A dynamic and static condition infusion set flow control characteristic test system

Abstract

The present application relates to the technical field of medical instrument testing, in particular to a kind of infusion apparatus flow control characteristic test system under dynamic and static conditions, comprising: signal acquisition module, signal structured module, fingerprint construction module, feature extraction module and compliance evaluation module. By collecting original flow time series signal and environmental parameters, the signal is decomposed and reorganized at multiple levels, forming a structured signal layer containing steady-state, transient and environmental coupling components. Based on this structure, a dynamic flow fingerprint is generated, and an adaptive weight network is used to iteratively calculate a set of key sensitive features. According to the response trajectory of the feature set under simulated abnormal conditions, a graph is generated to determine whether the flow control characteristics are compliant. The present application realizes multi-dimensional fine analysis and intelligent feature extraction of infusion apparatus flow control behavior, and improves the depth and reliability of the test.

Description

Technical Field

[0001] This invention relates to the field of medical device testing technology, and in particular to a testing system for the flow control characteristics of an infusion set under dynamic and static conditions. Background Technology

[0002] Currently, testing the flow control characteristics of infusion sets mainly relies on measuring the average flow rate under constant static pressure or performing simple dynamic simulations. Existing technologies typically directly acquire the raw readings of the flow meter and assess whether the flow rate is within the nominal range through threshold comparison or calculation using a fixed formula. For the effects of dynamic disturbances, such as infusion pump pulsation, patient movement, or changes in ambient temperature, existing methods mostly perform independent testing or treat them as interference and filter them out, lacking systematic coupled analysis.

[0003] These conventional technical solutions have shortcomings. Directly using raw flow signals for analysis results in mixed signal components, making it impossible to effectively isolate and quantify the individual effects of steady-state pressure, transient disturbances, and the coupling effects of environmental parameters on flow control. This leads to crude test results, failing to deeply reveal the intrinsic control mechanisms and performance boundaries of the infusion set under real-world, complex operating conditions. Furthermore, feature extraction heavily relies on fixed parameters preset by human experience, lacking adaptive learning capabilities based on the data itself. This makes it difficult to automatically and accurately capture the key sensitive features that determine the stability of flow control, limiting the intelligence and generalization capabilities of the testing system.

[0004] The current technological challenge is how to achieve refined and structured analysis of flow control behavior under combined dynamic and static conditions, and how to automatically and reliably identify the features that best reflect the essential attributes of the system from complex data, thereby providing a deeper basis for compliance assessment that goes beyond simple threshold judgments. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a test system for the flow control characteristics of infusion sets under dynamic and static conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a test system for the flow control characteristics of an infusion set under dynamic and static conditions, comprising: The signal acquisition module collects the raw flow time-series signal and related environmental parameters of the infusion set under different static pressure and dynamic disturbance modes, forming a raw signal set. The signal structuring module performs multi-level decomposition and recombination on the original signal set to generate a structured signal layer containing steady-state components, transient components, and environmental coupling components. The fingerprint construction module, based on the structured signal layer, constructs a dynamic flow fingerprint that reflects the inherent evolution law of flow control; The feature extraction module uses the dynamic traffic fingerprint to drive an adaptive weight network and iteratively calculates the key sensitive feature set of the traffic control system. The compliance assessment module generates a compliance judgment map of flow control characteristics based on the response trajectory of the key sensitive feature set under simulated abnormal operating conditions.

[0007] As a further aspect of the present invention, the original flow time-series signal and associated environmental parameters of the infusion set under different static pressure and dynamic disturbance modes are collected to form an original signal set, specifically including: Under multiple preset static pressure reference points, the stable flow output signal of the infusion set is continuously collected to obtain the static pressure flow signal; Under multiple preset frequencies and amplitudes of mechanical vibration disturbance, the real-time flow fluctuation signal of the infusion set is collected synchronously to obtain the dynamic disturbance flow signal; Throughout the test, temperature and humidity data of the environment where the infusion set is located, as well as pressure data of key points in the infusion tubing, were continuously collected to obtain a sequence of related environmental parameters. The static pressure flow signal, the dynamic disturbance flow signal, and the associated environmental parameter sequence are aligned along the time axis and packaged to form the original signal set.

[0008] As a further aspect of the present invention, the original signal set is decomposed and recombined at multiple levels to generate a structured signal layer containing steady-state components, transient components, and environmental coupling components, specifically including: Empirical mode decomposition is performed on each traffic time-series signal in the original signal set to separate multiple intrinsic mode functions arranged from high frequency to low frequency; From the multiple intrinsic mode functions, modes with an energy percentage exceeding a preset threshold and a frequency distribution in an extremely low range are selected and aggregated and reconstructed into the steady-state components. From the plurality of intrinsic mode functions, modes that are strongly correlated with the frequency of the applied mechanical vibration disturbance are selected and aggregated and reconstructed into the transient components; Calculate the mutual information value between the associated environmental parameter sequence and the remaining modes in the plurality of intrinsic mode functions, and fuse the modes with mutual information values ​​greater than a set threshold with the corresponding environmental parameters to generate the environmental coupling component; The steady-state component, the transient component, and the environmental coupling component are collectively organized into the structured signal layer.

[0009] As a further aspect of the present invention, constructing a dynamic flow fingerprint that reflects the inherent evolution law of flow control based on the structured signal layer specifically includes: For the steady-state components in the structured signal layer, calculate their sample entropy at multiple different time scales to form a steady-state complexity vector; For the transient components in the structured signal layer, extract their peak sequence and decay rate sequence in different perturbation periods to form a transient response vector; For the environmental coupling component in the structured signal layer, the rate of change of its fluctuation amplitude and phase lag angle under different environmental parameter gradients are analyzed to form an environmental sensitivity vector; The steady-state complexity vector, the transient response vector, and the environmental sensitivity vector are concatenated in time sequence and mapped to a dense vector of fixed dimension through an encoder network. The dense vector is the dynamic traffic fingerprint.

