Data-driven methods for predicting medical equipment failures and intelligent maintenance platforms

By establishing a control output baseline and evaluating the compensation degradation index, combined with the similarity of historical pre-fault trajectories, the problem of failing to identify latent degradation of medical devices in existing technologies has been solved, achieving highly sensitive fault warning and precise maintenance.

CN123069205APending Publication Date: 2026-11-10JIANGSU BEIZHEN MEDICAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611291849.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for predicting medical device failures fail to effectively utilize the latent degradation information of closed-loop controllers, resulting in the masking of mild degradation and making it difficult to identify early signs of failure in a timely manner.

Method used

By acquiring operational data from medical equipment, a control output baseline is established, the compensation degradation index is assessed, and similarity matching is performed with historical pre-failure trajectories to achieve fault prediction and maintenance operations.

Benefits of technology

It enables precise quantification of the physical compensation energy and control cost within closed-loop control, improving the foresight and reliability of diagnosis, and transforming it into precise pre-treatment based on conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069205A_ABST
    Figure CN123069205A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of medical equipment monitoring technology. It discloses a data-driven method for predicting medical equipment faults and an intelligent maintenance platform. The method includes: acquiring raw control data streams; extracting health status data segments to establish a control output baseline, and obtaining baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline under health status; generating a compensation degradation index based on the deviation between the current control compensation energy and the baseline compensation energy parameter, and further evolving it temporally to generate the current health status trajectory; extracting historical pre-fault trajectories along a time anchor point to construct a component degradation trajectory template, and performing similarity matching with the current health status trajectory to obtain the component's trajectory similarity; predicting component faults based on the compensation degradation index and trajectory similarity, and triggering corresponding maintenance operations based on the fault prediction results, thereby improving the foresight and reliability of the diagnosis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical equipment monitoring technology, and more specifically, to a data-driven method for predicting medical equipment failures and an intelligent maintenance platform. Background Technology

[0002] With the rapid development of industrial control and information technology, modern medical equipment such as ventilators, infusion pumps, and hemodialysis machines generally adopt closed-loop control systems, which adjust the controller output through real-time feedback to ensure that the equipment maintains high-precision and stable operation under complex working conditions.

[0003] In existing technologies, there are generally two methods for predicting the failure of medical equipment: one is to judge the failure based on the output variable exceeding the limit, that is, to trigger a failure warning when the feedback signal exceeds the preset boundary; the other is to directly train the model based on the original operating data and classify the failure or predict the remaining life by learning the feature differences between historical normal data and failure data. However, neither of the above methods considers the additional compensation behavior of the closed-loop controller in order to maintain the set target value, thus ignoring the implicit degradation information contained in the controller output signal, which leads to the serious masking of mild degradation and makes it difficult to identify the early signs of failure in time. For example, a ventilator adjusts the blower speed and proportional valve opening through the controller to make the actual airway pressure and inspiratory flow close to the set ventilation target. When the blower bearing wears, the impeller accumulates dust, or the airway resistance increases, the ventilator can still maintain the set pressure and flow, but the controller needs to increase the blower drive duty cycle or increase the proportional valve opening compensation amount, that is, there is an implicit degradation phenomenon where the output indicators appear to be completely normal, but the control compensation cost continues to increase.

[0004] In view of this, the present invention proposes a data-driven method for predicting medical equipment failures and an intelligent maintenance platform to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a data-driven method for predicting medical equipment failures and an intelligent maintenance platform, comprising: S1. Acquire the operational data of the medical equipment during control execution, and record it in time sequence to form the raw control data stream; S2. Extract health status data segments from the original control data stream to establish the control output baseline, and obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline under the health status. S3. Evaluate the response deviation between the current controller output signal and the control output baseline. Based on the deviation between the current control compensation energy and the baseline compensation energy parameter, generate a compensation degradation index and further evolve it over time to generate the current health status trajectory. S4. Extract the historical trajectory before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component. S5. Based on the compensation degradation index and trajectory similarity, perform fault prediction on the component and trigger corresponding maintenance operations according to the fault prediction results.

[0006] Furthermore, methods for forming the original control data flow include: Collect operational data when the medical equipment is in operation and executing control. The operational data includes controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records. The maintenance record time, fault confirmation time, and component replacement record in the historical maintenance records are associated with the corresponding component operating time; the controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records for each group are arranged in timestamp order to form the original control data stream.

[0007] Furthermore, methods for establishing a control output baseline include: The time interval with no fault confirmation time and no component replacement record is selected from the historical maintenance records, and the corresponding operation record is extracted from the original control data stream based on the time interval, and all of them are marked as health status data segments; the health status data segments are divided into several control condition groups based on the same equipment operation mode and the same set target value. The controller output signals in each control condition group are subjected to quantile regression processing. Under the condition of setting the target value and the feedback signal as input variables, the regression coefficients and intercept terms of the controller output signals at the lower, middle and upper quantiles are obtained. The regression coefficients and intercept terms at the midpoints are used to analyze the corresponding target values ​​and feedback signals to obtain the control output baselines for each control condition group.

