Novel infusion alarm and infusion alarm system

Abstract

The invention relates to the technical field of medical monitoring, and discloses a novel infusion alarm and an infusion alarm system. The system comprises an infusion state monitoring module, an alarm decision module and an alarm execution module. A multi-source sensing unit of the infusion state monitoring module is distributed in a liquid level detection area, a drip chamber area and a patient contact area of an infusion tube assembly and collects optical refractive index, gravity change and real-time image data, and a data fusion unit performs heterogeneous signal fusion on the data to generate a multi-mode infusion state signal. A state analysis unit of the alarm decision module receives the signal and outputs a liquid level prediction signal and an infusion abnormal risk signal, and an alarm rule engine generates an alarm trigger threshold according to the signal. And a sound-light alarm unit of the alarm execution module responds to the threshold value to start a sound-light alarm sequence, a remote notification unit generates a graded alarm notification based on the abnormal risk signal, and a man-machine interaction interface outputs a comprehensive risk index.

Description

Technical Field

[0001] The present invention relates to the technical field of medical monitoring, in particular to a novel infusion alarm and an infusion alarm system. Background Art

[0002] During clinical infusion therapy, real-time monitoring of the infusion status and timely handling of abnormal situations play an important role in ensuring the safety of patient treatment. The current infusion monitoring methods used in clinical practice mostly rely on regular inspections by medical staff or manual observation by patients and their families, which has obvious limitations. On the one hand, medical staff are busy with their daily work and it is difficult to pay continuous attention to each infusion patient. Especially in departments with a large number of patients, it is easy to delay the end of infusion or the handling of abnormal situations due to long inspection intervals. On the other hand, patients and their families often lack professional medical knowledge and have limited ability to identify abnormal situations that may occur during the infusion process, which can easily lead to misjudgment or omission. Most existing infusion alarm devices use a single sensing method for monitoring, for example, triggering an alarm by detecting only changes in the liquid level within the infusion tube. This type of alarm device with a single sensing method is easily interfered with by external environmental factors in actual applications, resulting in low alarm accuracy and frequent false alarms or missed alarms. For example, when the infusion tube is deformed by external force, it may affect the accuracy of liquid level detection, causing the alarm device to falsely alarm; for example, when the infusion speed is slow, a single liquid level detection may not be able to capture the state of the infusion about to end in time, resulting in missed alarms. Existing infusion alarm devices are relatively simple in their alarm modes, typically only emitting simple audible and visual alarm signals, and the alarm signals are not graded according to the severity of the abnormality. This makes it difficult for medical staff to quickly determine the urgency of the abnormality after receiving the alarm signal, thus affecting treatment efficiency. Furthermore, most of these alarm devices lack remote notification capabilities. If medical staff are not near the patient room, they will not receive the alarm information in a timely manner, which may delay patient treatment. Summary of the Invention

[0003] The purpose of the present invention is to provide a new infusion alarm device and an infusion alarm system to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides a novel infusion alarm system, comprising: An infusion status monitoring module, comprising a multi-source sensing unit and a data fusion unit. The multi-source sensing unit is arranged in the liquid level detection area, the drip bucket area, and the patient contact area of ​​the infusion tube assembly to collect optical refractive index data, gravity change data, and real-time image data. The data fusion unit receives the optical refractive index data, gravity change data, and real-time image data and performs heterogeneous signal fusion to generate a multimodal infusion status signal. An alarm decision module, comprising a state analysis unit and an alarm rule engine. The state analysis unit receives a multimodal infusion state signal and outputs a liquid level prediction signal and an infusion abnormality risk signal. The alarm rule engine generates an alarm trigger threshold based on the multimodal infusion state signal and the liquid level prediction signal. An alarm execution module includes an audible and visual alarm unit, a remote notification unit, and a human-computer interaction interface. The audible and visual alarm unit initiates an audible and visual alarm sequence in response to an alarm trigger threshold. The remote notification unit generates a graded alarm notification based on an abnormal infusion risk signal. The human-computer interaction interface processes the liquid level prediction signal and the abnormal infusion risk signal and outputs a comprehensive risk index.

[0005] Preferably, the infusion status monitoring module includes an optical sensing unit and a gravity sensing unit. The optical sensing unit obtains the liquid level position change data in the infusion tube through infrared spectroscopy analysis technology and forms a liquid level fluctuation signal. The gravity sensing unit monitors the quality change data of the infusion bag in real time and forms a quality attenuation signal. The data fusion unit receives the liquid level fluctuation signal and the quality attenuation signal to realize time domain alignment and feature fusion, and generates the multimodal infusion status signal. The multimodal infusion status signal includes a liquid level analysis signal, a flow calculation signal and an abnormality detection signal.

[0006] Preferably, the state analysis unit includes a liquid level prediction model and an adaptive threshold detection model based on a time convolution network, the liquid level analysis signal includes liquid level height time series data, drip rate frequency data and tube pressure fluctuation data, the flow calculation signal includes unit time flow value, cumulative infusion total amount and preset remaining time data, the abnormality detection signal includes bubble concentration data, pipeline blockage probability data and infusion component abnormality index, the liquid level height time series data, drip rate frequency data and tube pressure fluctuation data are input into the time convolution network model to output the liquid level prediction signal, the bubble concentration data, pipeline blockage probability data and infusion component abnormality index are fused through the adaptive threshold detection model and the infusion abnormality risk signal is output.

[0007] Preferably, the alarm rule engine includes a dynamic threshold calculation submodule and a multi-level alarm strategy library. The dynamic threshold calculation submodule generates a dynamic alarm trigger threshold through a nonlinear mapping relationship based on the remaining liquid volume value in the liquid level prediction signal, the preset remaining time data and the bubble concentration threshold in the infusion abnormality risk signal. The multi-level alarm strategy library stores alarm response strategies for different infusion abnormality categories, and triggers the corresponding alarm response strategy when the infusion abnormality risk signal exceeds the preset risk level.

[0008] Preferably, the alarm execution module also includes an alarm priority determination unit and a cross-device linkage unit. The alarm priority determination unit receives a comprehensive risk index and historical alarm record data, generates a current alarm response priority through a weighted decision tree algorithm, and the cross-device linkage unit activates the corresponding sound and light alarm sequence and remote notification channel according to the alarm response priority. The remote notification unit sends an encrypted alarm data packet to a preset monitoring terminal through a wireless communication protocol.

[0009] Preferably, the human-computer interaction interface includes a three-dimensional infusion simulator and a risk visualization unit. The three-dimensional infusion simulator renders the liquid flow state and bubble movement trajectory inside the infusion pipeline in real time. The risk visualization unit generates a dynamic risk heat map based on the comprehensive risk index, and automatically retrieves the microscopic imaging data of the pipeline blockage position when the abnormal infusion risk signal exceeds the high risk threshold.

[0010] Preferably, the system also includes a self-learning optimization module, which includes a historical data analysis unit and a rule update engine. The historical data analysis unit continuously collects infusion completion rate data, false alarm number data and medical staff intervention records. The rule update engine dynamically adjusts the dynamic alarm trigger threshold in the alarm rule engine and the response parameters of the multi-level alarm strategy library based on the historical data analysis results.