[0010] As a further aspect of the present invention, the dynamic flow fingerprint is used to drive an adaptive weight network to iteratively calculate the key sensitive feature set of the flow control system, specifically including: Initialize a fully connected neural network as the adaptive weight network, with its input layer dimension matching the dimension of the dynamic traffic fingerprint; The dynamic traffic fingerprint is input into the adaptive weight network, and the hidden layer of the adaptive weight network outputs a set of initial feature weights. The initial feature weights are used to weight and fuse all signal components in the structured signal layer to generate an initial fused signal. The initial fusion signal is fed back to an auxiliary input channel of the adaptive weight network, concatenated with the dynamic traffic fingerprint, and then passed through the adaptive weight network again to update the feature weights. Repeat the weighted fusion and weight update steps until the change in the feature weights is less than the convergence threshold. At this point, the signal components with the highest weight values ​​are identified as the key sensitive feature set.

[0011] As a further aspect of the present invention, the key sensitive feature set includes ultra-low frequency fluctuation energy extracted from the steady-state component, resonance gain under specific frequency disturbance extracted from the transient component, and temperature-flow transfer delay time extracted from the environmental coupling component.

[0012] As a further aspect of the present invention, generating a compliance determination map of flow control characteristics based on the response trajectory of the key sensitive feature set under simulated abnormal operating conditions specifically includes: Construct a set of simulated abnormal operating conditions, which includes simulations of partial blockage in pipelines, relaxation of roller clamps, and fatigue of infusion set materials. Under each simulated abnormal operating condition, the original signal of the infusion set is reacquired, and through the processing flow of the structured signal layer and the dynamic flow fingerprint, the feature value sequence corresponding to the key sensitive feature set under the simulated abnormal operating condition is obtained. The feature value sequence of the key sensitive feature set under normal working conditions is used as the reference trajectory, and the feature value sequence obtained under abnormal working conditions is used as the test trajectory. The deviation vector of each test trajectory relative to its corresponding reference trajectory is calculated. Using different anomaly types in the simulated abnormal working condition set as one dimension and different features in the key sensitive feature set as another dimension, the calculated deviation vector values ​​are filled into the corresponding positions to form a two-dimensional matrix, which is the compliance judgment map.

[0013] As a further aspect of the present invention, the deviation vector is calculated using a dynamic time warping algorithm, which is used to measure the minimum cumulative distance in shape between two feature value sequences.

[0014] As a further aspect of the present invention, the compliance determination map is further used for online monitoring, specifically in the following manner: The system acquires signals from the current infusion process in real time and extracts feature values ​​from the current key sensitive feature set to form a real-time feature trajectory. Calculate the real-time deviation vector between the real-time feature trajectory and the reference trajectory; The real-time deviation vector is matched with the pre-defined threshold regions of each abnormal pattern in the compliance judgment map; When the real-time deviation vector falls into a certain abnormal mode threshold area, a warning signal corresponding to the abnormal mode is triggered.

[0015] As a further aspect of the present invention, the warning signal includes visual warnings, acoustic warnings, and a text report containing the type of anomaly and the degree of deviation.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A multi-level decomposition and recombination operation is performed on the original flow time-series signal to generate a structured signal layer containing steady-state components, transient components, and environmental coupling components. This technique separates the original mixed signal according to its physical generation mechanism, allowing the steady-state component representing the basic flow under constant pressure, the transient component reflecting fluctuations caused by instantaneous disturbances, and the environmental coupling component reflecting the slow effects of environmental parameters to be quantified independently. This process completes the transformation from single-dimensional amplitude information to multi-dimensional physical cause information, enabling the analysis of flow control behavior to be performed based on signal components of different properties. It achieves decoupling and accurate measurement of various influencing factors under complex operating conditions, providing a high-resolution signal foundation for a deeper understanding of the mechanism of flow control.

[0017] A dynamic traffic fingerprint is constructed to drive an adaptive weighted network, which extracts a set of key sensitive features through iterative computation. This network dynamically adjusts its internal connection weights and parameters based on the characteristics of the input structured signal, autonomously optimizing the feature selection path and evaluation criteria during iteration. This process frees feature extraction from fixed rules or prior knowledge, guiding it instead through the data itself and its inherent patterns. The resulting feature set more fundamentally characterizes the system's behavioral patterns and state transition properties under different dynamic and static conditions, exhibiting higher sensitivity and specificity to subtle changes in system performance.

[0018] Based on a structured signal layer and adaptively extracted key sensitive feature sets, a compliance judgment map of flow control characteristics is generated under simulated abnormal operating conditions. This map is not based on a single threshold judgment, but rather a comprehensive evaluation model integrating multi-component evolution trajectories and high-dimensional feature response patterns. This upgrades compliance assessment from a traditional binary "pass / fail" judgment to a deep diagnostic that reflects performance degradation trends, identifies vulnerable points, and recognizes abnormal patterns, resulting in outputs containing richer dimensional and hierarchical information. Attached Figure Description

[0019] Figure 1 This is a timing diagram of the infusion set flow control characteristic testing system under dynamic and static conditions described in this invention; Figure 2 A flowchart for constructing dynamic traffic fingerprints; Figure 3 A three-dimensional feature vector distribution map of dynamic traffic fingerprints; Figure 4 Radar charts showing the flow control characteristics of infusion sets under multiple operating conditions; Figure 5 This is a graph showing the evolution of feature weights during the iterative process of the adaptive weighted network. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] See Figure 1 The signal acquisition module synchronously acquires the raw flow time-series signal of the infusion set and its associated environmental parameters under preset static pressure benchmark and dynamic disturbance modes, thus forming an unprocessed raw signal set. The signal structuring module performs multi-level decomposition and recombination operations on this raw signal set, separating and reconstructing steady-state components, transient components, and environmental coupling components representing different physical meanings through algorithms. These components together constitute a hierarchical structured signal layer. Based on this, the fingerprint construction module extracts multi-dimensional quantitative indicators from the structured signal layer and fuses and encodes them to generate a dynamic flow fingerprint that can condense and reflect the inherent evolution law of flow control. This dynamic flow fingerprint then drives the adaptive weight network in the feature extraction module to perform iterative calculations. Through repeated weighted fusion and weight adjustment, the adaptive weight network filters and calculates the key sensitive feature set that best represents the system behavior from the structured signal layer. The compliance assessment module simulates a series of abnormal operating conditions, observes and calculates the difference between the response trajectory of the key sensitive feature set under these abnormalities and the normal benchmark, and generates a compliance judgment map for comprehensive judgment based on these differences, thereby completing a comprehensive assessment of the infusion set's flow control characteristics.