[0008] Furthermore, methods for obtaining baseline control compensation energy include: The regression coefficients and intercept terms corresponding to the upper and lower quantiles are analyzed to obtain the upper and lower boundary values ​​of the control output for each control condition group. The controller output signal in the health status data segment is compared with the corresponding upper and lower boundary values ​​of the control output. If it is between the upper and lower boundary values ​​of the control output, the corresponding running record is marked as a normal fluctuation record; otherwise, it is marked as an abnormal fluctuation record. Calculate the deviation of the controller output signal marked as a normal fluctuation record relative to the control output baseline, and select the maximum absolute value of the deviation within each control condition group as the upper limit of the control output deviation under the corresponding equipment operating mode; The absolute value of the control error signal and the corresponding absolute value of the controller output deviation, which are coupled and marked as normal fluctuation records, are superimposed over the complete time interval of the health state data segment to obtain the cumulative control compensation energy value. The cumulative control compensation energy values ​​within each control condition group are summarized, and the right tail data exceeding the set threshold value are subjected to distribution fitting. The values ​​corresponding to the high confidence quantiles in the fitting results are selected as the baseline control compensation energy of the corresponding control condition group.

[0009] Furthermore, methods for generating the compensatory degradation index include: Based on the current equipment operating mode and the set target value, the corresponding control output baseline is selected, the deviation from the current controller output signal is evaluated, and the instantaneous value of the current control compensation power is obtained by combining the absolute value of the current control error signal. Starting from the start timestamp recorded in the current operation phase, the instantaneous values ​​of the current control compensation power at each moment are accumulated to obtain the current control compensation energy; The deviation between the current control compensation energy value and the baseline control compensation energy is normalized by controlling the upper limit of the output deviation, thus forming the current compensation degradation index.

[0010] Furthermore, methods for generating the current health state trajectory through further temporal evolution include: Within a fixed time window, the current compensation degradation index, control error signal, and component running time of the component are processed on a unified scale and combined to form the current health status value of the component. Arrange multiple consecutive current health status values ​​in chronological order, and the temporal evolution forms the current health status trajectory.

[0011] Furthermore, methods for obtaining the trajectory similarity of components include: The fault confirmation time of the corresponding component is selected from the historical maintenance records and marked as the time anchor point; with the time anchor point as the end point, the records corresponding to the timestamps that continuously exceed the natural fluctuation range of the healthy state are traced back in reverse order from late to early, and the timestamps of the records are used as the starting point of degradation. Extract the sequence of health status values ​​between the degradation start point and the end point to form the historical pre-fault trajectory; perform unified time window processing on the historical pre-fault trajectory to form the historical degradation trajectory template of the component; The current health status trajectory is time-series registered with the corresponding historical degradation trajectory template. The trajectory distance between the two is analyzed and normalized to obtain the trajectory similarity between the current health status trajectory of the component and the historical degradation trajectory template.

[0012] Furthermore, methods for predicting component failures based on the compensation degradation index and trajectory similarity include: The top K historical pre-fault trajectories are selected in descending order of trajectory similarity as a candidate set of similar trajectories; the time span between the fault confirmation time and the degradation start point of each historical pre-fault trajectory in the candidate set of similar trajectories is taken as the total degradation duration. Using the trajectory similarity of each historical trajectory before the failure as the weight, the total degradation duration in the candidate set of similar trajectories is weighted to obtain the predicted remaining operating life; starting from the current timestamp, the time length corresponding to the predicted remaining operating life is pushed forward, and the timestamp reached by the push is used as the predicted failure occurrence timestamp. The component corresponding to the historical trajectory before the fault with the maximum trajectory similarity is combined with the predicted fault occurrence timestamp to form the fault prediction result.

[0013] Furthermore, methods for triggering corresponding maintenance operations based on fault prediction results include: The time interval between the current timestamp and the predicted fault occurrence timestamp is used as the predicted remaining maintenance window duration and compared with a preset maintenance threshold; where the maintenance threshold includes an emergency maintenance threshold and a regular maintenance threshold; If the predicted remaining maintenance window duration does not exceed the emergency maintenance threshold, the medical equipment will be stopped and an emergency maintenance warning message will be displayed; if it is between the emergency maintenance threshold and the regular maintenance threshold, the medical equipment will continue to operate and a planned maintenance prompt message will be displayed; if it exceeds the regular maintenance threshold, the medical equipment will continue to operate and a normal operating status message will be displayed.

[0014] Furthermore, the data-driven intelligent maintenance platform for predicting medical equipment failures includes: The runtime sequence acquisition module is used to generate the raw control data stream; The baseline parameter construction module is used to establish a control output baseline during the period when the equipment is in a healthy state, and to obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline in the healthy state. The health trajectory analysis module is used to generate the current health status trajectory based on the deviation between the current control compensation energy and the baseline compensation energy parameters, and further based on the temporal evolution of the compensation degradation index. The trajectory template registration module is used to extract historical trajectories before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component. The fault prediction and control module is used to predict component faults based on the compensation degradation index and trajectory similarity, and trigger corresponding maintenance operations based on the fault prediction results.