[0011] Preferably, the sound and light alarm unit includes a multi-band sound generator and a parallel light source array. The multi-band sound generator switches warning sounds of different frequencies according to the alarm response priority. The parallel light source array projects a dynamic warning pattern on the surface of the infusion bag and starts the rotating warning beam when the comprehensive risk index reaches the highest level.

[0012] Preferably, the remote notification unit includes an encryption transmission submodule and an emergency contact polling mechanism. The encryption transmission submodule performs chaotic encryption on the abnormal infusion risk signal and transmits it via low-power Bluetooth. The emergency contact polling mechanism automatically switches to the backup contact list when no confirmation signal is received from the monitoring terminal within a preset time period.

[0013] Preferably, the present invention also includes a new infusion alarm device, which uses the above-mentioned infusion alarm system.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This new infusion alarm system significantly improves the comprehensiveness and accuracy of infusion status monitoring through the collaborative work of multi-source sensing units and data fusion units. The multi-source sensing units are deployed in multiple key areas of the infusion tube assembly and can simultaneously collect optical refractive index data, gravity change data, and real-time image data. These different types of data reflect various status information during the infusion process from multiple dimensions. The data fusion unit fuses these heterogeneous signals to generate a multimodal infusion status signal that combines the advantages of various types of data, effectively avoiding the problem of susceptibility to external interference under a single sensing method, making the monitoring of information such as infusion level, drip rate, and patient contact site status more reliable. The state analysis unit within the alarm decision module conducts in-depth analysis of multimodal infusion status signals. The output level prediction signal can predict the changing trend of the infusion level in advance, allowing medical staff to prepare in advance and avoid the risks associated with delayed infusion completion. The alarm rule engine generates alarm trigger thresholds based on the multimodal infusion status signals and the level prediction signal, making the alarm trigger conditions more scientific and reasonable, and reducing the possibility of false alarms and missed alarms. The setting of the alarm execution module enriches the implementation methods of the alarm and improves the effectiveness of the alarm. The sound and light alarm unit responds to the alarm trigger threshold to start the sound and light alarm sequence, which can promptly remind patients and nearby medical staff in the ward; the remote notification unit generates graded alarm notifications based on the abnormal infusion risk signal, so that medical staff who are not near the ward can also be informed of abnormal infusion conditions in a timely manner, and can judge the urgency of the situation according to the level of the alarm notification, give priority to emergency situations, and improve the efficiency of medical response. The human-computer interaction interface processes relevant signals and outputs a comprehensive risk index, presenting the risk status of the infusion process in an intuitive way, which is convenient for medical staff to quickly understand the overall situation and make corresponding treatment decisions. The system organically combines multiple modules to form a complete closed loop from monitoring, analysis to alarm. It not only improves the intelligence level of infusion monitoring, but also provides more comprehensive and efficient support for clinical infusion management, helps to optimize the allocation of medical resources, and improves the safety and comfort of patients during the infusion process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a working principle diagram of the novel infusion alarm system of the present invention; Figure 2 This is the working principle diagram of the infusion status monitoring module; Figure 3 This is the working principle diagram of the state analysis unit; Figure 4 This is the working principle diagram of the alarm execution module. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] See also Figure 1 The present invention provides a new infusion alarm and an infusion alarm system, which includes: an infusion status monitoring module, an alarm decision module and an alarm execution module.

[0018] The infusion status monitoring module consists of a multi-source sensing unit and a data fusion unit. The multi-source sensing units are deployed in the liquid level detection area, drip chamber, and patient contact area of ​​the infusion tube assembly, respectively collecting optical refractive index data, gravity change data, and real-time image data. The data fusion unit receives and fuses these heterogeneous signals to generate a multimodal infusion status signal consisting of a liquid level analysis signal, a flow calculation signal, and anomaly detection signals. The state analysis unit of the alarm decision module receives the multimodal infusion status signals and outputs a liquid level prediction signal using a temporal convolutional network model. This signal is then combined with an adaptive threshold detection model to generate an infusion anomaly risk signal. The alarm rule engine dynamically calculates the alarm trigger threshold based on the multimodal infusion status signal and the liquid level prediction signal. The audible and visual alarm unit of the alarm execution module activates the audible and visual alarm sequence in response to the alarm trigger threshold. The remote notification unit generates a graded alarm notification based on the infusion anomaly risk signal. The human-computer interface processes the liquid level prediction signal and the infusion anomaly risk signal and outputs a comprehensive risk index.

[0019] Example 1: See Figure 2 The optical sensing unit of the infusion status monitoring module adopts infrared spectroscopy analysis technology. It transmits an infrared light beam of a specific wavelength to penetrate the wall of the infusion tube. The receiving end detects the change in the refractive index of the light beam, captures the liquid surface position fluctuation in real time and generates a liquid level fluctuation signal. The infrared light source uses a narrow-band LED, and the wavelength range is controlled between 850nm and 950nm to reduce ambient light interference. The photoelectric sensor array at the receiving end collects the transmitted light intensity data at a sampling frequency of 100 times per second, and eliminates the baseline drift through the differential amplification circuit. The calculation of the liquid level fluctuation signal is based on the Beer-Lambert law. When the liquid level in the infusion tube drops, the liquid medium in the light beam path decreases, and the light intensity at the receiving end increases accordingly. The system converts the light intensity change into liquid level height data through the calibration curve.

[0020] The gravity sensing unit integrates a high-precision strain gauge and is mounted on the load-bearing component of the infusion bag suspension bracket. The strain gauge utilizes a full-bridge circuit design with a sensitivity coefficient of 2.0, a measuring range of 0 to 5 kg, and a resolution of 0.1 g. The gravity signal is converted by a 24-bit ADC and digitally filtered to eliminate noise caused by mechanical vibration. The mass decay signal is calculated using a sliding average algorithm, updating the remaining mass of the infusion bag every 10 seconds. This information is then converted to the remaining volume based on the density of the drug solution. The system automatically performs a zero-point calibration at the start of an infusion to eliminate the influence of the bag's own weight.

[0021] The data fusion unit uses a sliding window mechanism to align the liquid level fluctuation signal and the mass decay signal in the time domain. The sampling interval for optical sensor data is 10ms, and the sampling interval for gravity data is 100ms. The system uses linear interpolation to unify the two to the same time base. The feature extraction stage uses a wavelet transform, using the Db4 wavelet basis to perform a five-layer decomposition of the signal, extracting approximate coefficients and detail coefficients as fusion features. The liquid level analysis signal is processed using the Kalman filter algorithm, and the state equation is modeled as a linear change process of the liquid level height. The observation equation combines the optical and gravity measurements to output smoothed liquid level time series data.

