudden changes without the need to add additional dedicated sensors, effectively reducing the hardware cost and deployment complexity of the system.

[0045] S1 also includes: timestamp alignment of the time series of the weight sensor and the liquid level smart sensor to extract data within the same time slice; time series deviation is the time difference between the liquid level response and the weight change; dynamic proportional coefficient is the real-time equivalent density.

[0046] A high-precision system-level real-time clock module is introduced into the hardware acquisition center to assign a unified system timestamp to each frame of raw data packet transmitted from the two independent sensors. After receiving the data, the processor uses a time window sliding and interpolation reconstruction algorithm to remove isolated data points with misaligned timestamps and extract data segments within the same physical time interval. Within the extracted data segment, the time corresponding to the start point of height change minus the time corresponding to the start point of weight change is the time difference between the liquid level response and the weight change; the ratio of the mass increment converted to a volume value within this extracted segment to the volume reading of the actual height change is the real-time equivalent density.

[0047] In discrete digital signal processing, the timestamp alignment mechanism acts as an absolute synchronous reference coordinate system for two asynchronous signal buses. By strictly limiting data extraction to the "same time slice," the system eliminates interference phase introduced by bus communication delays and hardware sampling time differences. This ensures that when the microprocessor extracts the "time difference" feature, the difference is purely due to the real dynamic delay caused by the physical viscous friction of the fluid flowing through the pipe, rather than the processing delay of the underlying electronic components, thus ensuring the absolute mathematical rigor of the real-time equivalent density and deviation calculations.

[0048] It completely eliminates the system accumulation error caused by asynchronous sampling of sensor hardware, ensures the absolute accuracy of multi-dimensional sensing response feature extraction, and provides high-fidelity data input for mapping the underlying physical signal to the higher-level feature space.

[0049] In S2, the calculation steps for timing deviation and dynamic scaling factor include:

[0050] The rate of gravity increment is obtained by calculating the first derivative of the cumulative weight increment data.

[0051] The rate of liquid level rise is obtained by calculating the first derivative of the liquid level height increment data.

[0052] The ratio of the rate of increase in gravity to the rate of rise in the liquid level is used as the dynamic proportionality coefficient.

[0053] In the microprocessor's arithmetic unit, a digital differentiation algorithm is executed on the extracted discrete points of the time series. A small sampling time step constant is set, and the cumulative weight at the current moment is subtracted from the cumulative weight of the previous clock cycle, then divided by the time step constant to obtain the gravitational increment rate reflecting the mass flow rate. Similarly, the rate of liquid level rise is obtained by differentiating the height data. Subsequently, the processor executes a division instruction and outputs the ratio of these two rates of change.

[0054] This step transforms the algebraic analysis of static accumulation into a calculus-dynamic analysis of motion characteristics in the mathematical model. At the physics level, the first derivative of the accumulated weight represents the mass inflow rate of the fluid, while the first derivative of the liquid level represents the volumetric accumulation velocity of the fluid in the vertical direction. When the algorithm ratios these two derivatives, it effectively eliminates the time dimension variable, directly yielding the "mass increase corresponding to a unit change in height." Given a fixed cross-sectional area of ​​the drainage container, this dynamic ratio is physically directly proportional to the instantaneous density of the inflowing liquid. Through differential differentiation, the system extremely sensitively amplifies transient components in the signal, enabling even extremely subtle changes in fluid properties to be significantly captured in the derivative domain.

[0055] It enables real-time and transient capture of fluid dynamic characteristics, greatly enhancing the system's sensitivity and response speed to sudden small-flow discharges or hidden fluid property changes, and improving the real-time performance of parameter calculation.

[0056] When determining whether the deviation from the baseline pure water density threshold is in S3, the pure water density needs to be dynamically calibrated in conjunction with the ambient temperature. When the absolute value of the deviation difference between the dynamic proportional coefficient and the baseline pure water density threshold is greater than the preset allowable deviation range, fluid property classification and identification is initiated. When the absolute value of the deviation difference is less than or equal to the preset allowable deviation range, fluid property classification and identification is skipped, and the property classification result is directly determined as pure urine.

