A method, device, and medium for fault identification of an analytical analyte sensor.

By acquiring historical signal contour data of the analyte sensor, extracting morphological features, and utilizing a fault identification model, the problem of difficulty in efficiently identifying fault modes of the analyte sensor in existing technologies is solved, and accurate identification and classification of sensor faults are achieved.

CN123069095APending Publication Date: 2026-11-10JIANGSU YUWELL POCT BIOLOGICAL TECH CO LTD +3
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510591594.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-07
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies struggle to efficiently and accurately identify the fault modes of analyte sensors, especially when processing non-stationary signals, which can easily lead to misjudgments and an inability to accurately pinpoint the specific fault type.

Method used

By acquiring the monitoring signals from the analyte sensor within a preset time period prior to the current moment, extracting signal contour data and determining morphological feature data, and using a fault identification model to classify the morphological feature data, the fault modes of the sensor are identified.

Benefits of technology

It achieves efficient and timely identification of sensor fault modes for analytes, overcomes the limitations of fixed signal threshold judgment, and accurately identifies the specific fault type of the sensor.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069095A_ABST
    Figure CN123069095A_ABST
Patent Text Reader

Abstract

This application provides a method, device, and medium for fault identification of an analytical analyte sensor, belonging to the field of medical device technology. The method includes acquiring monitoring signals from the analytical analyte sensor within a preset time period prior to the current moment; extracting signal contour data from the monitoring signals; determining morphological feature data corresponding to the monitoring signals based on the signal contour data; and determining the fault mode of the analytical analyte sensor based on the morphological feature data and a fault identification model; wherein the fault identification model is trained based on morphological feature data from multiple historical monitoring signals. Thus, the above method efficiently, timely, and accurately identifies the fault mode of the analytical analyte sensor.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of medical device technology, and in particular to a method, device and medium for fault identification of an analytical sensor. Background Technology

[0002] The analyte sensor provides dynamic data support for users' health management by monitoring the concentration of analytes in their subcutaneous tissue fluid in real time. However, during long-term implantation, the analyte sensor is susceptible to factors such as biofilm formation, electrode oxidation, and external mechanical stress interference, which can easily lead to various failure modes, such as signal drift and noise spikes, severely affecting the accuracy of the monitoring data.

[0003] Currently, traditional failure detection methods mostly rely on statistical analysis or fixed threshold judgments. For example, they determine noise levels by calculating signal variance or identify signal drops based on preset amplitude thresholds. These methods perform poorly when dealing with non-stationary signals, struggle to distinguish the coupling effect between analyte data fluctuations and sensor failures, and are prone to misjudgment. Furthermore, traditional methods cannot accurately pinpoint specific failure mode types.

[0004] Therefore, there is an urgent need for a technical solution for fault identification of analytical analyte sensors to overcome the current technical problem of difficulty in efficiently and accurately identifying the fault modes of analytical analyte sensors. Summary of the Invention

[0005] To address the aforementioned issues, embodiments of this application provide a method, device, and medium for fault identification of analytical analyte sensors, thereby achieving efficient and accurate identification of fault modes of analytical analyte sensors.

[0006] In a first aspect, embodiments of this application provide a fault identification method for an analytical material sensor, the method comprising:

[0007] Acquire monitoring signals from the analyte sensor within a preset time period prior to the current time;

[0008] Extract the signal contour data of the monitoring signal, and determine the morphological feature data corresponding to the monitoring signal based on the signal contour data;

[0009] The fault mode of the analyte sensor is determined based on the morphological feature data and the fault identification model; wherein the fault identification model is trained based on the morphological feature data of multiple historical monitoring signals.

[0010] In one implementation of this application, the analyte sensor includes multiple preset fault modes, and determining the fault mode includes:

[0011] The morphological feature data is input into the fault identification model to obtain probability values ​​corresponding to multiple preset fault modes;

[0012] The preset fault mode corresponding to the highest probability value among the multiple probability values ​​is taken as the fault mode.

[0013] In one implementation of this application, the fault identification model includes a classification module and a selection module, and determining the fault mode includes:

[0014] The morphological feature data is input into the classification module to obtain probability values ​​corresponding to multiple preset fault modes;

[0015] Multiple probability values ​​are input into the selection module to obtain the maximum probability value, and the preset fault mode corresponding to the maximum probability value is taken as the fault mode.

[0016] In one implementation of this application, the function for determining the probability value is as follows:

[0017]

[0018] in, Let represent multiple probability values ​​corresponding to the i-th monitoring signal, t represent the number of iterations, T represent the total number of iterations, and i represent the i-th monitoring signal.

[0019] In one implementation of this application, the function for determining the maximum probability value is as follows:

[0020]

[0021] in, The maximum probability value; These are multiple probability values ​​corresponding to the i-th monitoring signal.

[0022] In one implementation of this application, before determining the failure mode, the method further includes:

[0023] The morphological feature data of multiple historical monitoring signals are converted into corresponding feature vector data respectively;

[0024] Each feature vector data is labeled with a corresponding preset fault mode, forming multiple training samples;

[0025] The fault identification model is obtained by training an initial classification model using the multiple training samples.

[0026] In one implementation of this application, the morphological feature data is converted into the feature vector data, as shown below:

[0027] x i =[x i1 ,x i2 ,x i3 ,xi4 ,x i5 ]

[0028] Where i represents the i-th monitoring signal; x i x represents the feature vector data corresponding to the i-th monitoring signal; i1 ,x i2 ,x i3 ,x i4 ,x i5 The corresponding morphological characteristic data representing the i-th monitoring signal include the intervals of steep rise or fall of the signal, the interval of colored noise signal, the signal pulse jump point, the interval of continuous abnormal high value, and the interval of continuous abnormal low value.