[0022] The flow calculation signal calculates the flow rate per unit time based on the mass decay rate and the infusion tube diameter parameters. The system stores the nominal inner diameter data of different infusion tubes. When a mass change is detected, the instantaneous flow rate is calculated in combination with the liquid level height change rate. The cumulative total infusion volume is obtained by numerical integration, and the time step is set to 1 minute. The abnormal detection signal is generated through multimodal data cross-validation: a sudden change in the optical refractive index may indicate the mixing of bubbles. At this time, the system compares the continuity of the gravity signal. If the gravity changes smoothly, the presence of bubbles is confirmed; the judgment of pipeline blockage is based on the degree of deviation between the liquid level drop rate and the gravity change rate. When the difference between the two exceeds the threshold, a blockage warning is triggered.

[0023] The optical sensing unit monitors the drip chamber using a reflective measurement scheme. The infrared light source and receiver are mounted at a 45° angle. When a droplet passes through the detection area, the light beam scatters on the droplet surface, and the receiver captures the pulsed light intensity changes. Drip rate frequency data is extracted using a peak detection algorithm. The system counts the number of pulses per minute and converts it into a drip rate value. Tube pressure fluctuation data is acquired using a flexible pressure sensor attached to the outer wall of the infusion tube to measure the resistance change caused by tube wall deformation. The pressure signal is normalized and then input into the liquid level prediction model along with the liquid level data.

[0024] The data fusion unit outputs three types of signals: a level analysis signal records the current liquid level and its changing trend; a flow calculation signal provides real-time flow rate and remaining time estimate; and an anomaly detection signal identifies bubble concentration, risk of line blockage, and abnormal drug composition. The multimodal infusion status signal is generated using a weighted decision mechanism, with optical data weighted at 0.6, gravity data at 0.3, and pressure data at 0.1. If a sensor fails, the system automatically adjusts the weight distribution to ensure continuous monitoring.

[0025] The system's anti-interference design includes an ambient light compensation algorithm for the optical sensor. This algorithm uses a reference sensor to collect ambient light intensity and subtract background noise from the primary signal. The gravity sensing unit utilizes a temperature compensation circuit to mitigate the effects of thermal drift in the strain gauges. The data fusion unit features an outlier rejection function, automatically discarding and interpolating data points outside of the physically reasonable range. Real-time image data is captured by a micro-camera with a field of view covering critical areas of the infusion tubing assembly. Image processing algorithms identify the liquid surface edge and bubble outlines, complementing and verifying the optical measurement results.

[0026] The infusion status monitoring module's hardware is housed in a medical-grade plastic casing, and the internal circuit board is sealed to prevent moisture. A hydrophobic coating is applied to the optical sensor window to prevent droplets from adhering to and affecting measurement accuracy. The system is powered by a rechargeable lithium battery with a battery life of over 72 hours and support for charging-while-use mode. All sensor data is transmitted to the main control unit via an RS-485 bus. The communication protocol includes a CRC checksum to ensure data transmission reliability.

[0027] The calibration process is automatically executed upon system startup, including dark current calibration of the optical sensor, zero point calibration of the gravity sensor, and baseline calibration of the pressure sensor. Manual calibration can be triggered through the user interface, and the system guides the user through setting the standard liquid level reference point. Historical data storage records the last 30 days of monitoring data and supports USB export for offline analysis. The modular design allows for individual replacement of faulty sensors, with the system automatically recognizing the new device and recalibrating.

[0028] The layout of the multi-source sensing units was optimized through fluid dynamics simulation. The optical sensor in the liquid level detection area is installed 15 cm from the drip chamber, avoiding the bend in the infusion tubing. The gravity sensing unit is rigidly connected to the infusion bag's suspension point to reduce measurement errors caused by the mechanical structure. The temperature sensor in the patient contact area is isolated to prevent skin contact from affecting temperature measurement accuracy. The time synchronization accuracy of each sensor is controlled within 10ms to ensure the effective time alignment of heterogeneous data.

[0029] The system's fault self-diagnosis function regularly checks sensor status. Optical sensors verify optical path integrity by emitting self-test signals; gravity sensors perform full-scale self-tests, applying standard weights to verify measurement linearity; and pressure sensors check for zero drift. When a fault is detected, the system displays a specific error code on the human-machine interface and automatically switches to redundant sensor operation.

[0030] Example 2: See Figure 3 The liquid level prediction model of the state analysis unit adopts a time convolutional network structure. The network input layer receives three types of time series data from the infusion state monitoring module: liquid level measurement value, drip rate frequency sampling value and tube pressure fluctuation reading. The liquid level data is input at a rate of 10 samples per second. The median filter is used in the data preprocessing stage to remove transient interference and retain the overall trend of the liquid level change. The drip rate frequency data comes from the monitoring of the drip bucket area by the optical sensor. The system counts the number of droplets per minute and converts it into Hertz units. The tube pressure fluctuation data is collected by a flexible pressure sensor. The original signal is band-pass filtered to extract the effective components in the range of 0.1Hz to 10Hz. This frequency band contains the main pressure characteristics in the infusion pipeline.

[0031] The architecture of the temporal convolutional network consists of six layers of dilated causal convolutions, with kernel width set to 3 in each layer and exponentially increasing dilation factors. The dilation factor of the first layer is 1, followed by 2, 4, 8, 16, and 32 in subsequent layers. This design enables the network to capture liquid level patterns at different time scales. Each convolution layer is followed by a Reluctant Unit (ReLU) activation function and batch normalization to prevent gradient vanishing and accelerate convergence. A skip connection mechanism transfers shallow-layer features directly to the deep network, preserving multi-scale temporal information. The network output layer uses a fully connected structure, mapping the temporal features into liquid level predictions for the next five minutes, with a configurable prediction step size of one minute.

[0032] The adaptive threshold detection model processes three types of abnormal indicators: bubble concentration data, pipeline blockage probability data, and infusion composition abnormality indicators. The bubble concentration data comes from the scattered light intensity analysis of the optical sensor. When bubbles appear in the infusion pipeline, the transmittance of the infrared light beam will show periodic fluctuation characteristics. The system calculates the number and amplitude of fluctuation peaks per unit time and generates a bubble concentration indicator based on historical baseline data. The pipeline blockage probability data is obtained through pressure gradient analysis. Under normal infusion conditions, the tube pressure fluctuation shows a smooth periodicity. When blockage occurs, the pressure waveform will show a steep rise or continuous high pressure characteristics. The system extracts the time domain statistics of the pressure signal, including the mean, variance, and peak-to-peak value, and inputs them into the logistic regression model to calculate the blockage probability.

[0033] The abnormal infusion composition indicator uses a spectral matching algorithm. The optical sensor collects the transmission spectrum of the drug solution in the visible light band and compares it with the pre-stored standard spectrum library for similarity. The standard spectrum library contains the characteristic absorption peak data of common infusion drugs at specific concentrations. The matching algorithm calculates the correlation coefficient between the real-time spectrum and the standard spectrum. When the correlation coefficient is lower than the threshold, a component abnormality warning is triggered. The adaptive threshold detection model uses a sliding window statistical method with a window size set to 5 minutes to calculate the moving average and standard deviation of each indicator. The threshold of the bubble concentration indicator is set to 3 times the standard deviation of the baseline value, the warning threshold of the pipeline blockage probability is 0.7, and the judgment threshold for abnormal infusion composition is a spectral correlation coefficient of 0.5.