[0015] The technical effects and advantages of the data-driven medical equipment fault prediction method and intelligent maintenance platform of this invention are as follows: 1. This invention extracts health status data segments to divide control condition groups, applies quantile regression to construct a control output baseline, and combines upper and lower boundary deviations for anomaly filtering, thereby achieving precise quantification of the physical compensation energy and control cost within the closed-loop control. Furthermore, it integrates the control error and controller output deviation over time within the health time interval to calibrate the baseline control compensation energy, eliminating the interference of sudden changes in operating conditions on degradation evaluation. This enables highly sensitive degradation early warning in the early stages of latent degradation before obvious faults occur in equipment or components, significantly improving the foresight and reliability of the diagnosis. 2. This invention uses the fault confirmation time in historical maintenance records as the time anchor point, traces back the degradation index in reverse order according to the timestamp, determines the degradation starting point, generates a historical degradation trajectory template by extracting the health status value within the degradation period, and performs time axis registration and normalized distance calculation between the health status trajectory and the template. This achieves nonlinear alignment and high morphological matching quantification of the entire historical degradation evolution process under different maintenance cycles, providing a highly confident benchmark for subsequent remaining service life prediction and fault timestamp estimation. In turn, the entire process is linked to form a diagnostic closed loop, realizing the transformation from passive maintenance to precise pre-maintenance based on condition. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the principle of the data-driven medical device fault prediction method of the present invention. Figure 2 This is a system schematic diagram of the data-driven intelligent maintenance platform for predicting and maintaining medical equipment faults according to the present invention. Figure 3 This is a flowchart illustrating how the baseline control compensation energy is obtained in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 and Figure 3 As shown, the data-driven medical device fault prediction method described in this embodiment includes: S1. Acquire the operational data of the medical equipment during control execution and record it in time series to form the raw control data stream.

[0019] In the closed-loop control operation of medical equipment such as ventilators, the controller continuously sends drive inputs to actuators such as blower speed and proportional valve opening to maintain target values. Even when the actuators experience slight degradation such as bearing wear, impeller dust accumulation, or increased air resistance, the feedback signal can still remain near the set target value, only manifesting as a hidden degradation phenomenon of continuously increasing control compensation costs. To obtain operational data that reflects this hidden degradation phenomenon of continuously increasing control compensation costs, data is synchronously collected for medical equipment that is in operation and performing control.

[0020] Methods for forming raw control data streams include: The system collects operational data when the medical equipment is in operation and under control. This operational data includes controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records.

[0021] For example, for a ventilator, the controller output signals include the blower drive duty cycle and proportional valve opening; the set target values ​​include the set airway pressure and the set inspiratory flow rate; the feedback signals include the actual airway pressure, the actual inspiratory flow rate, and the blower speed; the control error signals include the pressure deviation between the set airway pressure and the actual airway pressure; and the device operating mode includes the currently active specific control strategy type, such as pressure-controlled ventilation mode and volume-controlled ventilation mode.

[0022] For example, the historical maintenance record states that "the blower bearing was abnormal, the fault was confirmed at 10:00 on March 20, 2025, and the blower replacement was completed at 12:00 on March 20, 2025".

[0023] The maintenance record time, fault confirmation time, and component replacement record in the historical maintenance records are associated with the corresponding component operating time; the controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records for each group are arranged in timestamp order to form the original control data stream.

[0024] S2. Extract health status data segments from the original control data stream to establish the control output baseline, and obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline under the health status.

[0025] Methods for establishing a control output baseline include: Each historical maintenance record is parsed to identify and filter timestamps with no fault confirmation and no component replacement records; the filtered timestamps are then divided into continuous, uninterrupted time intervals. Based on this time interval, corresponding operation records are extracted from the original control data stream. These operation records include controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, maintenance record times, fault confirmation times, and component replacement records. Each operation record extracted from each time interval is marked as a health status data segment.

[0026] Based on the same equipment operating mode and the same set target value, health status data segments are divided into several control condition groups. All operating records within each control condition group have the same equipment operating mode and the same set target value; and each control condition group is uniquely identified by a combination of the equipment operating mode code and the set target value; each control condition group is used for subsequent processing of the controller output signals within that control condition group to establish the corresponding control output baseline.

[0027] The controller output signals in each control condition group are subjected to quantile regression processing. Under the condition of setting the target value and the feedback signal as input variables, the regression coefficients and intercept terms of the controller output signals at the lower, middle and upper quantiles are obtained.

[0028] The specific steps are as follows: traverse each control condition group, extract the set target value, feedback signal, and controller output signal from the health status data segment within the control condition group; use the set target value and feedback signal as independent variables and the controller output signal as dependent variable to construct a quantile regression objective function; In the objective function of quantile regression, the weight parameters of the asymmetric loss function are set as the proportion of the lower quantile (e.g., 0.05), the proportion of the middle quantile (0.50), and the proportion of the upper quantile (0.95), respectively, corresponding to the three representative conditional quantiles of the lower, middle, and upper ends.