[0057] A high-sensitivity temperature sensor is integrated into the control circuit to collect the ambient temperature at the interface of the drainage container in real time. The microprocessor uses a temperature-density thermodynamic reference table stored in its internal non-volatile memory to continuously look up and update the current baseline pure water density threshold. In each data processing loop, the dynamic scaling factor is subtracted from the temperature-calibrated pure water density threshold. Only when the absolute value of this difference exceeds the pre-written fault-tolerant dead zone in the processor will the main control program call and execute the computationally intensive fluid property classification and identification subroutine.

[0058] This step introduces a dynamic reference coordinate system and hysteretic state machine control logic. First, since the calculation of fluid density is highly dependent on the reference benchmark, the temperature compensation algorithm ensures that the benchmark pure water threshold, serving as the anchor point for judgment, is always physically correct under different room temperature environments, eliminating the system static error introduced by thermal expansion and contraction. Second, the system sets a "dead zone threshold" for triggering conditions at the software level. Clinical drainage processes are often accompanied by liquid dripping, splashing, and bubble bursting; these physical phenomena can produce brief numerical spikes. The preset deviation allowance acts as the judgment threshold valve for the low-pass filter, ensuring that the algorithm ignores minor disturbances caused by background white noise. Subsequent classification and matching are only triggered when the detected equivalent density undergoes a real, significant, and sustained physical transition, thereby achieving conditional on-demand scheduling of computing resources.

[0059] It completely eliminates the reference point drift error caused by changes in ambient temperature, ensuring the robustness of the system; it effectively shields the sensor hardware noise and invalid calculation triggers caused by small physical disturbances, significantly reducing the dynamic power consumption of the device and eliminating false alarms.

[0060] The clinical fluid characteristics database includes: pure urine, serous drainage fluid, high-viscosity drainage fluid, bloody drainage fluid, and purulent drainage fluid, along with their corresponding density ranges and wall adhesion compensation coefficients; bloody drainage fluid is graded according to bleeding concentration.

[0061] The device's flash memory blocks are pre-programmed with a multidimensional relational data table based on large-scale in vitro clinical calibration. Each row of this table represents a fluid label, covering pure urine, serous drainage, high-viscosity drainage, purulent drainage, and bloody drainage, which is further subdivided into mild, moderate, and severe types. Each label corresponds to a set of rigorous physical parameter range fields: on the one hand, it includes the upper and lower limits of the characteristic density range reflecting the substance's composition; on the other hand, it includes the characteristic value of the residual wall adhesion compensation coefficient reflecting the fluid dynamic friction characteristics of the tube and container walls.

[0062] This feature library plays a core role in the overall architecture as both a "prior knowledge model" and a "dimensional translation dictionary." Body fluids under various pathological conditions possess highly distinctive rheological characteristic spectra: for example, serous fluid has a density close to pure water and excellent fluidity; purulent fluid is rich in cellular debris and protein networks, exhibiting strong non-Newtonian fluid viscosity despite its moderate density; and bloody fluids of different concentrations show a strict stepwise increase in overall equivalent density with increasing hematocrit. Once the system's underlying layer extracts the physical characteristics of the real-time fluids, it directly enters this data structure for traversal addressing or interval matching, mapping and decoding engineering measurements into physiological and pathological states that medical personnel can directly read, and simultaneously outputting the corresponding error correction coefficients.

[0063] A direct bridge has been successfully built between the underlying physical sensing data and the top-level clinical medical diagnosis, enabling the instrument to not only have physical measurement capabilities, but also to output medical-grade drainage fluid property assessment reports, greatly expanding the clinical application dimensions of the instrument.

[0064] The delay feature characterizes the response hysteresis caused by fluid adhesion to the wall; the mapping and identification process in S3 is as follows: the dynamic scaling factor and the delay feature are used as multidimensional sensing response features, and similarity matching is performed with the features in the clinical fluid feature library to match the corresponding fluid properties.

[0065] The delay feature is obtained by calculating the digital time interval between the peak of the mass increment derivative curve and the peak of the liquid level increment derivative curve. A spatial recognition algorithm is executed within the microprocessor, combining the extracted current dynamic scaling factor and the delay feature into a two-dimensional real-valued coordinate vector. This real-time input vector is then compared to the pre-calibrated center coordinate vectors of various fluids in the feature library, using, for example, the Euclidean distance formula to find the minimum distance solution. The algorithm determines that the shortest distance among the pre-defined center coordinates represents the highest feature similarity, and thus outputs its corresponding fluid characteristic label.