[0034] The training data for the liquid level prediction model comes from historical records of clinical infusion processes. The dataset contains 200 sets of time series data for complete infusion processes, with each set lasting between 2 and 8 hours. Data augmentation technology expands the training sample to five times its original size by time-stretching the original data, adding Gaussian noise, and randomly varying the sampling interval. The network is trained using the Adam optimizer, with an initial learning rate set to 0.001, which decays every 50 training cycles. The loss function uses a smoothed L1 loss to strike a balance between mean squared error and absolute error to prevent outliers from having a significant impact on the training process. An early stopping strategy is used during training, terminating training when the validation set loss has not decreased for 10 consecutive cycles.

[0035] The adaptive threshold detection model utilizes an online learning mechanism for parameter adjustment. The system continuously records the handling results of each alarm event and medical staff feedback. When a false alarm occurs, the threshold of the relevant indicator is automatically adjusted upward by 5%; when a false alarm occurs, the threshold is correspondingly adjusted downward by 3%. The adjustment range adopts a gradual strategy, with larger changes initially to facilitate rapid convergence, and gradually decreasing the adjustment step size to improve stability. The model maintenance interface allows medical staff to manually adjust the sensitivity of specific indicators, limiting the adjustment range to within ±20% of the baseline value to prevent parameters from deviating from the appropriate range.

[0036] The state analysis unit operates in two parallel threads: real-time data processing and periodic model updates. The real-time data processing thread receives sensor data in a 100ms cycle, performs preprocessing, and then feeds it into the liquid level prediction model. The prediction results, combined with anomaly detection indicators, generate an infusion status report. The periodic model update thread automatically starts each morning to check for new training data. If so, it performs incremental training to update model parameters. The model version management function records the timestamp and modification details of each update, supporting rollback to previous versions to address performance degradation after an update.

[0037] Anomaly detection signals are processed hierarchically using a state machine mechanism. The system defines three risk states: observation, warning, and emergency. The observation state triggers internal system recording but does not generate an alarm, marking potential risk trends. The warning state activates visual cues on the human-computer interface. The emergency state triggers audible and visual alarms and remote notifications. State transition conditions are based on the duration and severity of the anomaly indicator. For example, the bubble concentration indicator must exceed the limit for three consecutive sampling cycles before it is upgraded from observation to warning. The system maintains an abnormal event log, recording the time, indicator value, and handling measures of each state transition for subsequent analysis.

[0038] The credibility of liquid level prediction signals is assessed using an ensemble learning approach. The system runs three prediction models with identical structures but different initialization parameters in parallel, and determines the final output through a voting mechanism. When the difference in prediction results between models exceeds a threshold, the system marks the prediction for that period as low credibility and displays a corresponding warning on the human-computer interface. The credibility indicator also influences the alarm triggering strategy. For prediction results in low-credibility periods, the system automatically raises the alarm threshold to reduce the possibility of false alarms. The prediction results are displayed in the form of interval estimates. In addition to the central prediction value, a 95% confidence interval is also displayed to help medical staff understand the uncertainty range of the prediction.

[0039] The system optimizes real-time performance using a multi-level caching strategy. Raw sensor data is stored in a circular buffer, preprocessed feature data is stored in an in-memory database, and prediction results and anomaly indicators are written to persistent storage. Computationally intensive tasks, such as convolution and spectrum analysis, are accelerated through instruction set optimization and parallel computing. Time-critical tasks, such as alarm detection, run on real-time operating system threads, ensuring response latency of no more than 200ms. A resource monitoring module continuously tracks CPU and memory usage, automatically reducing data processing frequency when load is excessive to maintain stable system operation.

[0040] The interface between the status analysis unit and other modules utilizes a publish-subscribe model. Infusion status reports are encapsulated in JSON format and include fields such as timestamp, predicted liquid level, abnormality indicators, and risk level. Modules that subscribe to this data include an alarm rules engine and a human-computer interface. The system ensures real-time and consistent data distribution. The interface protocol defines a comprehensive error handling mechanism. If the receiver does not respond for an extended period, the sender automatically retransmits and logs the communication failure. The data encryption module performs field-level encryption on sensitive medical information, ensuring patient privacy and compliance with medical data security standards.

[0041] Hardware acceleration technology is applied to key computational steps in the state analysis unit, with convolutional neural network inference deployed on a dedicated AI accelerator chip, achieving low-power, high-performance computing. The statistical computation portion of the anomaly detection algorithm is offloaded to a coprocessor, freeing up main CPU resources. The system supports hot-swappable hardware upgrades, with new computing resources automatically added to the processing cluster, expanding overall analytical capabilities. The power management unit dynamically adjusts processor frequency based on workload, optimizing energy efficiency while ensuring real-time performance.

[0042] Maintenance and diagnostic functions are integrated into the state analysis unit firmware. The system regularly performs self-tests to verify the integrity of the computing core and memory. The debug interface outputs detailed operation logs, including input data, intermediate features, and output results for each inference cycle, for troubleshooting. Performance analysis tools analyze the time distribution of each processing stage to help optimize algorithms and resource allocation. The firmware supports remote secure updates. Update packages are automatically installed after digital signature verification, and basic monitoring functions are maintained during the update process.

[0043] Example 3: See Figure 4 The dynamic threshold calculation submodule of the alarm rule engine uses a nonlinear mapping algorithm to process liquid level prediction signals and infusion abnormality risk signals. This module receives input data including the remaining liquid volume value, the preset remaining time data, and the bubble concentration threshold, and generates the dynamic alarm trigger threshold using the following relationship:

[0044] in, Indicates the dynamic alarm trigger threshold, is the baseline alarm threshold, and is the adjustment coefficient of the remaining liquid volume, is the current remaining liquid volume, is the initial total capacity of the infusion bag, is the bubble concentration influencing factor, is the bubble concentration value detected in real time, is the safe upper limit of bubble concentration. This formula uses an exponential decay term to reflect the trend of increasing alarm sensitivity as the remaining liquid decreases, while introducing a linear suppression term to reduce the risk of false alarms caused by bubble interference.

[0045] The multi-level alarm strategy library uses a tree structure to store response strategies for different types of infusion anomalies. Nodes are divided into four layers: the root node corresponds to the no-alarm state, the first-level sub-nodes handle low-risk anomalies (such as a temporary excess of bubble concentration), the second-level sub-nodes target medium-risk events (such as a partial blockage in the pipeline), and the third-level sub-nodes manage high-risk conditions (such as abnormal infusion composition). Each node stores three types of parameters: trigger conditions, response actions, and escalation rules. Trigger conditions are defined by Boolean expressions, such as "the bubble concentration continues to exceed the limit for 60 seconds and the liquid level drop rate is lower than the threshold"; response actions include the selection of audible and visual alarm modes, remote notification content templates, and interface prompt priorities; and escalation rules determine when to jump from the current node to a higher-risk node, such as "if the current alarm is not confirmed within 5 minutes, it will automatically escalate to the next level."