[0029] In one implementation of this application, extracting the signal contour data of the monitoring signal includes:

[0030] Perform a closing operation on the monitoring signal to obtain the upper boundary contour data;

[0031] Perform an opening operation on the monitoring signal to obtain the lower boundary contour data;

[0032] The signal contour data is obtained by performing a closing operation on the upper boundary contour data and the lower boundary contour data respectively.

[0033] In one implementation of this application, determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes:

[0034] Differential calculations are performed on the upper boundary contour data and the lower boundary contour data in the signal contour data to determine the differential array group;

[0035] The curve boundary expansion area value is calculated based on the difference array group, and the colored noise signal interval is determined based on the curve boundary expansion area value.

[0036] The colored noise signal range is determined as the morphological feature data.

[0037] In one implementation of this application, determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes:

[0038] Differential calculations are performed on the upper boundary contour data and the lower boundary contour data in the signal contour data to determine the differential array group;

[0039] Within a preset sliding window, the difference values ​​of each difference column in the difference column group are accumulated.

[0040] The steep rise or fall interval of the signal is determined based on the cumulative result of the difference value, and the steep rise or fall interval of the signal is determined as the morphological feature data.

[0041] In one implementation of this application, determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes:

[0042] The detrending signal curve is determined by subtracting the corresponding coordinate points of the monitoring signal from those of the signal contour data.

[0043] The signal pulse collision point is determined by comparing the detrended signal curve with a preset threshold.

[0044] The signal pulse collision point is determined as the morphological feature data.

[0045] In one implementation of this application, determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes:

[0046] Select the signal value before and after the signal pulse collision point, and replace the signal pulse collision point with the average of the before and after signal values ​​to obtain the preprocessed monitoring signal;

[0047] The abnormal high value duration interval is determined based on the preprocessed monitoring signal and the first preset judgment threshold, and the abnormal high value duration interval is determined as the morphological feature data.

[0048] In one implementation of this application, after obtaining the preprocessed monitoring signal, the method further includes:

[0049] Calculate the average signal value within the preset sliding window corresponding to the preprocessed monitoring signal;

[0050] The abnormally low value duration interval is determined based on the average signal value and the second preset judgment threshold, and the abnormally low value duration interval is determined as the morphological feature data.

[0051] Secondly, embodiments of this application also provide a fault identification device for an analytical material sensor, the device comprising:

[0052] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a fault identification method for an analytical sensor as described in the first aspect above.

[0053] Thirdly, embodiments of this application also provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as a fault identification method for an analytical material sensor as described in the first aspect above.

[0054] Compared with the prior art, the significant advantages of this application are as follows:

[0055] The technical solution of this application acquires the analyte sensor monitoring signal within a preset time period before the current moment, extracts and deeply analyzes the signal contour data of the monitoring signal, thereby determining the morphological feature data of the monitoring signal. This overcomes the limitation of the prior art in determining sensor failure by using a fixed signal threshold, and realizes the timely and accurate extraction of key features of the fault mode. At the same time, it uses a fault identification model to efficiently and accurately classify multiple fault modes of the morphological feature data of the sensor monitoring signal, and clarifies the specific fault mode corresponding to the sensor with monitoring signal fault features. This solves the technical problem that it is currently difficult to efficiently and accurately identify the fault modes of analyte sensors. Attached Figure Description

[0056] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0057] Figure 1 This is a flowchart illustrating a fault identification method for an analytical material sensor according to an embodiment of this application.

[0058] Figure 2 This is a schematic diagram showing the delay corresponding to the opening operation of the signal in a fault identification method for an analytical material sensor according to an embodiment of this application;

[0059] Figure 3 This is a schematic diagram showing the delay corresponding to the signal closing operation in a fault identification method for an analytical material sensor according to an embodiment of this application;

[0060] Figure 4 This is a schematic diagram of the raw signal with multiple marker intervals in a fault identification method for an analyte sensor according to an embodiment of this application;

[0061] Figure 5 This is a schematic diagram illustrating the extraction of signal pulse collisions in a fault identification method for an analytical material sensor according to an embodiment of this application.

[0062] Figure 6 This is a schematic diagram of the fault mode determination result in a fault identification method for an analytical material sensor according to an embodiment of this application;

[0063] Figure 7 This is a schematic diagram of the structure of a fault identification device for an analytical material sensor in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] This application provides a method, device, and medium for fault identification of analytical analyte sensors, which addresses the current technical problem of difficulty in efficiently and accurately identifying fault modes of analytical analyte sensors.

[0066] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0067] This application provides a method for fault identification of an analytical material sensor, such as... Figure 1 As shown, the method may include steps S101-S103:

[0068] S101, acquire the monitoring signal of the analyte sensor within a preset time period before the current time.

[0069] The monitoring signals acquired and processed in this application are signals within a preset time period prior to the current moment. The preset time period can be simply understood as a time sliding window, for example, 500 milliseconds. The application acquires monitoring signals collected by the analyte sensor within the 500 milliseconds preceding the current moment, and performs subsequent steps S102-S103 on these signals. If the preset time period condition is not met before the current moment (e.g., only 200 milliseconds have elapsed since the analyte sensor started operating), the application continues to wait for 300 milliseconds until the preset time period condition is met. Then, the monitoring signals within the preset time period are acquired, and subsequent steps are executed. Step S101, by monitoring the monitoring signals in real time and analyzing the monitoring signals prior to the current moment, can promptly determine whether there are fault characteristics in the monitoring signals, allowing the user to take immediate countermeasures.