[0029] The linear programming simplex algorithm is used to solve for the minimum values ​​of the objective functions corresponding to the proportions at the lower, middle, and upper quantiles, respectively; the regression coefficients and intercepts at the lower, middle, and upper quantiles are obtained.

[0030] The regression coefficients and intercept terms at the median quantile are used to analyze the corresponding target values ​​and feedback signals to obtain the control output baseline for each control condition group. Specifically, the regression coefficients and intercept terms at the median quantile for each control condition group are read; wherein, the regression coefficients at the median quantile include the median quantile regression coefficients corresponding to the target values ​​and the median quantile regression coefficients corresponding to the feedback signals. Read the set target value and feedback signal from each running record in the control condition group; multiply the set target value by the median regression coefficient corresponding to the set target value to obtain the first product value; multiply the feedback signal in each running record by the median regression coefficient corresponding to the feedback signal to obtain the second product value; add the first product value and the second product value and then add them to the median intercept term. The sum is used as the control output baseline of the running record in the control condition group. Repeat the operation on each running record in the control condition group to obtain the control output baseline value corresponding to each running record; the control output baseline values ​​corresponding to all running records in all control condition groups constitute the control output baseline of each control condition group.

[0031] Methods for obtaining baseline control compensation energy include: The regression coefficients and intercept terms corresponding to the upper and lower quantiles are analyzed to obtain the upper and lower boundary values ​​of the control output for each control condition group. The specific steps are as follows: Read the setpoint value and feedback signal from each operating record within each control condition group; substitute the setpoint value and feedback signal into the linear relationship determined by the regression coefficient and intercept term at the lower quantile, respectively, to calculate the lower boundary value of the control output for that operating record; substitute the setpoint value and feedback signal values ​​into the linear relationship determined by the regression coefficient and intercept term at the upper quantile, respectively, to obtain the upper boundary value of the control output for that operating record; repeat this process to obtain the lower and upper boundary values ​​of the control output for each operating record.

[0032] Among them, the upper and lower boundary values ​​of the control output refer to the lower and upper limits of the normal fluctuation range of the controller output signal. Together, they constitute the normal fluctuation boundary of the controller output signal value under healthy conditions.

[0033] The controller output signal in the health status data segment is compared with the corresponding upper and lower boundary values ​​of the control output. If it is between the upper and lower boundary values ​​of the control output, the corresponding operation record is marked as a normal fluctuation record; otherwise, it is marked as an abnormal fluctuation record. Iterate through each operating record marked as a normal fluctuation record in the health status data segment, read the controller output signal and control output baseline in each normal fluctuation record, calculate the deviation of the controller output signal from the control output baseline (i.e., the controller output signal minus the control output baseline corresponding to that record), and take the absolute value of the difference to form the absolute deviation value. Within each control condition group, select the maximum absolute deviation value as the upper limit of the control output deviation for the corresponding equipment operating mode. The upper limit of the control output deviation is used to characterize the maximum allowable deviation of the controller output signal from the control output baseline under healthy conditions.

[0034] Extract the absolute value of the control error signal and the corresponding absolute value of the controller output deviation from the records marked as normal fluctuations. Multiply the absolute value of the control error signal in each normal fluctuation record by the corresponding absolute value of the controller output deviation to obtain the instantaneous value of the control compensation power for that normal fluctuation record. Starting from the beginning timestamp of the health status data segment, traverse the data one record at a time in ascending order of timestamps, performing the accumulation operation only on the records marked as normal fluctuations, until the entire time interval has been traversed to obtain the accumulated control compensation energy value.

[0035] Among them, the absolute value of the control error signal reflects the deviation between the set target value and the feedback signal, and the controller output deviation value reflects the output force adjusted by the controller to eliminate the deviation. The cumulative sum of the two over time is the total compensation cost paid by the controller to maintain the set target value.

[0036] Summarize the cumulative control compensation energy values ​​in each control condition group to make each control condition group correspond to a set of cumulative control compensation energy value arrays; arrange all values ​​in the cumulative control compensation energy value arrays corresponding to the control condition group in ascending order; extract all values ​​that are above the set threshold value after arrangement as the right tail data of the control condition group. The right-tailed data are fitted with a generalized Pareto distribution to obtain the fitting parameters of the generalized Pareto distribution. The values ​​corresponding to the high confidence quantiles (e.g., 0.99) in the fitting results are selected as the baseline control compensation energy for the corresponding control case group.

[0037] The specific percentile position of the threshold value is determined based on the sample size and data distribution of the cumulative control compensation energy value array; the setting of the high confidence quantile is determined based on the safety requirements of the medical device, that is, the higher the safety requirements of the medical device, the higher the high confidence level value, making the device more sensitive to latent degradation.