[0023] In one embodiment of the present invention, the signal acquisition module continuously acquires the stable flow output signal of the infusion set under multiple preset static pressure reference points to obtain static pressure flow signals. Simultaneously, the module synchronously acquires the real-time flow fluctuation signal of the infusion set under multiple preset frequency and amplitude mechanical vibration disturbances to obtain dynamic disturbance flow signals. Throughout the test, ambient temperature data, humidity data, and pressure data at key points of the infusion tubing are continuously acquired to obtain a sequence of associated environmental parameters. The aforementioned static pressure flow signals, dynamic disturbance flow signals, and associated environmental parameter sequences are aligned and packaged along a unified time axis to form the original signal set required for subsequent processing. The signal structuring module performs empirical mode decomposition on each flow time-series signal in the original signal set, separating each signal into multiple intrinsic mode functions (IMFs) arranged from high frequency to low frequency. From these IMFs, modes with an energy percentage exceeding a preset threshold and a frequency distribution in an extremely low range are selected, and these modes are aggregated and reconstructed into steady-state components characterizing the system's basic output capability. From multiple intrinsic mode functions (EMFs), modes strongly correlated with the applied mechanical vibration disturbance frequency are selected and aggregated and reconstructed into transient components characterizing the forced response of the system. The mutual information values ​​between the associated environmental parameter sequence and the remaining EEMs are calculated. Modes with mutual information values ​​exceeding a set threshold are fused with their corresponding environmental parameters to generate environmental coupling components characterizing the coupling relationship between environmental factors and flow fluctuations. The steady-state components, transient components, and environmental coupling components are collectively organized into a structured signal layer for use by subsequent modules.

[0024] In practical implementation, the signal acquisition module of the infusion set flow control characteristic testing system operates on a controlled test platform. This platform can precisely adjust the static pressure at the infusion set inlet and apply programmable mechanical vibrations. For example, the platform's preset static pressure reference points are 50 mmHg, 100 mmHg, and 150 mmHg. At each static pressure reference point, the signal acquisition module continuously acquires the stable flow output signal from the infusion set's output end using a high-precision flow sensor at a sampling frequency of 100 times per second, with a sampling duration set to 60 seconds, thereby obtaining static pressure flow signals corresponding to different static pressures. In practical implementation, when dynamic disturbances are applied, the vibration table of the test platform operates according to a preset program. For example, two sinusoidal mechanical vibration disturbances are applied: one with a frequency of 5 Hz and an amplitude of 0.5 mm, and the other with a frequency of 10 Hz and an amplitude of 0.3 mm. Under each vibration mode, the signal acquisition module synchronously acquires the real-time flow fluctuation signal of the same infusion set, also at a sampling frequency of 100 times per second, obtaining the dynamic disturbance flow signal corresponding to the disturbance mode. In practice, throughout the static and dynamic testing process described above, the temperature sensor, humidity sensor, and three pressure sensors installed at the infusion line inlet, middle section, and drip chamber in the test environment work continuously. The temperature sensor and humidity sensor record data once per second, while the pressure sensor and flow sensor collect data synchronously at a frequency of 100 times per second, thereby obtaining a sequence of associated environmental parameters that are strictly time-synchronized with the flow signal.

[0025] In some embodiments, after the signal acquisition module completes data acquisition, it transmits the static pressure-flow signal, dynamic disturbance flow signal, and associated environmental parameter sequence to the data processing unit. The data processing unit, based on the high-precision timestamps embedded in each sensor data packet, aligns all data to the same time axis with millisecond-level precision. The aligned data is packaged into a structured data file containing time dimensions, multiple signal dimensions, and environmental parameter dimensions. This structured data file constitutes the original signal set. In a specific implementation, the signal structuring module calls the Empirical Mode Decomposition (EMD) algorithm to process each flow time-series signal in the original signal set. Taking a flow signal acquired under a static pressure of 100 mmHg as an example, the EMD algorithm decomposes it into eight intrinsic mode functions (EMFs) arranged from high to low frequency and a residual component. The signal structuring module calculates the energy percentage of each EMF and analyzes its dominant frequency, selecting EMFs with an energy percentage exceeding 60% of the total energy and a dominant frequency below 0.1 Hz. These selected EMFs are then superimposed and reconstructed; the reconstructed signal is defined as the steady-state component. The signal structuring module then analyzes the correlation between all intrinsic mode functions and the applied mechanical vibration disturbance frequency, calculates the amplitude of the power spectral density of each intrinsic mode function at the disturbance frequency, filters out intrinsic mode functions whose amplitude exceeds a set threshold, and aggregates and reconstructs these intrinsic mode functions into transient components.

[0026] In some embodiments, for the intrinsic mode functions (EMFs) remaining after empirical mode decomposition, the signal structuring module calculates their mutual information value with each parameter in the associated environmental parameter sequence. The mutual information value quantifies the nonlinear dependency between the two variables. EMFs with a mutual information value greater than 0.3 with any environmental parameter are selected. For example, an EMF with a mutual information value of 0.45 with the environmental temperature sequence is selected. The signal structuring module fuses the selected EMFs with their corresponding environmental parameter sequences. One method of fusion is to multiply the environmental parameter sequence as a modulation signal with the EMF to generate an environmental coupling component. Finally, the signal structuring module organizes the reconstructed steady-state component, transient component, and generated environmental coupling component according to the same time base, forming a structured signal layer containing three clearly defined physical signal layers. In a specific implementation, the calculation of the mutual information value used to select environmental coupling components involves a formula used to select from multiple candidate modes. The formula is specifically expressed as the selection criteria: Where: symbol Represents the remaining 1st mode after empirical mode decomposition. One intrinsic mode function, symbol Represents the first in the sequence of associated environmental parameters An environmental parameter (such as temperature), symbol Represents the preset mutual information threshold value, symbol This represents a function operation that calculates the mutual information between two variables.