[0066] This step represents a technological leap from one-dimensional threshold judgment to two-dimensional feature space pattern recognition. In rheological principles, the hysteresis time of fluid movement within a pipe is directly proportional to the fluid's kinematic viscosity and the shear stress of the pipe wall. By extracting the "response hysteresis," the system logically quantifies the viscous damping physical process into specific time-dimensional digital features. Since relying solely on equivalent density easily leads to blind spots where the densities of severely infected purulent fluid and mildly bloody fluid overlap, the introduction of delay features allows the system to construct a two-dimensional feature vector space of "equivalent density - viscous delay." This multi-dimensional mapping matching mechanism utilizes the orthogonality of multiple feature dimensions in geometric space, avoiding the recognition confusion caused by one-dimensional data overlap. This ensures that each clinical fluid has its unique and non-overlapping mapping cluster on the two-dimensional plane, thereby significantly improving the classification and analysis capabilities of complex mixed fluids.

[0067] It breaks through the limitations of traditional single physical quantity criteria, which are prone to overlapping and misjudgment. It uses the response hysteresis as the key orthogonal dimension for discrimination, and can accurately quantify the viscosity of fluids without the need for additional expensive microfluidic viscometers, which greatly improves the characteristic detection rate of pus and high-viscosity drainage fluid.

[0068] The formula for nonlinear compensation conversion is:

[0069]

[0070] in, For actual volume results, For cumulative weight increment data, As the dynamic density base, This is the wall-mounted compensation coefficient.

[0071] After completing the fluid property hit, the processor extracts the absolute data of the cumulative weight increment from the current time slice cache. The instruction then addresses the feature library and calls the dynamic density base that perfectly corresponds to the currently determined trait. and the corresponding wall-mounted compensation coefficient The arithmetic logic unit first performs a division operation to obtain the basic correction volume for the current true density. Then, it multiplies this volume by a magnification term that includes compensation coefficients, completing the multiplication calculation process for nonlinear compensation, and finally obtaining the true restored volume. .

[0072] An adaptive mathematical model incorporating a gain factor overcomes the fundamental architectural flaw of traditional instruments that rely on constants for linear calculations. Traditional drainage measurement models implicitly assume that "all fluid passing through the weight sensor will fall to the bottom of the container." However, high-viscosity fluids inevitably form a thick static adhesion layer on the inner surface of slender tubes. While this fluid triggers an increase in the mass of the suspension system, it does not create an observable apparent volume. In the formula, Part of the work is responsible for the initial correction of volume calculation deviations caused by changes in the inherent specific gravity of the fluid; while This is a dynamic gain amplifier based on a viscous dynamics classification model. Its mathematical essence is to use feedback compensation theory to proportionally restore the "hidden loss volume" physically trapped by the pipeline by multiplying it by a positive gain coefficient. This joint algebraic modeling method compensates for the mass distribution dissipation error in complex flow path systems at its source.

[0073] It completely solves the problem of hidden volume loss caused by body fluid adhering to the wall and residue, so that the final output volume data is free from the interference of physical retention in the flow path, and truly achieves accurate volume restoration with high adaptability to various body fluid rheological properties.

[0074] Calculate the cumulative classification amount within a specified time period, including: establishing independent classification and measurement ledgers based on different characteristics; and accumulating the actual volume results into the corresponding classification and measurement ledgers to generate the 24-hour cumulative classification amount.

[0075] Within the system's memory data structure, multiple independent register address spaces or data table columns are partitioned, each mapped to a different trait label such as clean, bloody, or purulent. Each time the processor completes a single volume... After calculation and compensation, based on the simultaneously generated trait classification attribute identifiers, an addressing instruction is initiated to find the specific classification register that perfectly matches the target category, and an accumulation operation is performed to... The data is appended to the existing sum in the register and overwritten. Simultaneously, a hardware timer or system real-time clock runs at the system's underlying level. Every 24-hour time window, a summary interrupt is automatically triggered, packaging the data from each independent category register and generating structured charts and reports.