[0070] It should be noted that the analytes monitored by the analyte sensor in this application include one or more of blood glucose, blood ketones, uric acid, and lactic acid. Taking blood glucose as an example, the corresponding sensor is a blood glucose monitoring sensor, which is built into the blood glucose monitoring device. The execution subject of the fault identification method in this application can be the microcontroller in the chip of the blood glucose monitoring device itself, or it can be the chip or server of the terminal device. These are merely examples, and the execution subject is not limited to these. This application does not make any specific limitations on this. When the execution subject is the chip or server of the terminal device, the terminal device should be able to establish a communication connection with the blood glucose monitoring device and receive and analyze the monitoring signals collected by the blood glucose monitoring device.

[0071] It should be noted that the monitoring signal of the analyte sensor includes one or more of the following: current signal, voltage signal, or analyte concentration data.

[0072] S102, extract the signal contour data of the monitoring signal, and determine the morphological feature data corresponding to the monitoring signal based on the signal contour data.

[0073] In this embodiment of the application, the signal contour data for extracting the monitoring signal includes:

[0074] The upper boundary contour data is obtained by performing a closing operation on the monitoring signal, the lower boundary contour data is obtained by performing an opening operation on the monitoring signal, and the signal contour data is obtained by performing a closing operation on the upper boundary contour data and the lower boundary contour data respectively.

[0075] In other words, the analyte sensor can collect monitoring signals within a preset time period and send them to the aforementioned execution entity, such as a microcontroller. The microcontroller further uses morphological operations to extract the signal contour data of the real-time monitoring signal. Specifically, the monitoring signal is processed by closing operations to extract the upper boundary contour data, i.e., the morphological features corresponding to the upper boundary contour of the monitoring signal; the real-time monitoring signal is processed by opening operations to extract its lower boundary contour data, i.e., the morphological features corresponding to the lower boundary contour of the monitoring signal; then, the extracted upper and lower boundary contour data are processed by closing operations respectively to obtain the signal contour data of the monitoring signal.

[0076] In the above embodiments of this application, after performing a combination of opening and closing operations on the upper and lower boundaries, a closing operation is further performed to remove burrs from the contour edges, resulting in smooth and continuous signal contour data.

[0077] Furthermore, in implementing the closing and opening operations, this application employs maximum and minimum value filters to implement the basic dilation and erosion algorithms, and introduces a sliding window for real-time calculation, with a window size N. wThese can be preset and are not specifically limited here. The closing operation is dilation followed by erosion, used to remove small noise or details from a signal; the opening operation is erosion followed by dilation, used to fill small holes or broken connections in a signal. The dilation operation can be viewed as a maxima filter; it uses structuring elements to find local maxima in the signal and expands along these directions, resulting in a profile that reflects the diffusion of the original signal's maxima. Conversely, the erosion operation is a minima filter; it is used to reduce the signal by finding local minima to narrow the signal range, ultimately obtaining a profile that reflects the diffusion of the signal's minimum values.

[0078] The filter in this application is a causal filter, whose output depends only on current and past data, not future data. During the aforementioned calculation process, the maximum or minimum value can only be accurately calculated when the sliding window is completely filled with data. The latest output data does not immediately affect the output; it must wait until it is fully evaluated within the entire window, introducing an estimation delay. In other words, the signal processed in this application is a monitoring signal within a preset time period.

[0079] For estimating latency, for example, when extracting upper boundary contour data, a total of four separate dilation and erosion operations are required through two closing operations. Therefore, the real-time computation latency is 4 × (N). w / / 2+1), / / represents integer division. Similarly, when extracting the lower boundary contour data, a total of four separate dilation and erosion operations are required through one opening operation and one closing operation. The real-time computation delay is 4×(N). w / / 2+1), / / represents integer division. For example... Figure 2 , Figure 3 As shown, the gray lines represent monitoring signals. Figure 2 The Opened Signal diagram illustrates the delay corresponding to the opening operation on the signal. The green Batch Opened line represents the lower boundary of the batch opening operation with the estimated delay added above, and the red PointwiseOpened line represents the lower boundary of the pointwise opening operation with the estimated delay added above. Figure 3 The Closed Signal diagram illustrates the delay corresponding to the closing operation on the signal. The green Batch Closed line represents the upper boundary of the batch closing operation with the estimated delay added above, and the red Pointwise Closed line represents the upper boundary of the pointwise closing operation with the estimated delay added above.

[0080] Ideally, the real-time monitoring signal (current signal) acquired by a blood glucose monitoring sensor exhibits a high degree of consistency. Specifically, for a normally functioning blood glucose monitoring sensor, from a mathematical morphology perspective, the upper and lower boundaries of the current signal almost completely coincide, indicating minimal fluctuations in the signal over time and that the current value is distributed within a preset normal range. This signal consistency and stability reflect the sensor's good operating condition. Conversely, any other signal characteristics indicate varying degrees of signal anomalies, or even sensor malfunctions. In the embodiments of this application, the types of sensor malfunction characteristics include one or more of the following: colored noise, steep signal rise / fall, signal pulse overlap, abnormally high signal values, and abnormally low signal values. The morphological characteristics of analyte sensor malfunctions are shown in Table 1.

[0081] Table 1 Fault Morphological Characterization Table

[0082]

[0083] The various fault morphological representations in the table above all have corresponding morphological feature data. This application can use the following embodiments to extract the morphological feature data of each morphological representation, and then perform sensor fault identification.