[0027] In one embodiment of the present invention, see [reference] Figure 2 The fingerprint construction module constructs a dynamic flow fingerprint based on the structured signal layer. For the steady-state components in the structured signal layer, its sample entropy at multiple different time scales is calculated, thus forming a steady-state complexity vector describing the output stability and intrinsic complexity. For the transient components in the structured signal layer, its peak sequence and decay rate sequence within different disturbance periods are extracted to form a transient response vector describing the system's dynamic response capability. For the environmental coupling components in the structured signal layer, its fluctuation amplitude change rate and phase lag angle under different environmental parameter gradients are analyzed to form an environmental sensitivity vector describing the system's environmental sensitivity. The above steady-state complexity vector, transient response vector, and environmental sensitivity vector are concatenated according to their temporal relationship to form a high-dimensional joint vector. This joint vector is then input into a pre-trained encoder network, which maps it into a fixed-dimensional dense vector. This dense vector is the dynamic flow fingerprint that comprehensively reflects the intrinsic evolution law of flow control.

[0028] In specific implementation, the fingerprint construction module receives a structured signal layer from the signal structuring module. The structured signal layer contains steady-state components, transient components, and environmental coupling components. The fingerprint construction module calculates the sample entropy of the steady-state components at multiple different time scales. The selection of time scales is based on the analysis requirements for the long-term stability of the flow signal. For example, the scale factor τ is selected as four scales: 1, 2, 3, and 4. The steady-state component time series is coarsened, and the sample entropy value of each coarsened sequence is calculated. The calculation of sample entropy involves setting the embedding dimension m and the similarity tolerance r. The four calculated sample entropy values ​​are arranged in ascending order of scale to form a four-dimensional steady-state complexity vector. In practical implementation, for transient components, the fingerprint construction module first identifies the complete cycle corresponding to each mechanical vibration disturbance in the component. For example, it identifies five consecutive complete disturbance cycles, extracts the positive and negative peak values ​​of the flow fluctuation in each cycle, and forms a peak sequence consisting of ten peak data. At the same time, it calculates the attenuation rate of the fluctuation in each disturbance cycle. The attenuation rate is calculated using the logarithmic attenuation rate formula, which is the natural logarithm of the ratio of two adjacent peaks in the same direction, thus obtaining an attenuation rate sequence containing five attenuation rate values. The peak sequence and the attenuation rate sequence are combined into a fifteen-dimensional transient response vector.

[0029] In some embodiments, for the environmental coupling component, the fingerprint construction module analyzes its response characteristics under environmental parameter gradients. For example, it analyzes the fluctuation of the environmental coupling component during a test phase where the temperature parameter increases linearly. It calculates the average change in the fluctuation amplitude of the environmental coupling component for every 1 degree Celsius temperature change, divides this change by the initial amplitude to obtain the fluctuation amplitude change rate, and simultaneously calculates the phase lag angle between the environmental coupling component fluctuation and the environmental temperature change sequence. The phase lag angle is obtained by calculating the cross-correlation function of the two sequences and finding the time delay corresponding to the maximum cross-correlation value. The time delay is then converted into an angle. When multiple environmental parameters exist, such as temperature and inlet pressure, the fluctuation amplitude change rate and phase lag angle of the environmental coupling component with respect to each parameter are calculated separately. All calculation results are arranged in order of parameter category to form an environmental sensitivity vector. It can be understood that after the fingerprint construction module completes the construction of the above three vectors, it concatenates them in the order of steady-state complexity vector, transient response vector, and environmental sensitivity vector to form a high-dimensional joint feature vector. The dimension of this joint feature vector is the sum of the dimensions of each sub-vector.

[0030] In practice, the concatenated joint feature vector is input into a pre-trained encoder network. This encoder network is a feedforward neural network with three fully connected layers. The first layer maps the high-dimensional input to 128 dimensions, the second to 64 dimensions, and the third outputs a 32-dimensional dense vector. The encoder network is trained using joint feature vectors generated from a large amount of historical test data as input, and is trained in an unsupervised manner using an autoencoder or a contrastive learning approach. The final output, a 32-dimensional dense vector, is the dynamic traffic fingerprint. The calculation of sample entropy follows a standard formula used to quantify the complexity of the time series data. The formula is specifically expressed as: Where: symbol Represents the embedding dimension, symbol The similarity tolerance is typically taken as a proportion of the standard deviation of the time series, with the sign... The symbol represents the length of data points in a time series. Indicates that in dimension The number of template vector pairs that satisfy the similarity condition, sign Indicates that in dimension The number of template vector pairs that satisfy the similarity condition, sign This represents the natural logarithm operation. Optionally, the encoder network can be trained using a variational autoencoder structure, constraining the distribution of the latent space (i.e., dense vectors) by introducing a KL divergence term into the loss function. It can be understood that the fixed-dimensional design of the dynamic flow fingerprint ensures the consistency of the input to the subsequent adaptive weight network.