[0046] The alarm priority determination unit uses a weighted decision tree algorithm to process the comprehensive risk index and historical alarm record data. The feature vector of the decision tree contains five indicators: the current risk score, the historical frequency of similar alarms, patient vital signs, environmental factors, and time parameters (such as night mode). The weights of each feature are obtained through offline training, and the training data is derived from alarm handling records in real scenarios. Each non-leaf node of the decision tree corresponds to a binary judgment condition, such as "Is the comprehensive risk index greater than 75?" The leaf node outputs the specific alarm response priority (level 1-5). The branch path is dynamically adjusted during operation. For example, when a patient is detected in intensive care, the initial priority of all alarms is automatically increased by one level.

[0047] The cross-device linkage unit enables the coordinated work of the alarm execution module and the hospital information system. The unit maintains a device registry to record the communication addresses and status information of various monitoring terminals. When the alarm is triggered, the unit first checks the online status of the target device and gives priority to the most recently active device to establish a connection. The communication protocol adopts a hybrid mode: the HL7 standard format is used for text-based alarm notifications to ensure compatibility with the hospital information system; a proprietary binary protocol is used for real-time data streams to ensure transmission efficiency. The generation process of the encrypted alarm data packet includes three steps: the original data is serialized into JSON format, AES-256 encryption is performed using a symmetric key generated based on the infusion bag ID, and finally a digital signature is attached to prevent tampering.

[0048] The execution logic of the audio and visual alarm unit is strictly tied to the alarm priority level. A low-frequency beep (500Hz square wave, 1-second interval) corresponds to a priority 1 alarm, primarily alerting nearby medical staff. A medium-frequency intermittent tone (1kHz pulse, 0.5-second duty cycle) is used for priority 2. A high-frequency continuous tone (2.5kHz sine wave) triggers priority levels 3 and above. The light array's warning modes are similarly graded: a single-color slow flash (yellow, 1Hz) indicates low priority, a two-color alternating flash (yellow and red, 2Hz) indicates medium priority, and a full red high-speed flash (5Hz) is used for emergencies. When the comprehensive risk index reaches the highest level, the unit activates the rotating warning beam. The motor drives the laser module to rotate at 10rpm, creating a dynamic light spot on the ward wall, enhancing the visual warning effect.

[0049] The emergency contact polling mechanism of the remote notification unit adopts a timeout retry strategy. The system maintains two independent contact lists: the main list contains regular contacts such as responsible nurses and doctors on duty, and the backup list extends to relevant personnel such as department directors and pharmacists. A 5-minute countdown starts after the first alarm is sent. If no confirmation signal is received from the recipient, the alarm is automatically sent to the next contact in the backup list. The polling order follows the principle of job hierarchy and professional relevance. For example, for alarms of abnormal infusion components, the pharmacist's polling order will be advanced. A detailed log is recorded for each contact attempt, including the sending time, receiving device and response status. These data are used for subsequent optimization of the polling strategy.

[0050] The alarm information display of the human-computer interaction interface adopts a layered design. The main interface maintains an overview of the overall status of the infusion process, including a bar graph of the remaining liquid volume, a real-time digital display of the drip rate, and a risk level indicator. When an alarm is triggered, the interface automatically pops up a secondary details window, displaying the abnormality type, occurrence time, current disposal recommendations, and historical statistics of similar events in columns. The third-level in-depth information is hidden through a sliding panel, and medical staff need to actively expand and view it. It contains technical details such as the sensor's raw data waveform, the prediction model confidence interval, and the alarm threshold calculation process. The interface supports multiple interaction methods: touch screen operation for routine confirmation and alarm cancellation, physical buttons are reserved for emergency silencing, and voice commands can query the alarm handling guide.

[0051] The system's fault-tolerance mechanism operates throughout the entire alarm execution process. The network communication layer implements automatic reconnection and cached transmission. When the wireless signal is interrupted, the alarm information is temporarily stored locally and reissued when the connection is restored. The device fault detection module continuously monitors parameters such as the LED brightness and speaker impedance of the audio and visual alarm unit, automatically switching to a backup alarm mode if a hardware anomaly is detected. Data consistency is ensured through a transaction processing mechanism. The generation, transmission, and confirmation of any alarm event are recorded in persistent storage, and the complete context can be restored after a system restart.

[0052] The alarm rule engine's configuration interface provides fine-grained parameter adjustment. Clinical administrators can modify default thresholds, delay parameters, and escalation conditions for various alarm types. Adjustments take effect in real time and are synchronized to all associated devices. The system retains modification history and supports parameter rollback and solution comparison. For key parameters, modifications require dual authentication to prevent security risks caused by misoperation. The rule testing sandbox function allows administrators to simulate various abnormal scenarios and verify that alarm responses meet expectations without affecting actual monitoring work.

[0053] Time synchronization services ensure accurate and consistent timestamps for alarm events. The system uses the NTP protocol to synchronize with the hospital's time server, automatically correcting any local clock drift exceeding 100ms. Key time nodes are recorded with millisecond accuracy, providing accurate time series data for subsequent analysis. For offline operations, the system uses a hardware RTC module to maintain a time base, automatically compensating for time differences upon reconnection.

[0054] The alarm execution module uses an intelligent scheduling strategy to manage energy consumption. During normal monitoring, the audible and visual alarm unit operates in a low-power standby mode, maintaining power only to essential circuits. When the prediction model indicates an increased risk probability, the unit pre-heats and enters a rapid response state. The wireless communication module dynamically adjusts transmit power based on network quality, reducing power consumption by over 30% in areas with good signal strength.

[0055] Example 4: A three-dimensional infusion simulator with a human-computer interaction interface is constructed based on a real-time fluid dynamics model. The model describes the liquid flow state in the infusion line through the following coupling relationship:

[0056] in, represents the fluid state parameter, is the velocity potential function, is the kinematic viscosity coefficient of the liquid, is the pipeline resistance factor. The discrete solution of this formula uses the finite volume method to divide the infusion tube into 1000 micro-element grids, and the pressure, flow velocity, and flow direction parameters are calculated for each grid node. The 3D rendering engine receives these physical parameters and simulates the flow effect of the liquid medicine through a particle system. The diameter of a single particle is set to 0.1mm, and the color gradient changes from dark at the inlet to light at the outlet, intuitively displaying the liquid medicine update process. The bubble motion trajectory is rendered using special marker particles. The bubble particles have a translucent appearance and dynamic size change characteristics. When the bubble concentration is detected to be increased, the particle size is enlarged by 150% to improve visibility.

[0057] The heat map generation algorithm of the risk visualization unit uses bilinear interpolation technology to divide the two-dimensional expansion surface of the infusion pipeline into a temperature field grid. The risk value of each grid point is The risk index is determined by the following factors: predicted liquid level deviation, real-time drip rate fluctuation, and pipeline pressure gradient. Five risk levels are assigned, corresponding to blue (safe), green (normal), yellow (caution), orange (warning), and red (danger). The heat map refreshes at 30 frames per second, and color transitions utilize HSL color space interpolation to ensure visual continuity. When the comprehensive risk index exceeds a preset threshold, the heat map automatically overlays black contour lines to mark the boundaries of high-risk areas.