[0038] The baseline control compensation energy represents the maximum reasonable value that the accumulated control compensation energy can achieve when the controller output signal is within the healthy fluctuation range. The baseline control compensation energy can serve as a reference benchmark for determining whether the current control compensation energy deviates abnormally from the healthy level: if the current control compensation energy is lower than or close to the baseline control compensation energy, it indicates that the compensation cost currently borne by the medical device is still within the healthy benchmark range; if the current control compensation energy is consistently higher than the baseline control compensation energy, it indicates that the medical device has incurred additional compensation costs exceeding the healthy benchmark to maintain the set target value, i.e., there is implicit degradation.

[0039] S3. Evaluate the response deviation between the current controller output signal and the control output baseline. Based on the deviation between the current control compensation energy and the baseline compensation energy parameters, generate a compensation degradation index and further evolve it over time to generate the current health status trajectory.

[0040] Methods for generating the compensatory degradation index include: Read the device operating mode and set target value corresponding to the current operating record from the original control data stream to determine the control condition group to which the current operating record belongs; based on the current device operating mode and set target value, select the corresponding control output baseline, evaluate the deviation from the current controller output signal, and combine the absolute value of the current control error signal to obtain the instantaneous value of the current control compensation power.

[0041] The instantaneous value of the current control compensation power is used to characterize the instantaneous compensation force exerted by the medical device controller at the current moment to eliminate the deviation between the set target value and the feedback signal.

[0042] Starting from the start timestamp recorded in the current operation phase, the instantaneous values ​​of the current control compensation power at each moment are accumulated to obtain the current control compensation energy.

[0043] Current control compensation energy refers to the total amount of compensation cost that the medical device controller continuously pays to maintain the set target value from the moment of power-on or startup until the moment of current operation during the current operating phase.

[0044] The deviation between the current control compensation energy value and the baseline control compensation energy is normalized by controlling the upper limit of the output deviation, forming the current compensation degradation index. The control compensation energy difference is obtained by subtracting the baseline control compensation energy from the current control compensation energy value; the compensation energy deviation rate is obtained by dividing the control compensation energy difference by the baseline control compensation energy; the upper limit of the control output deviation corresponding to the current running record is read, and the compensation energy deviation rate is divided by the upper limit of the control output deviation to obtain the normalized compensation energy deviation rate; the normalized compensation energy deviation rate is mapped to the 0-1 interval using a sigmoid function to obtain the compensation degradation index corresponding to the current running record.

[0045] Further methods for generating the current health status trajectory through temporal evolution include: Within a fixed time window (e.g., 10 seconds, determined based on the control cycle and response speed of the medical device), the current compensation degradation index, control error signal, and component runtime are processed using a unified scale. Specifically, the maximum and minimum values ​​of the compensation degradation index, control error signal, and component runtime are extracted from all runtime records within the fixed time window. The compensation degradation index is then subtracted from the minimum compensation degradation index within the window and divided by the difference between the maximum and minimum values ​​within that window to obtain the standardized compensation degradation index. Similarly, the standardized control error signal and standardized component runtime are obtained. These three standardized data points are then weighted to obtain the current health status value.

[0046] The standardized degradation compensation index directly quantifies the extra compensation cost incurred by the controller to maintain the set target value, and is the most direct quantitative representation of implicit degradation; therefore, its weight coefficient is set to the highest. The control error signal reflects the instantaneous deviation between the set target value and the actual feedback signal at the current moment. When the controller is working normally, this deviation is usually controlled within a small range, and its indicative role in the degree of degradation is second only to the degradation compensation index; therefore, the control error signal is given the second highest weight coefficient. Component runtime reflects the cumulative usage time of the equipment, but the degree of degradation of different components may vary significantly under the same cumulative runtime; therefore, the component runtime is given the lowest weight coefficient. For example, the weight coefficient for the degradation compensation index is set to 0.60, the weight coefficient for the control error signal is 0.25, and the weight coefficient for the component runtime is 0.15.

[0047] The system moves backward with a fixed sliding step size to form multiple current health status values. These multiple current health status values ​​are then arranged in chronological order, and the temporal evolution forms the current health status trajectory.

[0048] S4. Extract the historical trajectory before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component.

[0049] Methods for obtaining the trajectory similarity of components include: The fault confirmation time of the corresponding component is selected from the historical maintenance records and marked as the time anchor point. Using the time anchor point as the end point, the process is reversed in the order of the timestamp from late to early, and the timestamp corresponding to the record where the compensation degradation index continuously exceeds the natural fluctuation range of the healthy state is taken as the degradation starting point.

[0050] The natural fluctuation range of health status refers to the range of values ​​in which the compensatory degradation index fluctuates around a certain central level under normal conditions when a medical device is in a healthy state and operating in a specific operating mode.

[0051] The upper and lower boundaries of the natural fluctuation range of the healthy state are determined by the upper and lower boundary values ​​of the control output: that is, the normalized compensation energy deviation rate, obtained by using the upper limit of the control output deviation as the normalized denominator, is constrained to a finite numerical interval centered at zero. This finite numerical interval, after being mapped by the S-shaped function, falls within the approximately linear response region of the S-shaped function. The upper and lower boundaries of this linear response region correspond to the upper and lower boundaries of the natural fluctuation range of the compensation degradation index under the healthy state. If the compensation degradation index continuously exceeds the upper boundary of this natural fluctuation range, it indicates that the deviation of the controller output signal from the control output baseline has exceeded the normal random fluctuation range under the healthy state.