[0031] In one embodiment of the present invention, the feature extraction module uses a dynamic traffic fingerprint to drive an adaptive weight network for iterative computation. First, a fully connected neural network is initialized as the adaptive weight network, with its input layer dimension matching that of the dynamic traffic fingerprint. The dynamic traffic fingerprint is input into this adaptive weight network, and the hidden layers output a set of initial feature weights corresponding to each signal component in the structured signal layer. These initial feature weights are used to weight and fuse all signal components in the structured signal layer, generating an initial fused signal. This initial fused signal is fed back to an auxiliary input channel of the adaptive weight network, where it is concatenated with the original dynamic traffic fingerprint. The concatenated vector is then passed through the adaptive weight network again to update a new set of feature weights. This process of generating a new signal through weighted fusion and feeding it back to update the weights is repeated, forming a closed iterative loop. The iteration process terminates when the change in feature weights during the iteration is less than a preset convergence threshold. At this point, the signal components with the highest final feature weight values ​​are identified as the key sensitive feature set of the traffic control system. Key sensitive feature sets typically include ultra-low frequency fluctuation energy extracted from steady-state components, resonant gain under specific frequency disturbances extracted from transient components, and temperature-flow transfer delay time extracted from environmental coupling components.

[0032] In the specific implementation, the feature extraction module receives a dynamic traffic fingerprint from the fingerprint construction module. The dynamic traffic fingerprint is a 32-dimensional dense vector. The feature extraction module initializes a fully connected neural network as an adaptive weight network. The number of nodes in the input layer of the adaptive weight network is set to 32 to match the dimension of the dynamic traffic fingerprint. The adaptive weight network contains a hidden layer with 64 nodes and an output layer. The number of nodes in the output layer is equal to the total number of signal components in the structured signal layer, such as steady-state components, transient components, environmental coupling components, and their possible sub-components. The number of nodes in the output layer is set to 10. In the specific implementation, the dynamic traffic fingerprint is input into the adaptive weight network. The hidden layer of the adaptive weight network processes the input 32-dimensional vector through an activation function. The output layer maps the output of the hidden layer to a set of 10 initial feature weights. These 10 initial feature weights correspond one-to-one with the 10 signal components of the structured signal layer, and the values ​​of the initial feature weights range from 0 to 1. The initial feature weights are used to perform weighted fusion on all 10 signal components in the structured signal layer. The weighted fusion operation is to multiply the data sequence of each signal component by the corresponding initial feature weight, and then add all the weighted sequences together to generate an initial fused signal. The initial fused signal is a time series of the same length as the original signal.

[0033] In practice, the initial fused signal is fed back to an auxiliary input channel of the adaptive weight network. The dimension of the auxiliary input channel is adapted to the length of the initial fused signal. At the input of the adaptive weight network, the dynamic traffic fingerprint and the flattened vector of the initial fused signal are concatenated to form an extended input vector. This extended input vector is then passed through the adaptive weight network again, which updates its calculations based on the extended input vector, generating a set of updated feature weights at the output layer. This weighted fusion and weight update process is repeated: the newly obtained feature weights are used to weight the signal components of the structured signal layer to generate a new fused signal. This new fused signal is then concatenated with the original dynamic traffic fingerprint and input into the adaptive weight network to obtain the next round of feature weights. This process forms a closed iterative loop. The iteration continues until the change between the feature weight vectors calculated in two adjacent iterations is less than a preset convergence threshold (set to 0.001). At this point, the iteration terminates, and the adaptive weight network outputs the final feature weight vector. The feature extraction module identifies the three signal components with the highest weight values ​​in the final feature weight vector as the key sensitive feature set. It is understandable that the key sensitive feature set includes ultra-low frequency fluctuation energy extracted from steady-state components, resonance gain under specific frequency perturbations extracted from transient components, and temperature-flow transfer delay time extracted from environmental coupling components.

[0034] In some embodiments, the iterative process of weighted fusion and weight update involves a convergence judgment formula, which is used to determine whether the iteration stops. The formula is specifically expressed as follows: Where: symbol Indicates the first The feature weight vector obtained after the iteration is denoted as... Indicates the first The feature weight vector obtained after the iteration is denoted as... The operation that calculates the Euclidean norm (L2 norm) of a vector is represented by the symbol. This represents the preset convergence threshold. Optionally, the hidden layers of the adaptive weight network can use the ReLU activation function, and the output layer can use the Softmax activation function to ensure that the sum of the feature weights is 1. In some embodiments, the auxiliary input channel processes the initial fused signal by downsampling to reduce the dimensionality of the expanded input vector and avoid excessively large inputs to the adaptive weight network. It can be understood that the final identified key sensitive feature set is the signal component considered most representative of the state of the flow control system after multiple iterative evaluations by the adaptive weight network.

[0035] See Figure 3This is a 3D feature vector distribution chart of dynamic flow fingerprint. This grouped bar chart shows the normalized feature values ​​of the three core dimensions of dynamic flow fingerprint—steady-state complexity, transient response, and environmental sensitivity—at five time scales in the infusion set flow control characteristic test, reflecting the multi-dimensional dynamic characteristics of the flow control system. The significant differences in the feature value distributions of the three vectors indicate that dynamic flow fingerprint can comprehensively characterize the flow control characteristics of the infusion set from three dimensions: steady-state accuracy, dynamic response, and environmental robustness, avoiding the limitations of single features. The dominant position of the transient response vector suggests that the stability of the infusion set under dynamic disturbances needs to be focused on, which is also a core risk point for infusion safety in clinical scenarios. The peak value of steady-state complexity at a medium time scale, and the time scale dependence of environmental sensitivity, provide a quantitative basis for optimizing test duration and environmental parameter sampling frequency.

[0036] In one embodiment of the present invention, the compliance assessment module generates a compliance judgment map based on the response trajectory of a key sensitive feature set under simulated abnormal operating conditions. First, a set of simulated abnormal operating conditions is constructed, including various preset abnormality types such as partial pipeline blockage simulation, roller clamp loosening simulation, and infusion set material fatigue simulation. Under each simulated abnormal operating condition, the signal acquisition process is restarted to obtain the raw signal from the infusion set, and through the processing flow of the signal structuring module and the fingerprint construction module, the feature value sequence corresponding to the key sensitive feature set under this abnormal operating condition is obtained. The feature value sequence obtained under normal operating conditions is set as the baseline trajectory, and the feature value sequences obtained under each abnormal operating condition are used as test trajectories. The deviation vector of each test trajectory relative to its corresponding baseline trajectory is calculated. The deviation vector is calculated using a dynamic time warping algorithm, which measures the minimum cumulative distance in shape between two feature value sequences. Using different anomaly types in the simulated abnormal operating condition set as one dimension and different features in the key sensitive feature set as another dimension, the calculated values ​​of each deviation vector are filled into their corresponding positions, thus forming a two-dimensional matrix, which is the compliance judgment map of flow control characteristics.