[0084] In this embodiment of the application, for the morphological characterization of colored noise, the morphological feature data corresponding to the monitoring signal is determined based on the signal contour data, specifically including:

[0085] Differential calculations are performed on the upper and lower boundary contour data of the signal contour data to determine the differential fractional group. The curve boundary expansion area value is calculated based on the differential fractional group. The colored noise signal interval is determined based on the curve boundary expansion area value. The colored noise signal interval is determined as the morphological feature data corresponding to the monitoring signal.

[0086] The aforementioned colored noise signal interval corresponds to colored noise in the monitoring signal. The oscillation of the monitoring signal can be determined through the aforementioned differential processing, area value calculation, and threshold filtering. The oscillation amplitude can be characterized by the area value of the curve boundary expansion; the larger the area value, the larger the oscillation amplitude. When the area value of the curve boundary expansion exceeds a preset area value threshold, the oscillation amplitude is determined to reach the level of colored noise. The abscissa interval of the monitoring signal where the area value of the curve boundary expansion exceeds the preset area value threshold is taken as the colored noise signal interval. This colored noise signal interval can be used as morphological characteristic data for judging fault modes. The aforementioned preset area value threshold can be set by the user according to the actual usage scenario; this application does not specifically limit it.

[0087] Specifically, this application performs differential calculations on the contour curve corresponding to the upper boundary contour data, and determines the corresponding first differential sequence based on the result of the differential calculation; simultaneously, it performs differential calculations on the contour curve corresponding to the lower boundary contour data, and determines the corresponding second differential sequence based on the result of the differential calculation. The first and second differential sequences form a differential sequence group. Subsequently, the absolute values ​​of corresponding items in the first and second differential sequences are added together, and the sum of each absolute value is used as the expansion area value of each curve boundary; wherein, each curve boundary expansion area value corresponds one-to-one with each abscissa point of the monitoring signal. The corresponding item can be understood as the abscissa correspondence of elements in the two sequences. Based on the comparison result of each curve boundary expansion area value with a preset area value threshold, the colored noise signal interval in the contour curve corresponding to the signal contour data is determined, and the colored noise signal interval is added to the morphological feature information.

[0088] like Figure 4 As shown, "Original Signal with Multiple Marked Intervals" is a schematic diagram of an original signal with multiple marked intervals. The gray "Original Signal" represents the original signal, the green "Upper" represents the upper boundary, the red "Lower" represents the lower boundary, the blue "Up" dots represent the upward jump points of the signal pulse, and the red "Down" dots represent the downward jump points of the signal pulse. The red area of ​​Marked Interval 1 (marked interval 1) represents a steeply rising interval; the green area of ​​Marked Interval 2 (marked interval 2) represents a steeply falling interval; the yellow area of ​​Marked Interval 3 (marked interval 3) corresponds to the colored noise signal interval; and the purple area of ​​Marked Interval 4 (marked interval 4) corresponds to the interval of abnormally low signal values.

[0089] In one embodiment of this application, regarding the morphological characterization of a steep rise / fall fault in the signal, the morphological feature data corresponding to the monitoring signal is determined based on the signal contour data, specifically including:

[0090] Differential calculations are performed on the upper and lower boundary contour data of the signal contour data to determine the differential column groups. Within a preset sliding window, the differential values ​​of each differential column in the differential column group are accumulated. Based on the accumulated differential values, the steep rising or falling interval of the signal is determined. The steep rising or falling interval of the signal is then identified as morphological feature data.

[0091] Steep signal rises / falls primarily target monitoring signals that, during stable operation, encounter sudden and sustained upward or downward trends, accompanied by oscillations within the overall trend. When extracting the steep rise and fall intervals corresponding to these steep rises / falls, similar to the identification of colored noise mentioned above, differential calculations are first performed to obtain a series of differential values. Subsequently, a preset sliding window is used to accumulate the differential values. The accumulated differential values ​​are then used to obtain the steep rise and fall intervals of the signal, extracting the morphological characterization of the steep rise / fall fault.

[0092] Specifically, by performing difference calculations on the contour curves corresponding to the upper boundary contour data in the above steps to determine the corresponding first difference column; and by performing difference calculations on the contour curves corresponding to the lower boundary contour data to determine the corresponding second difference column, a group of difference columns is obtained. Within a preset sliding window N... w Within the first and second difference columns, the difference values ​​are accumulated to obtain the upper boundary sum and the lower boundary sum. The upper boundary sum is compared with a first preset sum threshold, and the lower boundary sum is compared with a second preset sum threshold. Based on the two comparison results, the steep rise and fall intervals of the signal are determined and added to the morphological feature information. For example, if the upper boundary sum is positive and greater than the first preset sum threshold, the abscissa of the corresponding monitoring signal is added to the steep rise interval; if the lower boundary sum is less than the second preset sum threshold, the abscissa of the corresponding monitoring signal is added to the steep fall interval.

[0093] In another embodiment of this application, the morphological characterization of signal pulse jump faults refers to a brief spike or pulse in a one-dimensional monitoring signal that suddenly appears with an amplitude significantly higher than normal. The duration of the jump signal is very short, often occupying only a few sampling point intervals, after which it quickly returns to normal. This application determines the morphological feature data corresponding to the monitoring signal based on signal contour data, specifically including:

[0094] The detrended signal curve group is determined by subtracting the corresponding coordinate points of the monitored signal from those of the signal contour data. The detrended signal curve is then compared with a preset threshold to determine the signal pulse collision point. The signal pulse collision point is then identified as morphological feature data.