[0076] A data demultiplexing and hierarchical storage architecture based on state machine identifier attributes is adopted. While calculating the volume increment, the microprocessor uses characteristic tags as address pointers to forcibly deconstruct the mixed time-series volume data stream and distribute it to parallel digital channels for separate integration. This transforms the chaotic "total drainage volume integral" into a "multi-dimensional channel integral vector" strictly classified according to fluid physics properties at the information processing level. By accumulating the separated data, the device can reconstruct a precise, structured data ledger analyzed by type over a long-period time axis, ensuring that data from different sources are not contaminated or overwhelmed.

[0077] It provides automated and detailed classification ledger reports for the clinical medical care system, allowing attending physicians to intuitively grasp the independent total volume percentage of different grades of bloody or purulent fluid within 24 hours, providing high-precision data support for objectively judging the progress of postoperative bleeding control and infection regression.

[0078] The time-series comparison in S5 includes: establishing the time-series evolution trajectory of effluent traits, and calculating the difference in trait grades between the current trait classification result and the historical trait classification results within the preset time window.

[0079] The system microprocessor dynamically maintains a first-in-first-out (FIFO) data queue or a time-sliding window matrix in its running memory. This matrix, with system time as the primary key, continuously records the discrete-level state scalars of fluid properties at each sampling point, constructing a continuous data linked list. The algorithm periodically reads the latest calculated current state level value from the queue, subtracts the historical baseline level value located at the end of the time window, and calculates the first-order difference value of the property level. The magnitude and sign of this difference value are used to determine the slope and abrupt change range of the property trajectory.

[0080] This step extends the static current state identification to a dynamic time series trend analysis logic. By tracking the dynamic evolution path of discrete data through a sliding time window algorithm, the system no longer focuses solely on "which category the drainage fluid currently belongs to," but rather emphasizes calculating and capturing the time derivative of the data distribution. Comparing the difference between the current state scalar and the historical baseline scalar is equivalent to introducing monitoring of the rate of change in the condition along the time dimension. By quantifying the magnitude and speed of the jump in trait levels on the horizontal time axis, the system successfully transforms the instrument's passive recording and measurement mode of discharge into an active tracking and prediction model for the trend of disease deterioration or improvement.

[0081] It endows the system with powerful historical tracing and trend analysis capabilities, enabling continuous, automated, and holographic tracking of the evolution trajectory of excrement characteristics. Without the need for manual review and comparison of complex historical records, it can keenly detect potential changes in disease conditions that are not easily noticed.

[0082] Triggering conditions for trait mutation warnings include: sudden bleeding warnings caused by a sharp increase in serous isodense density due to a serous mutation, or warnings of aggravated infection due to increased viscosity reflected by an increase in delayed characteristics.

[0083] A multi-condition branch structure judgment logic is set in the high-level decision-making layer of the control program. When the comparison operation unit determines that the current trait level difference exceeds the limit, specifically when the equivalent density base value jumps across the preset level threshold in a short period of time, the sound and light alarm chip is driven, and the communication module is called to push the corresponding "sudden bleeding" message protocol packet to the nursing center terminal; when the operation judgment delay feature value shows a continuous abnormal nonlinear lengthening with the time window, the "infection aggravation" early warning message instruction is triggered and pushed.

[0084] This is the final step in connecting the instrument's underlying physical signal analysis technology with the upper-level clinical emergency intervention logic. The operating principle of this step is based on a strongly coupled correlation mapping model between clinical pathophysiological evolution and changes in macroscopic fluid dynamic parameters: when acute active bleeding occurs in a patient's lesion, the hematocrit of red blood cells in the drainage fluid increases instantaneously, which is reflected in the sensor as a step-like abrupt change in equivalent density; conversely, when secondary infection occurs postoperatively, a large amount of leukocytes and fibrin deposits accumulate in the drainage fluid, and the fluid exhibits strong non-Newtonian fluid characteristics, leading to a surge in frictional viscosity, which is reflected in the system's digital characteristics as a dramatic increase in timing delay. The system accurately decodes and translates these underlying abstract numerical transitions and temporal deviations into interruption warning commands that can directly guide clinical emergency procedures through a pre-set Boolean judgment tree gate circuit model.