[0037] In practice, the compliance assessment module first generates a set of simulated abnormal operating conditions for testing. The set of simulated abnormal operating conditions includes three specific types: partial blockage simulation, which is achieved by connecting a section of thin tubing with an inner diameter of 0.5 mm in series to the middle section of a normal infusion tubing; roller clamp relaxation simulation, which is achieved by rotating the adjustment knob of the infusion set roller clamp in the opposite direction by a specific angle to reduce its clamping force on the tubing; and infusion set material fatigue simulation, which is achieved by placing a section of tubing sample in a constant temperature and humidity chamber and applying periodic tensile loads for pre-aging treatment, and then installing the aged tubing sample onto the testing system. In practical implementation, for each abnormal condition in the set of simulated abnormal operating conditions, the system re-runs the complete test process. The signal acquisition module acquires the original signal of the infusion set under the same static pressure reference point and dynamic disturbance mode. The signal structuring module and fingerprint construction module process these newly acquired signals in sequence, and finally obtain the feature value sequence corresponding to the key sensitive feature set under the current abnormal operating condition. The key sensitive feature set includes ultra-low frequency fluctuation energy, resonance gain under specific frequency disturbance, and temperature-flow transfer delay time. Taking ultra-low frequency fluctuation energy as an example, under normal operating conditions, a feature value sequence fluctuating in the range of 0.95 to 1.05 units may be obtained, while under the simulated operating condition of partial blockage in the pipeline, a feature value sequence fluctuating in the range of 0.60 to 0.75 units may be obtained.

[0038] In implementation, the compliance assessment module pre-stores the feature value sequences of key sensitive feature sets under normal operating conditions and uses these normal sequences as baseline trajectories. Feature value sequences obtained under simulated pipeline partial blockage, roller clamp relaxation, and infusion set material fatigue conditions are used as test trajectories. The compliance assessment module calculates the deviation vector of each test trajectory relative to its corresponding baseline trajectory. The deviation vector is calculated using a dynamic time warping algorithm, which minimizes the cumulative distance by finding the optimal bending path between two sequences; this distance is the deviation. For each key sensitive feature, its deviation is calculated under three abnormal operating conditions, ultimately generating three deviation values. These deviation values ​​can be understood as being filled into a two-dimensional matrix. Different anomaly types in the simulated abnormal operating condition set constitute one dimension of the matrix, and different features in the key sensitive feature set constitute the other dimension. The resulting two-dimensional matrix is ​​the compliance judgment map of flow control characteristics. The compliance judgment map provides a quantitative and visual tool for comparing the impact of different anomaly types on different feature parameters.

[0039] In some embodiments, the formula for calculating the minimum cumulative distance between two sequences using the dynamic time warping algorithm is specifically expressed as follows: Where: symbol Indicates the calculated deviation, symbol This represents an alignment path that specifies the sequence. and sequence The correspondence between data points, symbols This represents the first reference trajectory sequence under normal operating conditions. The feature values ​​of each data point, with symbols This represents the test trajectory sequence under abnormal operating conditions. The feature values ​​of each data point, with symbols This represents all possible alignment paths. Find the path that minimizes the cumulative distance. Optionally, when constructing the compliance judgment map, a normalization step can be introduced, dividing the calculated original deviation by the standard deviation of its corresponding normal baseline trajectory sequence to obtain the relative deviation and enhance comparability. Refer to Table 1, which shows sample data of a compliance judgment map. Table 1 shows the deviation vector values ​​calculated for three key sensitive features under three simulated abnormal operating conditions.

[0040] Table 1: Exemplary Compliance Judgment Graph Data Table In some embodiments, the values ​​in the compliance judgment graph can be further converted into qualitative judgments through threshold settings. For example, an alarm threshold can be set for each feature-anomaly type combination. When the deviation exceeds the threshold, it is marked as "abnormal" at the corresponding position in the matrix; otherwise, it is marked as "normal". It can be understood that the process of generating the compliance judgment graph relies on the systematic testing and signal processing of simulated abnormal operating conditions in the early stage, thereby establishing a quantitative mapping relationship between abnormal patterns and feature responses.

[0041] See Figure 4 This is a radar chart showing the multi-condition flow control characteristics of an infusion set. The chart compares the performance of normal operating conditions (blue) with three abnormal operating conditions (tubing blockage, roller loosening, and material fatigue) across six key feature dimensions, intuitively reflecting the impact of different abnormal modes on the infusion set's flow control characteristics. The radar profiles of different abnormal operating conditions show significant differences, serving as a "fingerprint" for automatic anomaly type determination in online monitoring, thus improving the intelligence level of infusion safety. The radar chart quantifies the degree of impact of various anomalies on key features, providing clear quantitative basis for infusion set durability testing and fault simulation. Specific patterns of feature deviation can guide infusion set design optimization; for example, to address the resonance gain problem of roller loosening, the stiffness of the roller clamping structure can be optimized.

[0042] In one embodiment of the present invention, the compliance judgment map is further used for online monitoring of the infusion process. During online monitoring, the system acquires signals of the current infusion process in real time and extracts feature values ​​of the current key sensitive feature set through the same processing flow to form a real-time feature trajectory. The real-time deviation vector of this real-time feature trajectory and the pre-stored normal operating condition baseline trajectory are calculated. The calculated real-time deviation vector is matched and compared with the pre-marked threshold regions of each abnormal mode in the compliance judgment map. When the real-time deviation vector falls into a certain abnormal mode threshold region, the system triggers a warning signal corresponding to that abnormal mode. The warning signal can be presented in the form of visual warnings on the control panel, acoustic warnings issued by the system, and text reports containing the identified abnormal type and specific degree of deviation.