[0095] In other words, this application extracts signal pulse jump points from morphological feature data through coordinate value processing and threshold comparison. Specifically, the monitoring signal is subtracted from the corresponding coordinate points of the signal contour data (including upper and lower boundary contour data) to determine the detrended signal curve group based on the subtraction result. The corresponding coordinate point subtraction involves subtracting the ordinate values ​​of the same horizontal axis from the corresponding signal curves of the monitoring signal and signal contour data. The horizontal axis of the signal curve represents time, and the vertical axis represents current value. The detrended signal curve group includes curves corresponding to the upper and lower boundaries, respectively. The coordinate values ​​of each curve of the upper boundary detrended signal curve are compared with a first preset threshold to determine the signal pulse jump point based on the first comparison result. If the curve coordinate value is greater than the first preset threshold, the coordinate point corresponding to that curve coordinate value is the signal pulse jump point.

[0096] Subsequently, the signal curve of the real-time monitoring signal is subjected to differential calculation (i.e., the value of the later point is subtracted from the value of the earlier point; in discrete systems, calculating the difference is equivalent to calculating the derivative, which represents the rate of change of the data), and a preset sliding window N is used. w The results of each difference calculation are summed in a window (this summation can be achieved by convolving a vector of all 1s) to obtain the cumulative sum of each window. The coordinate values ​​of each curve of the lower boundary detrending signal curve are compared with a second preset threshold to determine the second comparison result, and the cumulative sum of each window is compared with a third preset threshold to determine the third comparison result. Based on the second and third comparison results, the lower jump point of the signal pulse is determined. If the coordinate value of the lower boundary detrending signal curve is greater than the second preset threshold and the cumulative sum of the windows is less than the third preset threshold, then the coordinate point corresponding to the coordinate value of the lower boundary detrending signal curve is determined as the lower jump point of the signal pulse. Then, the upper jump point and the lower jump point of the signal pulse are added to the morphological feature information.

[0097] like Figure 5 As shown, Figure 5 The diagram shows the signal pulse collisions extracted from the Original Signal. The original signal is represented by gray lines, the blue lines corresponding to Opening+Closing represent the lower boundary contour data, and the green lines corresponding to Closing+Closing represent the upper boundary contour data. In the Morphological Pulse curve, blue pulses represent the upper pulse jump points, and red pulses represent the lower pulse jump points.

[0098] The first, second, and third preset thresholds mentioned above are set by the user according to the actual scenario, and this application does not impose specific limitations on them. Since the upper jump point in the monitored signal usually manifests as a single-point pulse wave, the determination of the upper jump point of the signal pulse is achieved by threshold comparison. The lower jump point is often accompanied by signal oscillation, manifesting as a spike signal with a differential cumulative sum of 0. Therefore, the lower jump point of the signal pulse is detected by comparing the curve coordinate value with the second preset threshold and the window cumulative sum with the third preset threshold.

[0099] For the morphological characterization of abnormally high signal values, the signal suddenly surges to an extremely high level without immediate decline, or there is a decline process, but it continues to oscillate at a high level afterward. In another embodiment of this application, the morphological feature data corresponding to the monitoring signal is determined based on the signal contour data, specifically including:

[0100] The signal values ​​preceding and following the signal pulse collision point are selected, and the average of the preceding and following signal values ​​is used to replace the signal pulse collision point to obtain the preprocessed monitoring signal. Based on the preprocessed monitoring signal and a first preset judgment threshold, the abnormally high value duration interval is determined, and this abnormally high value duration interval is identified as the morphological feature data corresponding to the monitoring signal.

[0101] In other words, based on the upward and downward jump points of each signal pulse obtained from the above steps, the preceding and following signal values ​​of each corresponding signal pulse jump point (including the upward and downward jump points) are determined. This can be simply understood as the coordinates before and after the pulse jump signal, which include the preceding and following coordinates adjacent to the pulse jump point. The average of the preceding and following signal values ​​is used to replace the corresponding signal pulse jump point, thereby filtering the signal pulse jump points and obtaining the preprocessed monitoring signal. A first preset judgment threshold is used to filter the sustained excessive characteristics of the preprocessed monitoring signal; signals exceeding the first preset judgment threshold are considered to be in a sustained excessive state, representing abnormally high signal values, otherwise they are considered normal signals. The abnormally high value duration interval corresponding to the abnormally high signal values ​​in the preprocessed monitoring signal is determined, and this abnormally high value duration interval is added to the morphological feature information. The first preset judgment threshold can be set by the user according to the actual usage scenario or by experts; this application does not specifically limit this.

[0102] The morphological characterization of abnormally low signal values ​​differs from that of abnormally high signal values. Monitoring signals may frequently oscillate near their minimum values ​​due to physical contact, waterproofing, etc. In this process, the minimum value generated immediately upon reaching its bottom should not be considered a continuous bottoming process. Because frequent minimum values ​​manifest as frequent fluctuations and are not a continuous bottoming process, this segment of the signal can be identified as containing colored noise or violent oscillations. In another embodiment of this application, after obtaining the aforementioned preprocessed monitoring signal, the method further includes:

[0103] Calculate the mean signal value within a preset sliding window corresponding to the preprocessed monitoring signal. Determine the duration of abnormally low values ​​based on the mean signal value and a second preset judgment threshold, and define the duration of abnormally low values ​​as morphological feature data.