[0058] The trigger logic of the microscopic imaging system is deeply bound to the risk signal of abnormal infusion. The system continuously monitors the probability data of pipeline blockage. When the value exceeds 0.85 for 10 consecutive seconds, the micro endoscope is controlled to extend to the nearest suspicious position. The endoscope adopts a flexible fiber optic bundle design with a diameter of only 1.2mm. It can move to the target area along the preset channel on the outer wall of the infusion tube. The imaging module contains two sets of lenses: a wide-angle lens (viewing angle of 120°) is used to quickly locate the blockage point, and a microscope lens (magnification of 40X) provides detailed observation. The lighting system consists of 8 white light LEDs arranged in a ring. The brightness is automatically adjusted according to the ambient light to ensure that the image signal-to-noise ratio is better than 35dB. After the collected image data is processed by distortion correction and edge enhancement, it is superimposed on the corresponding position of the three-dimensional infusion simulator to form a multi-scale observation perspective.

[0059] The historical data analysis unit of the self-learning optimization module establishes a multidimensional feature warehouse, storing four key metrics: infusion completion rate, number of false alarms, medical staff intervention records, and equipment operation logs. The data warehouse utilizes a time series database structure, with each record containing a precise timestamp, event type, and associated parameters. The analysis engine performs batch processing tasks daily at dawn, primarily completing three tasks: feature extraction (such as calculating the weekly average false alarm rate), pattern recognition (discovering correlations between alarms and specific drug solution types), and anomaly detection (identifying trends in sensor performance degradation). The analysis results are stored as structured reports for the rule update engine to access.

[0060] The rule update engine uses incremental learning strategy to adjust system parameters. For dynamic alarm trigger thresholds, the adjustment amount Determined by the following rules: If a certain type of alarm is manually canceled by medical staff more than 3 times within 24 hours, the corresponding threshold will be increased If the system does not detect an actual risk event (confirmed by subsequent manual review), the threshold is reduced. The response parameters of the multi-level alarm policy library are updated using a game theory approach, using the efficiency of medical personnel as the payoff function and searching for the optimal response time window through iterative optimization. For example, for a bubble alarm, the system might adjust the initial response time from 5 minutes to 3 minutes, while extending the secondary reminder interval to 10 minutes.

[0061] The interactive features of the 3D infusion simulator support multiple operating modes. Use a pan gesture to view any section of the pipeline, pinch to zoom to adjust the viewing angle, and long-press a specific area to pop up a context menu. Menu options include "Mark Suspicious Points," "Compare to Historical Data," and "Generate Snapshot Report." The comparison function calls up infusion curves from similar cases in the historical database and displays them in a semi-transparent overlay to assist medical staff in determining whether the current situation is abnormal. The snapshot report automatically captures the current 3D scene and risk heat map, attaches a timestamp and key parameter readings, and can be directly shared with the consultation team through the hospital information system.

[0062] Advanced features of the Risk Visualization Unit include trend prediction maps. This feature extends the heat map display over a future timeline based on the output of the liquid level prediction model. The prediction map uses fuzzy rendering technology, gradually reducing transparency over time, to intuitively demonstrate the growing uncertainty of the prediction results. Medical staff can view the predicted risk distribution at different times by sliding the timeline, assisting in the development of intervention plans. The map also supports annotation, allowing text annotations or arrow marks to be added at specific time points. These annotations are automatically synchronized with the associated medical records.

[0063] The self-learning optimization module's feedback loop design consists of three levels. Primary feedback automatically fine-tunes parameters, such as adaptively updating sensor calibration coefficients. Intermediate feedback requires confirmation from the head nurse and involves structural adjustments to alarm rules. Advanced feedback triggers an expert review process. When the system detects contradictory learning signals (such as polarized responses to a particular alarm type), a consultation ticket is automatically generated and sent to the technical committee. Feedback processing results are pushed to relevant medical staff through the system's message center to ensure transparency.

[0064] The microscopic imaging system's image analysis algorithm enables automated blockage diagnosis. After acquiring an image of the pipeline interior, the system performs the following processing: Gaussian filtering to remove noise, Canny edge detection to extract the pipe wall contour, and morphological closing operations to fill small gaps. The degree of blockage is assessed based on the proportion of unobstructed cross-sectional area. When this proportion falls below 30%, the system overlays a red warning box on the image and automatically raises the alarm priority for that area. In the event of complete blockage, the system immediately triggers the highest level alarm, bypassing the conventional delayed judgment process.

[0065] The user interface's accessibility is designed with diverse usage scenarios in mind. High-contrast mode compresses the risk heatmap's color gamut to black and white grayscale, making it suitable for users with color vision impairments. Large text mode magnifies all digital readouts by 200%, making them easier to read from a distance. A voice announcement function provides aloud readings of the current risk status and pending alerts. The interface layout utilizes a responsive design, automatically adjusting control density and font size for tablet and desktop monitors. Tooltips are provided for all interactive elements; long-pressing any icon or data point displays provides detailed information.

[0066] The self-learning optimization module's performance monitoring continuously tracks learning outcomes. Key performance indicators include a decline in false alarm rates, reductions in alarm response times, and healthcare provider satisfaction scores. The monitoring dashboard uses statistical process control charts to display trends in these indicators. Abnormal fluctuations outside control limits automatically trigger root cause analysis. The analysis results are used to adjust the learning algorithm's hyperparameters, such as the weighting coefficient for historical data and the time decay factor. The system generates monthly learning effectiveness evaluation reports, listing the most significant rule changes and their impact, to help healthcare institutions understand the evolving logic of system behavior.

[0067] The 3D infusion simulator's physical parameter database contains the rheological properties of over 200 common medications. Once the system recognizes the infusion bag label (either by scanning a barcode or manually entering data), it automatically loads the corresponding medication's viscosity-temperature curve, surface tension coefficient, and density parameters. This data influences the fluid dynamics model's calculations in real time, ensuring the simulation is more realistic. The database supports manual expansion, allowing medical staff to add new medication parameters or modify existing data. All modifications must be electronically signed by the responsible pharmacy department.

[0068] The Risk Visualization Unit's multi-screen collaboration feature allows different displays to be projected onto secondary monitors. While the primary screen maintains the regular monitoring interface, secondary monitors can focus on microscopic images or forecast maps of high-risk areas. Display modes can be quickly switched by dragging and dropping, and layouts can be saved as personalized presets. During collaborative operation, all operational commands are recorded simultaneously for easy retrospective analysis. The system supports simultaneous connection of up to three external monitors, each with independently configurable refresh rate and resolution parameters.

[0069] Example 5: The multi-band sound generator of the audio and visual alarm unit utilizes digital direct synthesis technology and includes 32 pre-programmed warning tones covering the auditory-sensitive frequency range of 200Hz to 5kHz. The audio waveform can be selected from square, sine, or pulse waves, each corresponding to a different level of urgency. Low-frequency square waves (200-800Hz) are used for routine alerts, generating non-intrusive background alerts; medium-frequency sine waves (800-3kHz) are suitable for medium-risk situations requiring immediate attention; and high-frequency pulse waves (3-5kHz) are specifically designed for emergency situations requiring immediate intervention. The volume control utilizes a dynamic compression algorithm, automatically adjusting the output intensity based on the ambient noise level, maintaining a constant 60dB in quiet wards and increasing to 75dB in noisy environments. The acoustically optimized sound propagation pattern uses a waveguide structure to focus sound energy toward areas where medical staff reside, minimizing disruption to patients.