[0052] Extract the sequence of health status values ​​between the degradation start point and the end point to form the historical pre-fault trajectory; perform unified time window processing on the historical pre-fault trajectory to form the historical degradation trajectory template of the component; The current health status trajectory is temporally registered with the corresponding historical degradation trajectory template using a dynamic time warping algorithm to construct a cumulative distance matrix. The rows of the cumulative distance matrix correspond to each current health status value in the current health trajectory, and the columns of the cumulative distance matrix correspond to each historical health status value in the historical degradation trajectory template. The element value of each position in the cumulative distance matrix is ​​the minimum value among the absolute difference between the current health status value and the historical health status value corresponding to that position, plus the element values ​​of the positions to the left, below, and to the lower left of that position.

[0053] Starting from the bottom right corner of the cumulative distance matrix, backtracking towards the top left corner, each corresponding position along the backtracking path is recorded as the minimum curved path; the absolute difference between the current health status value and the historical health status value corresponding to each corresponding position on the minimum curved path is accumulated to obtain the trajectory distance; the total number of steps contained in the minimum curved path is extracted, the trajectory distance is divided by the total number of steps, and the normalized trajectory distance is converted into trajectory similarity in the interval of 0 to 1 through exponential decay operation. S5. Based on the compensation degradation index and trajectory similarity, perform fault prediction on the component and trigger corresponding maintenance operations according to the fault prediction results.

[0054] Methods for component failure prediction based on compensation degradation index and trajectory similarity include: The top K historical pre-fault trajectories (K value can be determined based on the total number of historical fault confirmation events corresponding to the target component in the historical maintenance records) are selected in descending order of trajectory similarity as a candidate set of similar trajectories; the time span between the fault confirmation time and the degradation start point of each historical pre-fault trajectory in the candidate set of similar trajectories is taken as the total degradation duration.

[0055] The total degradation duration represents the time it takes for a component to go from clearly entering a degradation state to eventually failing in historical failure cases.

[0056] Using the trajectory similarity of each historical pre-failure trajectory as weight, the total degradation duration in the candidate set of similar trajectories is weighted to obtain the predicted remaining operational lifespan. Specifically: the total degradation duration of each historical pre-failure trajectory is multiplied by the trajectory similarity corresponding to that historical pre-failure trajectory to obtain the weighted total degradation duration of each historical pre-failure trajectory; the weighted total degradation durations of all K historical pre-failure trajectories are summed to obtain the weighted total degradation duration sum; the trajectory similarity of all K historical pre-failure trajectories in the candidate set of similar trajectories is summed to obtain the trajectory similarity sum; the weighted total degradation duration sum is divided by the trajectory similarity sum, and the result is used as the predicted remaining operational lifespan.

[0057] Starting from the current timestamp, the time is shifted forward according to the predicted remaining operating life, and the timestamp reached is used as the predicted fault occurrence timestamp.

[0058] If the medical device is still running, the current absolute time value of the device's internal real-time clock is used as the current timestamp; if the medical device has been shut down, the timestamp of the last running record is used as the current timestamp.

[0059] The component corresponding to the historical trajectory before the fault with the maximum trajectory similarity is combined with the predicted fault occurrence timestamp to form the fault prediction result.

[0060] Methods for triggering corresponding maintenance operations based on fault prediction results include: The time interval between the current timestamp and the predicted fault occurrence timestamp is used as the predicted remaining maintenance window duration, and compared with a preset maintenance threshold. This maintenance threshold includes both emergency maintenance and routine maintenance thresholds, with the routine maintenance threshold being greater than the emergency maintenance threshold.

[0061] The emergency maintenance threshold refers to the lowest acceptable boundary value for the predicted remaining maintenance window duration; the routine maintenance threshold refers to the planned maintenance initiation boundary value for the predicted remaining maintenance window duration. Both the emergency and routine maintenance thresholds are dynamically set based on the medical device's safety level, the degradation rate of its failure modes, and its maintenance response capabilities.

[0062] If the predicted remaining maintenance window duration does not exceed the emergency maintenance threshold, it indicates that the medical equipment is very close to the predicted failure time, and the remaining operable time is insufficient to complete planned maintenance preparations or scheduling (such as spare parts procurement, maintenance personnel arrival, equipment usage schedule coordination, etc.). If operation continues, the equipment may experience a sudden failure in a very short time, leading to treatment interruption or patient safety risks; therefore, the medical equipment operation will be stopped and an emergency maintenance warning message will be displayed.

[0063] If the predicted remaining maintenance window duration is between the emergency maintenance threshold and the routine maintenance threshold, it indicates that the medical equipment still has a certain time margin before the predicted failure time. This time margin is sufficient to complete the preparatory work for planned maintenance (including confirming spare parts inventory, arranging maintenance personnel shifts, and coordinating equipment usage time); then the medical equipment will continue to operate and a planned maintenance prompt message will be displayed.