[0043] In practice, the online monitoring function of the compliance assessment map is activated on a monitoring terminal deployed in a clinical environment. This terminal is continuously connected to high-precision flow sensors, temperature sensors, and pressure sensors in the infusion tubing currently in use. The system acquires real-time flow signals, ambient temperature signals, and tubing pressure signals during the current infusion process. These real-time signals are input into the system as data streams. The signal structuring module and fingerprint construction module process the real-time data streams using the same algorithms and parameter configurations as in the offline testing phase. The feature extraction module, based on the dynamically generated flow fingerprint, drives an adaptive weighted network to quickly calculate the feature values ​​of the key sensitive feature set in the current state. The key sensitive feature set includes ultra-low frequency fluctuation energy, resonance gain under specific frequency disturbances, and temperature-flow transfer delay time. The system adds the three most recently calculated feature values ​​to the end of their respective feature sequences every minute, forming a real-time feature trajectory with a length of the most recent 10 minutes. The data structure of the real-time feature trajectory is consistent with the baseline trajectory used in the compliance assessment map construction phase. In practice, the system calculates the real-time deviation vector between the real-time feature trajectory and the normal operating condition baseline trajectory pre-stored in the database. The calculation of the real-time deviation vector also adopts the dynamic time warping algorithm. For each feature in the key sensitive feature set, the dynamic time warping algorithm calculates the minimum cumulative distance between its 10-minute real-time feature trajectory and the corresponding 10-minute baseline trajectory, thereby obtaining a real-time deviation vector containing three deviation values.

[0044] In some embodiments, the calculated real-time deviation vector is fed into a pattern matching unit. The pattern matching unit accesses pre-defined threshold regions for each abnormal mode in the compliance judgment map. These threshold regions are multi-dimensional spatial ranges statistically defined based on a large amount of offline simulation test data. For example, the threshold region corresponding to the "pipeline partial blockage simulation" abnormal mode can be defined as a three-dimensional spatial region where the ultra-low frequency fluctuation energy deviation is greater than 10.0, the resonance gain deviation under a specific frequency disturbance is between 2.0 and 4.5, and the temperature-flow transfer delay time deviation is less than 2.5. The pattern matching unit compares the real-time deviation vector with each abnormal mode threshold region to determine whether the real-time deviation vector falls within the geometric space defined by a certain threshold region. It can be understood that when the real-time deviation vector falls into the threshold region of, for example, the "roller clamp relaxation simulation" abnormal mode, the pattern matching unit generates a judgment signal corresponding to the "roller clamp relaxation simulation" abnormal mode. The triggering of the warning signal follows a spatial inclusion judgment logic, which can be formally expressed by a distance threshold formula to determine whether the real-time vector is close to a typical region of a certain abnormal pattern. The formula is specifically expressed as follows: Where: symbol Represents the calculated real-time deviation vector, with the symbol... This represents the center vector of a pre-defined threshold region for a specific anomaly pattern in the compliance assessment graph, with the symbol... The symbol represents the operation of calculating the Euclidean distance between two vectors. This represents the radius threshold set for this abnormal pattern, used to define the boundary range of the threshold region. Optionally, the pattern matching unit can employ a multi-level threshold strategy, such as setting threshold regions at three levels: "Attention," "Warning," and "Severe," with different radius thresholds corresponding to different levels. This enables tiered management of early warning systems.

[0045] In some embodiments, once a signal is detected, the system immediately triggers an early warning signal, which is issued by the alarm subsystem of the monitoring terminal. Visual warnings are manifested by the icon color of the corresponding infusion channel on the monitoring terminal screen changing from green to flashing red, while the warning light strip at the edge of the screen illuminates. Acoustic warnings are emitted intermittently by a buzzer built into the terminal, with the frequency corresponding to the alarm level. A text report is automatically generated and displayed in a dedicated alarm information bar on the screen. The text report format is "Alarm: [Abnormal Type] feature detected; Ultra-low Frequency Energy Deviation: [Value], Resonance Gain Deviation: [Value], Delay Time Deviation: [Value]", where [Abnormal Type] is replaced with the specific name of the matched pattern, and [Value] is replaced with the actual calculated value in the real-time deviation vector. It can be understood that the duration of the early warning signal and subsequent operations can be interactively controlled by medical personnel through the monitoring terminal interface according to clinical procedures.

[0046] See Figure 5 This is a graph showing the evolution of feature weights during the iterative process of an adaptive weight network. This line graph illustrates the dynamic changes in the weights of three key sensitive features over 10 iterations in the infusion set flow control characteristic test, reflecting the automatic learning process of the adaptive weight network regarding feature importance. The iterative process clearly demonstrates the dynamic adjustment of feature weights, ultimately forming a priority order of "resonance gain > ultra-low frequency fluctuation > temperature delay," providing a quantitative basis for subsequent test optimization. The convergence trend of the weights verifies the effectiveness of the adaptive weight network, which can automatically focus on the features with the most business value, reducing reliance on human experience. The decay of the temperature delay weight suggests that the sampling frequency of environmental parameters can be appropriately reduced, allowing more computational resources to be invested in monitoring dynamic disturbances and steady-state fluctuations, thus improving test efficiency.

[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A test system for the flow control characteristics of an infusion set under dynamic and static conditions, characterized in that, The system includes: The signal acquisition module collects the raw flow time-series signal and related environmental parameters of the infusion set under different static pressure and dynamic disturbance modes, forming a raw signal set. The signal structuring module performs multi-level decomposition and recombination on the original signal set to generate a structured signal layer containing steady-state components, transient components, and environmental coupling components. The fingerprint construction module, based on the structured signal layer, constructs a dynamic flow fingerprint that reflects the inherent evolution law of flow control; The feature extraction module uses the dynamic traffic fingerprint to drive an adaptive weight network and iteratively calculates the key sensitive feature set of the traffic control system. The compliance assessment module generates a compliance judgment map of flow control characteristics based on the response trajectory of the key sensitive feature set under simulated abnormal operating conditions.

2. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 1, characterized in that, The raw flow time-series signal and associated environmental parameters of the infusion set under different static pressure and dynamic disturbance modes are collected to form a raw signal set, specifically including: Under multiple preset static pressure reference points, the stable flow output signal of the infusion set is continuously collected to obtain the static pressure flow signal; Under multiple preset frequencies and amplitudes of mechanical vibration disturbance, the real-time flow fluctuation signal of the infusion set is collected synchronously to obtain the dynamic disturbance flow signal; Throughout the test, temperature and humidity data of the environment where the infusion set is located, as well as pressure data of key points in the infusion tubing, were continuously collected to obtain a sequence of related environmental parameters. The static pressure flow signal, the dynamic disturbance flow signal, and the associated environmental parameter sequence are aligned along the time axis and packaged to form the original signal set.

3. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 1, characterized in that, The process of multi-level decomposition and recombination of the original signal set to generate a structured signal layer containing steady-state components, transient components, and environmental coupling components specifically includes: Empirical mode decomposition is performed on each traffic time-series signal in the original signal set to separate multiple intrinsic mode functions arranged from high frequency to low frequency; From the multiple intrinsic mode functions, modes with an energy percentage exceeding a preset threshold and a frequency distribution in an extremely low range are selected and aggregated and reconstructed into the steady-state components. From the plurality of intrinsic mode functions, modes that are strongly correlated with the frequency of the applied mechanical vibration disturbance are selected and aggregated and reconstructed into the transient components; Calculate the mutual information value between the associated environmental parameter sequence and the remaining modes in the plurality of intrinsic mode functions, and fuse the modes with mutual information values ​​greater than a set threshold with the corresponding environmental parameters to generate the environmental coupling component; The steady-state component, the transient component, and the environmental coupling component are collectively organized into the structured signal layer.

4. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 1, characterized in that, Based on the structured signal layer, constructing a dynamic flow fingerprint that reflects the inherent evolution of flow control specifically includes: For the steady-state components in the structured signal layer, calculate their sample entropy at multiple different time scales to form a steady-state complexity vector; For the transient components in the structured signal layer, extract their peak sequence and decay rate sequence in different perturbation periods to form a transient response vector; For the environmental coupling component in the structured signal layer, the rate of change of its fluctuation amplitude and phase lag angle under different environmental parameter gradients are analyzed to form an environmental sensitivity vector; The steady-state complexity vector, the transient response vector, and the environmental sensitivity vector are concatenated in time sequence and mapped to a dense vector of fixed dimension through an encoder network. The dense vector is the dynamic traffic fingerprint.

5. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 1, characterized in that, Using the dynamic flow fingerprint to drive an adaptive weight network, the key sensitive feature set of the flow control system is iteratively calculated, specifically including: Initialize a fully connected neural network as the adaptive weight network, with its input layer dimension matching the dimension of the dynamic traffic fingerprint; The dynamic traffic fingerprint is input into the adaptive weight network, and the hidden layer of the adaptive weight network outputs a set of initial feature weights. The initial feature weights are used to weight and fuse all signal components in the structured signal layer to generate an initial fused signal. The initial fusion signal is fed back to an auxiliary input channel of the adaptive weight network, concatenated with the dynamic traffic fingerprint, and then passed through the adaptive weight network again to update the feature weights. Repeat the weighted fusion and weight update steps until the change in the feature weights is less than the convergence threshold. At this point, the signal components with the highest weight values ​​are identified as the key sensitive feature set.

6. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 5, characterized in that, The key sensitive feature set includes ultra-low frequency fluctuation energy extracted from the steady-state component, resonance gain under specific frequency perturbation extracted from the transient component, and temperature-flow transfer delay time extracted from the environmental coupling component.

7. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 1, characterized in that, Based on the response trajectory of the key sensitive feature set under simulated abnormal operating conditions, the compliance judgment map of flow control characteristics is generated, specifically including: Construct a set of simulated abnormal operating conditions, which includes simulations of partial blockage in pipelines, relaxation of roller clamps, and fatigue of infusion set materials. Under each simulated abnormal operating condition, the original signal of the infusion set is reacquired, and through the processing flow of the structured signal layer and the dynamic flow fingerprint, the feature value sequence corresponding to the key sensitive feature set under the simulated abnormal operating condition is obtained. The feature value sequence of the key sensitive feature set under normal working conditions is used as the reference trajectory, and the feature value sequence obtained under abnormal working conditions is used as the test trajectory. The deviation vector of each test trajectory relative to its corresponding reference trajectory is calculated. Using different anomaly types in the simulated abnormal working condition set as one dimension and different features in the key sensitive feature set as another dimension, the calculated deviation vector values ​​are filled into the corresponding positions to form a two-dimensional matrix, which is the compliance judgment map.

8. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 7, characterized in that, The deviation vector is calculated using a dynamic time warping algorithm, which measures the minimum cumulative distance between two feature value sequences in terms of shape.

9. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 7, characterized in that, The compliance determination graph is further used for online monitoring, specifically in the following manner: The system acquires signals from the current infusion process in real time and extracts feature values ​​from the current key sensitive feature set to form a real-time feature trajectory. Calculate the real-time deviation vector between the real-time feature trajectory and the reference trajectory; The real-time deviation vector is matched with the pre-defined threshold regions of each abnormal pattern in the compliance judgment map; When the real-time deviation vector falls into a certain abnormal mode threshold area, a warning signal corresponding to the abnormal mode is triggered.

10. The infusion set flow control characteristic testing system under dynamic and static conditions as described in claim 9, characterized in that, The warning signals include visual warnings, acoustic warnings, and text reports containing the type of anomaly and the degree of deviation.

Images