[0104] In other words, after obtaining the preprocessed monitoring signal according to the above steps, this application will further slide a preset sliding window along the horizontal axis of the preprocessed monitoring signal and calculate the average signal value within the window at each sliding position. Based on the average signal value and a second preset judgment threshold, signals less than the second preset judgment threshold are determined as abnormally low signal values; otherwise, they are considered normal signals. The corresponding abnormally low value duration interval is obtained based on the abnormally low signal value, and this interval is added to the morphological feature information. The second preset judgment threshold can be set by the user according to the actual usage scenario or by an expert; this application does not specifically limit this setting.

[0105] According to the above embodiments, morphological feature data including colored noise, steep rise / fall, signal pulse collisions, abnormally high signal values, and abnormally low signal values ​​can be obtained efficiently, timely, and accurately, so as to fully acquire the feature information of the monitoring signal and facilitate subsequent fault identification of the analyte sensor.

[0106] S103, based on morphological feature data and fault identification model, determine the fault mode of the analyte sensor.

[0107] The fault identification model is trained based on the morphological feature data of multiple historical monitoring signals.

[0108] In this embodiment of the application, the analyte sensor includes multiple preset fault modes, and the fault mode is determined as follows:

[0109] The morphological feature data is input into the fault identification model to obtain the probability values ​​corresponding to multiple preset fault modes. The preset fault mode corresponding to the maximum probability value among the multiple probability values ​​is taken as the fault mode.

[0110] In other words, this application has a pre-set fault identification model that can analyze the morphological feature data extracted above, obtain the probability value predicted for each preset fault mode, and take the preset fault mode corresponding to the highest probability value among the probability values ​​as the fault mode of the analyzed object sensor.

[0111] For example, a set of multiple preset fault modes is {EC1, EC2, EC3, EC4, EC5}, where each element corresponds to a preset fault mode. This application inputs real-time monitoring signals and their morphological characteristics into a fault identification model. The fault identification model predicts the probability values ​​corresponding to each preset fault mode, such as EC1 = 0.2, EC2 = 0.3, EC3 = 0.1, EC4 = 0.2, EC5 = 0.2. The maximum probability value is 0.3, and the fault mode of the analyte sensor is EC2.

[0112] In another embodiment of this application, the fault identification model includes a classification module and a selection module, and determining the fault mode includes:

[0113] The morphological feature data is input into the classification module to obtain the probability values ​​corresponding to multiple preset fault modes; the multiple probability values ​​are input into the selection module to obtain the maximum probability value, and the preset fault mode corresponding to the maximum probability value is taken as the fault mode.

[0114] In other words, the fault identification model of this application can include a classification module and a selection module, which respectively perform probability value classification calculation and the selection of the maximum probability value, in order to obtain the aforementioned fault modes. This modular approach shortens the development cycle in actual development work, improves development efficiency, and allows for easier updating and maintenance of the fault identification model, eliminating the need for overall updates and reducing the time cost of repetitive development of the fault identification model.

[0115] In one embodiment of this application, the function for determining the above probability value is as follows:

[0116]

[0117] in, Let t represent the multiple probability values ​​corresponding to the i-th monitoring signal, T represent the total number of iterations, and i represent the i-th monitoring signal. i Let be the feature vector data of the i-th monitoring signal.

[0118] This application uses the above formula to iteratively process the feature vector of the monitoring signal, and analyzes and obtains the probability values ​​of the monitoring signal corresponding to different dimensions of morphological feature data. For example, it includes the probability values ​​corresponding to the steep rise or fall interval of the signal, the colored noise signal interval, the signal pulse collision point, the interval of continuous abnormal high value, and the interval of continuous abnormal low value. That is, the above formula can output five probability values ​​of a monitoring signal in the above five dimensions.

[0119] In another embodiment of this application, the function for determining the maximum probability value is as follows:

[0120]

[0121] in, The maximum probability value; These are multiple probability values ​​corresponding to the i-th monitoring signal.

[0122] In other words, after obtaining the five probability values ​​corresponding to the same monitoring signal in the above five dimensions, this application can compare the maximum values ​​and extract the maximum probability value using the function corresponding to the set of independent variable points with the maximum value. It should be noted that the above example of this application only shows the probability values ​​of five dimensions. With the advancement of technology, in addition to the probability values ​​corresponding to the steep rise or fall interval of the signal, the colored noise signal interval, the signal pulse collision point, the interval of continuous abnormal high value, and the interval of continuous abnormal low value, the dimensions can be continuously expanded. This application does not make specific limitations on this.

[0123] In another embodiment of this application, the fault identification model needs to be trained to determine the fault mode; therefore, before determining the fault mode, the method further includes:

[0124] The morphological feature data of multiple historical monitoring signals are converted into corresponding feature vector data. Each feature vector data is labeled with a corresponding preset fault mode, forming multiple training samples. The initial classification model is trained using multiple training samples to obtain the fault identification model.

[0125] In other words, this application can extract corresponding morphological feature data from multiple historical monitoring signals and convert them into feature vector data. Experts or users can then label each feature vector data with a preset fault mode tag, such as EC1, EC2, EC3, EC4, and EC5 mentioned above. Specific examples of the preset fault mode tags are as follows: EC1 indicates sensor waterproofing failure, mainly due to increased resistance in electronic components caused by water ingress, thus affecting the analyte sensor's monitoring signal. EC2 indicates sensor outer membrane damage, mainly due to damage to the dense filter membrane of the sensor's molecular outer membrane, causing uncontrolled electrochemical enzyme reaction in the analyte sensor, thus affecting the measurement signal. EC3 indicates sensor detachment, mainly due to external factors causing the sensor's soft probe to detach from or show signs of detachment from human tissue fluid, thus affecting the monitoring signal. EC4 indicates poor sensor contact, mainly due to external factors causing interference in the signal transmission of the sensor's soft needle or its internal components, resulting in the monitoring signal being submerged in a large amount of noise. EC5 indicates sensor sensitivity attenuation, mainly due to external factors causing abnormal wear and tear on the sensor's enzymes, preventing the expected electrochemical reaction process from being achieved, thus affecting the analyte sensor's monitoring signal.