[0064] If the predicted remaining maintenance window duration exceeds the regular maintenance threshold, it indicates that the medical device has sufficient time leeway before the predicted failure time, and there is currently no maintenance requirement that requires immediate response or is planned in the near future. The device can continue to operate normally, so the medical device will continue to operate and display normal operating status information.

[0065] In this embodiment, by extracting health status data segments to divide control condition groups, applying quantile regression to construct the control output baseline, and combining upper and lower boundary deviations for anomaly filtering, the precise quantification of the physical compensation energy and control cost within the closed-loop control is achieved. Within the health time interval, the control error and controller output deviation are integrated over time to calibrate the baseline control compensation energy, eliminating the interference of sudden changes in operating conditions on degradation evaluation. This enables highly sensitive degradation early warning in the early stages of latent degradation before obvious faults occur in equipment or components, significantly improving the foresight and reliability of the diagnosis.

[0066] By using the fault confirmation time in historical maintenance records as the time anchor, and tracing back the degradation index line by line according to the timestamp, the degradation starting point is determined. By extracting the health status values ​​within the degradation period, a historical degradation trajectory template is generated. The health status trajectory and the template are then registered on the time axis and the normalized distance is calculated. This achieves nonlinear alignment and high morphological matching quantification of the entire historical degradation evolution process under different maintenance cycles. This provides a very high confidence benchmark for subsequent remaining service life prediction and fault timestamp estimation. In this way, the entire process is linked to form a diagnostic closed loop, realizing the transformation from passive maintenance to precise pre-maintenance based on condition.

[0067] Example 2, please refer to Figure 2 As shown in this embodiment, the data-driven intelligent maintenance platform for predicting medical equipment failures includes: The runtime sequence acquisition module is used to generate the raw control data stream; The baseline parameter construction module is used to establish a control output baseline during the period when the equipment is in a healthy state, and to obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline in the healthy state. The health trajectory analysis module is used to generate the current health status trajectory based on the deviation between the current control compensation energy and the baseline compensation energy parameters, and further based on the temporal evolution of the compensation degradation index. The trajectory template registration module is used to extract historical trajectories before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component. The fault prediction and control module is used to predict component faults based on the compensation degradation index and trajectory similarity, and trigger corresponding maintenance operations based on the fault prediction results.

[0068] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0069] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0070] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

[0071] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data-driven method for predicting medical device failures, characterized in that, The method includes: S1. Acquire the operational data of the medical equipment during control execution, and record it in time sequence to form the raw control data stream; S2. Extract health status data segments from the original control data stream to establish the control output baseline, and obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline under the health status. S3. Evaluate the response deviation between the current controller output signal and the control output baseline. Based on the deviation between the current control compensation energy and the baseline compensation energy parameter, generate a compensation degradation index and further evolve it over time to generate the current health status trajectory. S4. Extract the historical trajectory before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component. S5. Based on the compensation degradation index and trajectory similarity, perform fault prediction on the component and trigger corresponding maintenance operations according to the fault prediction results.

2. The data-driven medical device fault prediction method according to claim 1, characterized in that, The method for forming the raw control data stream includes: Collect operational data when the medical equipment is in operation and executing control. The operational data includes controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records. The maintenance record time, fault confirmation time, and component replacement record in the historical maintenance records are associated with the corresponding component operating time; the controller output signals, set target values, feedback signals, control error signals, equipment operating modes, component operating times, and historical maintenance records for each group are arranged in timestamp order to form the original control data stream.

3. The data-driven medical device fault prediction method according to claim 2, characterized in that, The method for establishing the control output baseline includes: The time interval with no fault confirmation time and no component replacement record is selected from the historical maintenance records, and the corresponding operation record is extracted from the original control data stream based on the time interval, and all of them are marked as health status data segments; the health status data segments are divided into several control condition groups based on the same equipment operation mode and the same set target value. The controller output signals in each control condition group are subjected to quantile regression processing. Under the condition of setting the target value and the feedback signal as input variables, the regression coefficients and intercept terms of the controller output signals at the lower, middle and upper quantiles are obtained. The regression coefficients and intercept terms at the midpoints are used to analyze the corresponding target values ​​and feedback signals to obtain the control output baselines for each control condition group.

4. The data-driven medical device fault prediction method according to claim 3, characterized in that, The method for obtaining baseline control compensation energy includes: The regression coefficients and intercept terms corresponding to the upper and lower quantiles are analyzed to obtain the upper and lower boundary values ​​of the control output for each control condition group. The controller output signal in the health status data segment is compared with the corresponding upper and lower boundary values ​​of the control output. If it is between the upper and lower boundary values ​​of the control output, the corresponding running record is marked as a normal fluctuation record; otherwise, it is marked as an abnormal fluctuation record. Calculate the deviation of the controller output signal marked as a normal fluctuation record relative to the control output baseline, and select the maximum absolute value of the deviation within each control condition group as the upper limit of the control output deviation under the corresponding equipment operating mode; The absolute value of the control error signal and the corresponding absolute value of the controller output deviation, which are coupled and marked as normal fluctuation records, are superimposed over the complete time interval of the health state data segment to obtain the cumulative control compensation energy value. The cumulative control compensation energy values ​​within each control condition group are summarized, and the right tail data exceeding the set threshold value are subjected to distribution fitting. The values ​​corresponding to the high confidence quantiles in the fitting results are selected as the baseline control compensation energy of the corresponding control condition group.