[0085] It fills the systemic gap in the field of intelligent auxiliary diagnosis and early warning of disease conditions in traditional metrology equipment, and elevates the role of the instrument from a simple data acquisition device to a bedside intelligent monitoring node; it provides warnings of symptom evolution through second-level response capability, which greatly shortens the golden intervention detection cycle for postoperative massive bleeding or serious complications and infections.

[0086] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion, characterized in that, Includes the following steps: S1. Obtain the cumulative weight increment data of the discharged material collected by the weight sensor within the same time slice, and the liquid level height increment data collected by the liquid level intelligent sensor. S2. Process the cumulative weight increment data and the liquid level increment data, calculate the temporal deviation between the gravity increment rate and the liquid level rise rate, and the dynamic proportional coefficient between the gravity increment rate and the liquid level rise rate. S3. Determine whether the dynamic proportional coefficient deviates from the benchmark pure water density threshold, extract the delay feature of the time deviation, and jointly map the dynamic proportional coefficient and the delay feature to a preset clinical liquid feature library for fluid property classification and identification to obtain property classification results. S4. Based on the feature entries hit by the trait classification results, obtain the corresponding dynamic density base and wall-hanging compensation coefficient, perform nonlinear compensation conversion on the weight-volume dimension, and calculate the actual volume result of a single discharge after adaptive calibration and the cumulative amount of classification within a specified time. S5. Compare the current trait classification result with the historical trait classification result over time. When it is determined that the trait classification result has undergone a cross-level mutation, trigger a trait mutation warning.

2. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, S1 further includes: aligning the time series of the weight sensor and the liquid level smart sensor with timestamps to extract data within the same time slice; the time series deviation is the time difference between the liquid level response and the weight change; the dynamic proportional coefficient is the real-time equivalent density.

3. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, In S2, the calculation steps for the timing deviation and dynamic scaling factor include: The rate of gravity increment is obtained by calculating the first derivative of the cumulative weight increment data. The rate of liquid level rise is obtained by calculating the first derivative of the liquid level height increment data. The ratio of the gravity increment rate to the liquid level rise rate is used as the dynamic proportionality coefficient.

4. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, In step S3, when determining whether the deviation from the baseline pure water density threshold is met, the pure water density needs to be dynamically calibrated in conjunction with the ambient temperature. When the absolute value of the deviation difference between the dynamic proportional coefficient and the baseline pure water density threshold is greater than the preset allowable deviation range, the fluid property classification and identification is initiated. When the absolute value of the deviation difference is less than or equal to the preset allowable deviation range, the fluid property classification and identification is skipped, and the property classification result is directly determined as pure urine.

5. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, The clinical fluid characteristic database includes: pure urine, serous drainage fluid, high-viscosity drainage fluid, bloody drainage fluid, and purulent drainage fluid, along with their corresponding density ranges and wall adhesion compensation coefficients; the bloody drainage fluid is graded according to the bleeding concentration.

6. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, The delay feature characterizes the response hysteresis caused by fluid adhesion to the wall; the mapping and identification process in S3 is as follows: the dynamic proportional coefficient and the delay feature are used as multidimensional sensing response features, and similarity matching is performed with the features in the clinical fluid feature library to match the corresponding fluid properties.

7. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, The formula for the nonlinear compensation conversion is: in, For the actual volume result, The cumulative weight increment data, The dynamic density base number is... The wall-mounted compensation coefficient is denoted as .

8. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, The calculation of the cumulative classification amount within a specified time period includes: establishing independent classification and measurement ledgers according to the classification results of different traits; and accumulating the actual volume results into the corresponding classification and measurement ledgers to generate a 24-hour cumulative classification amount.

9. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 1, characterized in that, The time-series comparison in S5 includes: establishing the time-series evolution trajectory of the effluent traits, and calculating the difference in trait grade between the current trait classification result and the historical trait classification result within a preset time window.

10. The method for automatic measurement and analysis of bedside intake and output volume using multi-source sensor fusion according to claim 9, characterized in that, The triggering conditions for the early warning of trait mutation include: a sudden bleeding warning caused by a sharp increase in density from serous mutation to severe bloody isodense, or an early warning of aggravated infection due to increased viscosity reflected by the increase in the delayed feature.

Images