[0126] The aforementioned feature vector data and their corresponding preset fault mode labels are used as training samples and input into the initial classification model for model training. The training continues until the model output accuracy is greater than the preset value, at which point the trained fault identification model is obtained. The preset value can be set by the user and is not specifically limited here.

[0127] In another embodiment of this application, the initial classification model of the fault identification model can be XGBoost, or other classification algorithm models can be selected; this application does not impose specific limitations on this. When XGBoost is selected as the initial classification mode, in order to extract effective feature vector data during the model training phase, a time sliding window can be moved with a fixed step size starting from the beginning position of the historical monitoring signal until the signal ends. Within each time sliding window, the original monitoring signal current and morphological feature data are extracted to form the i-th feature vector data x. i This is used as input to the XGBoost model to train the model and predict failure modes. i =[x i1 ,x i2 ,x i3 ,…,x in ], x in This represents the nth morphological feature data extracted from the i-th historical monitoring signal, containing the type of sensor fault characterization and its coordinate position in the historical monitoring signal. Through the fault identification model, it can construct feature vector data based on the morphological feature data of the input monitoring signal, and use its trained XGBoost model to output the predicted probability distribution corresponding to this feature vector data. Then determine based on the selected module The maximum probability value in Will The corresponding fault mode was determined to be the fault mode of the analyte sensor.

[0128] For example, the morphological feature data of the monitoring signal is converted into feature vector data, as shown below:

[0129] x i =[x i1 ,x i2 ,x i3 ,x i4 ,x i5 ]

[0130] Where i represents the i-th monitoring signal; x i x represents the feature vector data corresponding to the i-th monitoring signal; i1 ,x i2 ,x i3 ,x i4 ,x i5 The corresponding morphological characteristic data representing the i-th monitoring signal include the intervals of steep rise or fall of the signal, the interval of colored noise signal, the signal pulse jump point, the interval of continuous abnormal high value, and the interval of continuous abnormal low value.

[0131] The objective function of the XGBoost model in this application is as follows:

[0132]

[0133] Among them, Obj t y represents the objective function loss; t represents the number of iterations; n represents the number of monitored signals; l is the loss function, which is set by the user according to the actual use case and is not specifically limited here; i Let i be the true tag value of the i-th monitoring signal. Let Ω(f) be the probability value corresponding to the i-th monitoring signal in the t-round tree model, i.e., the updated prediction data; t ) is a preset regularization function used to control the complexity of the tree. It can be set by the user according to the actual use case, and there is no specific limitation here; constant is a preset constant term, which can be set by the user according to the actual use case, and there is no specific limitation here.

[0134] In addition, to further improve the predictive performance of the model, this application may also use Bayesian optimization method to optimize the parameters of the above XGBoost model, or other parameter optimization algorithms may be used for parameter optimization. This application does not make specific limitations on this.

[0135] By adopting the above technical solution, the monitoring signals of the analyte sensor within a preset time period before the current moment are acquired, and the signal contour data of the monitoring signals are extracted and deeply analyzed to determine the morphological feature data of the monitoring signals. This overcomes the limitation of the existing technology that uses a fixed signal threshold to determine sensor failure, and accurately extracts the key features of the fault mode. At the same time, the fault identification model is used to efficiently and accurately classify the morphological feature data of the sensor monitoring signals into multiple fault modes, and to identify the specific fault mode corresponding to the sensor with monitoring signal fault characteristics. This solves the current technical problem of difficulty in efficiently and accurately identifying the fault modes of analyte sensors.

[0136] Figure 6 This is a schematic diagram of the fault mode determination results provided in the embodiments of this application, such as... Figure 6 As shown:

[0137] Morphological Dilation and Erosion Filtering is a graph showing the curves after morphological dilation and erosion filtering. Gray represents the Original Signal, blue Opening (Erosion + Dilation) corresponds to the lower boundary (Erosion is performed first, followed by Dilation), and green Closing (Dilation + Erosion) corresponds to the upper boundary. Rise represents a steep rise in signal strength; Fall represents a steep fall in signal strength. Morphological Non-stable is a graph showing morphological instability, with Non-stable representing the colored noise signal range. Morphological Scale is a graph showing morphological scale, with Trend Signal representing the pre-processed monitoring signal, too large representing a persistent range of abnormally high values, and too small representing a persistent range of abnormally low values. Morphological EC Detection is a graph showing morphological EC detection, with EC1, EC2, EC3, EC4, and EC5 representing the fault modes corresponding to the maximum probability values ​​of the monitoring signal at this moment.

[0138] This application also provides a structural schematic diagram of a fault identification device for an analytical material sensor, such as... Figure 7 As shown, the device includes:

[0139] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0140] The system acquires monitoring signals from the analyte sensor within a preset time period prior to the current moment; extracts signal contour data from the monitoring signals; determines morphological feature data corresponding to the monitoring signals based on the signal contour data; and determines the fault mode of the analyte sensor based on the morphological feature data and a fault identification model; wherein the fault identification model is trained based on morphological feature data from multiple historical monitoring signals.