5. The data-driven medical device fault prediction method according to claim 4, characterized in that, The method for generating the compensatory degradation index includes: Based on the current equipment operating mode and the set target value, the corresponding control output baseline is selected, the deviation from the current controller output signal is evaluated, and the instantaneous value of the current control compensation power is obtained by combining the absolute value of the current control error signal. Starting from the start timestamp recorded in the current operation phase, the instantaneous values ​​of the current control compensation power at each moment are accumulated to obtain the current control compensation energy; The deviation between the current control compensation energy value and the baseline control compensation energy is normalized by controlling the upper limit of the output deviation, thus forming the current compensation degradation index.

6. The data-driven medical device fault prediction method according to claim 5, characterized in that, The method for further temporal evolution to generate the current health status trajectory includes: Within a fixed time window, the current compensation degradation index, control error signal, and component running time of the component are processed on a unified scale and combined to form the current health status value of the component. Arrange multiple consecutive current health status values ​​in chronological order, and the temporal evolution forms the current health status trajectory.

7. The data-driven medical device fault prediction method according to claim 6, characterized in that, The method for obtaining the trajectory similarity of the components includes: The fault confirmation time of the corresponding component is selected from the historical maintenance records and marked as the time anchor point; with the time anchor point as the end point, the records corresponding to the timestamps that continuously exceed the natural fluctuation range of the healthy state are traced back in reverse order from late to early, and the timestamps of the records are used as the starting point of degradation. Extract the sequence of health status values ​​between the degradation start point and the end point to form the historical pre-fault trajectory; perform unified time window processing on the historical pre-fault trajectory to form the historical degradation trajectory template of the component; The current health status trajectory is time-series registered with the corresponding historical degradation trajectory template. The trajectory distance between the two is analyzed and normalized to obtain the trajectory similarity between the current health status trajectory of the component and the historical degradation trajectory template.

8. The data-driven medical device fault prediction method according to claim 7, characterized in that, The method for predicting component failures based on compensation degradation index and trajectory similarity includes: The top K historical pre-fault trajectories are selected in descending order of trajectory similarity as a candidate set of similar trajectories; the time span between the fault confirmation time and the degradation start point of each historical pre-fault trajectory in the candidate set of similar trajectories is taken as the total degradation duration. Using the trajectory similarity of each historical trajectory before the failure as the weight, the total degradation duration in the candidate set of similar trajectories is weighted to obtain the predicted remaining operating life; starting from the current timestamp, the time length corresponding to the predicted remaining operating life is pushed forward, and the timestamp reached by the push is used as the predicted failure occurrence timestamp. The component corresponding to the historical trajectory before the fault with the maximum trajectory similarity is combined with the predicted fault occurrence timestamp to form the fault prediction result.

9. The data-driven medical device fault prediction method according to claim 8, characterized in that, The method for triggering corresponding maintenance operations based on fault prediction results includes: The time interval between the current timestamp and the predicted fault occurrence timestamp is used as the predicted remaining maintenance window duration and compared with a preset maintenance threshold; where the maintenance threshold includes an emergency maintenance threshold and a regular maintenance threshold; If the predicted remaining maintenance window duration does not exceed the emergency maintenance threshold, the medical equipment will be stopped and an emergency maintenance warning message will be displayed; if it is between the emergency maintenance threshold and the regular maintenance threshold, the medical equipment will continue to operate and a planned maintenance prompt message will be displayed; if it exceeds the regular maintenance threshold, the medical equipment will continue to operate and a normal operating status message will be displayed.

10. A data-driven intelligent maintenance platform for predicting medical equipment failures, used to implement the data-driven medical equipment failure prediction method according to any one of claims 1 to 9, characterized in that, The term includes: The runtime sequence acquisition module is used to generate the raw control data stream; The baseline parameter construction module is used to establish a control output baseline during the period when the equipment is in a healthy state, and to obtain the baseline control compensation energy based on the fluctuation range of the controller output signal relative to the control output baseline in the healthy state. The health trajectory analysis module is used to generate the current health status trajectory based on the deviation between the current control compensation energy and the baseline compensation energy parameters, and further based on the temporal evolution of the compensation degradation index. The trajectory template registration module is used to extract historical trajectories before the failure along the time anchor point to construct the degradation trajectory template of the component, and perform similarity matching with the current health status trajectory to obtain the trajectory similarity of the component. The fault prediction and control module is used to predict component faults based on the compensation degradation index and trajectory similarity, and trigger corresponding maintenance operations based on the fault prediction results.