[0070] The parallel light source array is composed of high-brightness RGB LEDs. Each group of LEDs is independently controllable, and 256 levels of brightness can be adjusted through PWM dimming. The array is installed on the top of the infusion alarm and projects light downward to cover the entire surface of the infusion bag. The standard warning mode uses a static color block display. Green indicates normal system operation, yellow indicates a low-level alarm, and red indicates a problem that requires immediate attention. The advanced warning mode activates dynamic patterns, including a clockwise rotating halo, a breathing pulsating effect, and a beam that sweeps left and right. The pattern generation algorithm takes into account the curved shape of the infusion bag, and pre-calibrated distortion correction ensures that the projected image remains geometrically accurate when viewed from different angles. The color temperature of the light source is coordinated with the ward lighting environment to avoid visual fatigue or misjudgment.

[0071] The rotating warning beam system integrates a red laser module and a stepper motor drive mechanism. The laser output power is strictly controlled within the Class 2 safety standard, and the beam diameter is expanded to 5mm through an optical lens group, forming a clearly visible light spot at a distance of 2 meters. The motor adopts micro-stepping drive technology, and the rotation speed can be adjusted steplessly in the range of 5-30rpm. The beam projection angle is automatically calibrated by a gravity sensor to ensure horizontal scanning in any installation posture. When the comprehensive risk index reaches the highest level, the system activates emergency mode, and the beam changes from continuous rotation to rapid flashing (8Hz) and the motor speed is increased synchronously to enhance the visual impact. The safety protection mechanism includes a motion detection sensor that immediately suspends the motor operation when the beam path is accidentally blocked.

[0072] The remote notification unit's encrypted transmission submodule provides end-to-end data protection. The encryption process utilizes a key derivation scheme based on the infusion bag identifier. The master key is generated using the SHA-3 algorithm from the device's unique identifier and a hash of the current date. Each communication session uses a different temporary key to effectively prevent replay attacks. The data encapsulation format complies with the extended standard for medical device communication protocols and consists of an encrypted payload, a digital signature, and an integrity check code. The transport layer utilizes adaptive protocol selection, using a low-latency BLE 5.0 connection when signal strength is good and switching to a LoRa wide-area network for longer-distance scenarios. The packet retransmission mechanism utilizes an exponential backoff algorithm, with an initial retry interval of 2 seconds and a maximum of no more than 60 seconds.

[0073] The emergency contact polling mechanism establishes a multi-level response network. The main contact list is sorted by division of responsibilities. The default first contact person is the responsible nurse, the second contact person is the doctor on duty, and the third contact person is the department head. The backup contact list is extended to relevant specialist consulting doctors and pharmacist teams. The polling strategy adopts intelligent timeout judgment. The waiting time for confirmation of routine alarms is 5 minutes, and the high-risk alarm is shortened to 2 minutes. The synchronous group mode is activated in extreme emergencies. The contact method is automatically selected according to the contact profile, including the hospital paging system, encrypted SMS and dedicated App push. A digital receipt is generated for each contact attempt, recording the exact sending time and receiving device model.

[0074] The alarm status persistent storage module ensures system reliability. All alarm events and their processing status are written to non-volatile memory, using write-ahead logging technology to prevent data corruption. The storage structure is organized by time slices, with a new data file generated every 24 hours and old files automatically compressed and archived. The stored content includes complete contextual information: trigger timestamp, sensor raw data snapshot, risk assessment process record, execution action list, and final disposal result. Storage encryption uses a hierarchical key scheme, with metadata protected by a device-wide universal key and sensitive medical information encrypted twice using a patient-specific key. Storage space management uses a circular buffer mechanism, automatically deleting the oldest non-critical event record when the capacity reaches 90%.

[0075] The environmental adaptation function of the audio and visual alarm unit enables intelligent adjustment. An ambient light sensor continuously monitors the patient room lighting conditions and dynamically adjusts the brightness of the LED array to ensure that the warning pattern is clearly visible in all lighting conditions. A sound propagation analysis algorithm uses a built-in microphone to collect echo characteristics, optimize the speaker output frequency response curve, and compensate for the attenuation caused by the room's acoustic characteristics. A temperature monitoring circuit prevents overheating of the laser module, automatically reducing output power when the internal temperature exceeds a safe threshold. Electromagnetic compatibility design includes RF shielding and filtering circuits to prevent interference with the normal operation of other medical equipment.

[0076] The remote notification unit's cross-platform compatibility is achieved through protocol conversion middleware. This middleware converts the system's native alarm signals into multiple standard medical communication protocols, including HL7, FHIR, and DICOM. The conversion process preserves all critical data fields and dynamically adjusts the level of detail based on the capabilities of the receiving system. For non-intelligent terminals, the system supports a degraded transmission mode, simplifying the rich alarm information into pre-defined standardized status codes. A reverse communication channel processes the receiving end's confirmation messages and remote control commands, executing the corresponding actions after strict authentication and permission checks.

[0077] The ergonomic design of the audible and visual alarm unit has undergone multiple rounds of clinical testing. Its dimensions conform to medical equipment installation specifications, and its corners are smoothly rounded to prevent scratches. The surface is constructed of an antibacterial composite material that withstands conventional disinfectants. The physical button layout is intuitive for emergency operations, with a prominent red emergency stop button and a protective cover to prevent accidental touches. The status indicator light uses a multi-color combination code, allowing you to determine system status by color and flashing pattern, even at difficult viewing angles. It supports wall-mounted, ceiling-mounted, and mobile bracket installation options to accommodate diverse ward layout requirements.

[0078] The exception handling process of the emergency contact polling mechanism covers various edge cases. When a contact is detected to be on vacation, the system automatically skips the contact and marks the status. In the event of a communication failure, the polling order is temporarily adjusted to prioritize other contact methods. For recurring unresponsive contacts, the system generates a management report prompting the update of contact information. After completing the standard polling process, major alarm events will additionally trigger the on-duty director's notification path. This function must be manually activated through two-factor authentication. Detailed logs are recorded for all exceptions, including the alternative measures taken by the system and the final resolution status.

[0079] The alarm execution module's energy optimization strategy balances response speed and battery life. Under normal monitoring conditions, the wireless communication module maintains a low-power listening mode, waking up every 10 seconds to check network status. The audio and visual alarm unit maintains only a minimum standby current when inactive. When the predictive model indicates an increase in risk probability, the system gradually preheats each subsystem: first activating the communication link, then starting the sound generator for preheating, and finally preparing the light source component. This phased wake-up strategy can reduce burst power peaks by 40%. The power management unit monitors battery health in real time and automatically adjusts the power allocation strategy when it detects a decrease in battery capacity due to aging.