[0141] This application embodiment also provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0142] The system acquires monitoring signals from the analyte sensor within a preset time period prior to the current moment; extracts signal contour data from the monitoring signals; determines morphological feature data corresponding to the monitoring signals based on the signal contour data; and determines the fault mode of the analyte sensor based on the morphological feature data and a fault identification model; wherein the fault identification model is trained based on morphological feature data from multiple historical monitoring signals.

[0143] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0144] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0145] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0146] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for fault identification of an analytical material sensor, characterized in that, The method includes: Acquire monitoring signals from the analyte sensor within a preset time period prior to the current time; Extract the signal contour data of the monitoring signal, and determine the morphological feature data corresponding to the monitoring signal based on the signal contour data; The fault mode of the analyte sensor is determined based on the morphological feature data and the fault identification model; wherein the fault identification model is trained based on the morphological feature data of multiple historical monitoring signals.

2. The fault identification method for an analytical material sensor according to claim 1, characterized in that, The analyte sensor includes multiple preset fault modes, and the fault modes are determined as follows: The morphological feature data is input into the fault identification model to obtain probability values ​​corresponding to multiple preset fault modes; The preset fault mode corresponding to the highest probability value among the multiple probability values ​​is taken as the fault mode.

3. The fault identification method for an analytical material sensor according to claim 1, characterized in that, The fault identification model includes a classification module and a selection module, and determining the fault mode includes: The morphological feature data is input into the classification module to obtain probability values ​​corresponding to multiple preset fault modes; Multiple probability values ​​are input into the selection module to obtain the maximum probability value, and the preset fault mode corresponding to the maximum probability value is taken as the fault mode.

4. A fault identification method for an analytical material sensor according to any one of claims 2 or 3, characterized in that, The function for determining the probability value is as follows: in, Let represent the multiple probability values ​​corresponding to the i-th monitoring signal, t represent the number of iterations, T represent the total number of iterations, and i represent the i-th monitoring signal.

5. A fault identification method for an analytical material sensor according to any one of claims 2 or 3, characterized in that, The function for determining the maximum probability value is as follows: in, The maximum probability value; These are multiple probability values ​​corresponding to the i-th monitoring signal.

6. The fault identification method for an analytical material sensor according to claim 1, characterized in that, Before determining the failure mode, the method further includes: The morphological feature data of multiple historical monitoring signals are converted into corresponding feature vector data respectively; Each feature vector data is labeled with a corresponding preset fault mode, forming multiple training samples; The fault identification model is obtained by training an initial classification model using the multiple training samples.

7. The fault identification method for an analytical material sensor according to claim 6, characterized in that, The morphological feature data is converted into the feature vector data, as shown below: x i =[x i1 ,x i2 ,x i3 ,x i4 ,x i5 ] Where i represents the i-th monitoring signal; x i x represents the feature vector data corresponding to the i-th monitoring signal; i1 ,x i2 ,x i3 ,x i4 ,x i5 The corresponding morphological characteristic data representing the i-th monitoring signal include the intervals of steep rise or fall of the signal, the interval of colored noise signal, the signal pulse jump point, the interval of continuous abnormal high value, and the interval of continuous abnormal low value.

8. The fault identification method for an analytical material sensor according to claim 1, characterized in that, The extraction of signal contour data from the monitoring signal includes: Perform a closing operation on the monitoring signal to obtain the upper boundary contour data; Perform an opening operation on the monitoring signal to obtain the lower boundary contour data; The signal contour data is obtained by performing a closing operation on the upper boundary contour data and the lower boundary contour data respectively.

9. The fault identification method for an analytical material sensor according to claim 1, characterized in that, The step of determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes: Differential calculations are performed on the upper boundary contour data and the lower boundary contour data in the signal contour data to determine the differential array group; The curve boundary expansion area value is calculated based on the difference array group, and the colored noise signal interval is determined based on the curve boundary expansion area value. The colored noise signal range is determined as the morphological feature data.

10. A fault identification method for an analytical material sensor according to claim 1, characterized in that, The step of determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes: Differential calculations are performed on the upper boundary contour data and the lower boundary contour data in the signal contour data to determine the differential array group; Within a preset sliding window, the difference values ​​of each difference column in the difference column group are accumulated. The steep rise or fall interval of the signal is determined based on the cumulative result of the difference value, and the steep rise or fall interval of the signal is determined as the morphological feature data.

11. The fault identification method for an analytical material sensor according to claim 1, characterized in that, The step of determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes: The detrending signal curve is determined by subtracting the corresponding coordinate points of the monitoring signal from those of the signal contour data. The signal pulse collision point is determined by comparing the detrended signal curve with a preset threshold. The signal pulse collision point is determined as the morphological feature data.

12. The fault identification method for an analytical material sensor according to claim 11, characterized in that, The step of determining the morphological feature data corresponding to the monitoring signal based on the signal contour data includes: Select the signal value before and after the signal pulse collision point, and replace the signal pulse collision point with the average of the before and after signal values ​​to obtain the preprocessed monitoring signal; The abnormal high value duration interval is determined based on the preprocessed monitoring signal and the first preset judgment threshold, and the abnormal high value duration interval is determined as the morphological feature data.

13. The fault identification method for an analytical material sensor according to claim 12, characterized in that, After obtaining the preprocessed monitoring signal, the method further includes: Calculate the average signal value within the preset sliding window corresponding to the preprocessed monitoring signal; The abnormally low value duration interval is determined based on the average signal value and the second preset judgment threshold, and the abnormally low value duration interval is determined as the morphological feature data.

14. A fault identification device for an analytical material sensor, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a fault identification method for an analytical sensor as described in any one of claims 1-13.

15. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured as a fault identification method for an analytical material sensor as described in any one of claims 1-13.