[0080] The system's self-check function covers all alarm-actuating components. Comprehensive diagnostic tests are performed daily, including LED brightness calibration, speaker frequency response testing, laser power measurement, and motor torque checks. When problems are detected, the system takes action based on severity: minor anomalies log maintenance reminders, significant defects trigger degraded operation, and critical failures immediately switch to a backup alarm mode and notify a technician. Self-check results are summarized in a device health report, and historical trends can be viewed through the management interface. Maintenance mode supports offline testing, and all diagnostic operations do not interfere with normal monitoring functions.

[0081] The remote notification unit's message priority management system implements intelligent traffic control. Transmission strategies are dynamically adjusted based on alarm level, recipient role, and current network conditions. High-priority messages receive dedicated bandwidth, temporarily interrupting lower-priority data transmission. Message queue management utilizes a weighted fairness algorithm to prevent a single alarm type from monopolizing communication resources. Transmission content optimization includes delta compression (sending only modified data fields) and intelligent caching (repeating alarms reuses previously transmitted static information). A network status dashboard displays real-time signal strength, packet loss rate, and transmission latency, helping administrators assess communication reliability.

[0082] The degraded operating mode of the audio and visual alarm unit ensures the reliability of core functions. When a partial component failure is detected, the system automatically reconfigures the remaining healthy resources. For example, if the speaker is damaged, the light source warning intensity is increased and the device's vibration motor is activated; if the LED array fails, the screen flashes full screen instead. Although the alarm effect is weakened in degraded mode, basic warning functions are still guaranteed. All degraded operations are accompanied by clear system status prompts to inform medical staff of the current restrictions. After the fault is restored, the system automatically performs a functional verification test and exits the degraded mode after confirming normal operation.

[0083] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0084] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A new infusion alarm system, characterized in that: include: An infusion status monitoring module, comprising a multi-source sensing unit and a data fusion unit. The multi-source sensing unit is arranged in the liquid level detection area, the drip bucket area, and the patient contact area of ​​the infusion tube assembly to collect optical refractive index data, gravity change data, and real-time image data. The data fusion unit receives the optical refractive index data, gravity change data, and real-time image data and performs heterogeneous signal fusion to generate a multimodal infusion status signal. An alarm decision module, comprising a state analysis unit and an alarm rule engine. The state analysis unit receives a multimodal infusion state signal and outputs a liquid level prediction signal and an infusion abnormality risk signal. The alarm rule engine generates an alarm trigger threshold based on the multimodal infusion state signal and the liquid level prediction signal. An alarm execution module includes an audible and visual alarm unit, a remote notification unit, and a human-computer interaction interface. The audible and visual alarm unit initiates an audible and visual alarm sequence in response to an alarm trigger threshold. The remote notification unit generates a graded alarm notification based on an abnormal infusion risk signal. The human-computer interaction interface processes the liquid level prediction signal and the abnormal infusion risk signal and outputs a comprehensive risk index.

2. The infusion alarm system according to claim 1, characterized in that: The infusion status monitoring module includes an optical sensing unit and a gravity sensing unit. The optical sensing unit obtains the liquid level position change data in the infusion tube through infrared spectroscopy analysis technology and forms a liquid level fluctuation signal. The gravity sensing unit monitors the mass change data of the infusion bag in real time and forms a mass attenuation signal. The data fusion unit receives the liquid level fluctuation signal and the mass attenuation signal to achieve time domain alignment and feature fusion, and generates the multimodal infusion status signal. The multimodal infusion status signal includes a liquid level analysis signal, a flow calculation signal and an abnormality detection signal.

3. The infusion alarm system according to claim 2, characterized in that: The state analysis unit includes a liquid level prediction model and an adaptive threshold detection model based on a time convolution network. The liquid level analysis signal includes liquid level height time series data, drip rate frequency data and tube pressure fluctuation data. The flow calculation signal includes unit time flow value, cumulative infusion total volume and preset remaining time data. The abnormality detection signal includes bubble concentration data, pipeline blockage probability data and infusion component abnormality index. The liquid level height time series data, drip rate frequency data and tube pressure fluctuation data are input into the time convolution network model to output the liquid level prediction signal. The bubble concentration data, pipeline blockage probability data and infusion component abnormality index are fused through the adaptive threshold detection model to output the infusion abnormality risk signal.

4. The infusion alarm system according to claim 3, characterized in that: The alarm rule engine includes a dynamic threshold calculation submodule and a multi-level alarm strategy library. The dynamic threshold calculation submodule generates a dynamic alarm trigger threshold through a nonlinear mapping relationship based on the remaining liquid volume value in the liquid level prediction signal, the preset remaining time data and the bubble concentration threshold in the infusion abnormality risk signal. The multi-level alarm strategy library stores alarm response strategies for different infusion abnormality categories. When the infusion abnormality risk signal exceeds the preset risk level, the corresponding alarm response strategy is triggered.

5. The infusion alarm system according to claim 4, characterized in that: The alarm execution module also includes an alarm priority determination unit and a cross-device linkage unit. The alarm priority determination unit receives a comprehensive risk index and historical alarm record data, and generates a current alarm response priority through a weighted decision tree algorithm. The cross-device linkage unit activates the corresponding sound and light alarm sequence and remote notification channel according to the alarm response priority. The remote notification unit sends an encrypted alarm data packet to a preset monitoring terminal through a wireless communication protocol.

6. The infusion alarm system according to claim 5, characterized in that: The human-computer interaction interface includes a three-dimensional infusion simulator and a risk visualization unit. The three-dimensional infusion simulator renders the liquid flow state and bubble movement trajectory inside the infusion pipeline in real time. The risk visualization unit generates a dynamic risk heat map based on the comprehensive risk index and automatically retrieves microscopic imaging data of the pipeline blockage location when the abnormal infusion risk signal exceeds the high-risk threshold.

7. The infusion alarm system according to claim 6, characterized in that: The system also includes a self-learning optimization module, which includes a historical data analysis unit and a rule update engine. The historical data analysis unit continuously collects infusion completion rate data, false alarm number data and medical staff intervention records. The rule update engine dynamically adjusts the dynamic alarm trigger threshold in the alarm rule engine and the response parameters of the multi-level alarm strategy library based on the historical data analysis results.

8. The infusion alarm system according to claim 7, characterized in that: The sound and light alarm unit includes a multi-band sound generator and a parallel light source array. The multi-band sound generator switches warning sounds of different frequencies according to the alarm response priority. The parallel light source array projects a dynamic warning pattern on the surface of the infusion bag and activates a rotating warning beam when the comprehensive risk index reaches the highest level.

9. The infusion alarm system according to claim 8, characterized in that: The remote notification unit includes an encryption transmission submodule and an emergency contact polling mechanism. The encryption transmission submodule performs chaotic encryption on the infusion abnormality risk signal and transmits it via low-power Bluetooth. The emergency contact polling mechanism automatically switches to the backup contact list when no confirmation signal is received from the monitoring terminal within a preset period of time.

10. A novel infusion alarm device, using the infusion alarm system according to any one of claims 1 to 9.

Images