A digital medical device quality data intelligent management platform

Abstract

The present invention relates to the field of data processing technology, and in particular to a digital medical device quality data intelligent management platform. The platform includes a processor and a memory, and the processor executes a computer program in the memory to implement the following steps: obtaining a sample set of any type of medical device, classifying the sample set to obtain a fault-free sample set, an unknown fault sample set, and a known fault sample set; obtaining the rationality of each K value within a preset K value range when classifying each sample in the unknown fault sample set using a K nearest neighbor algorithm, to obtain a new unknown fault sample set and a new known fault sample set; obtaining an optimal K value for each sample in the new unknown fault sample set, obtaining an optimal K value based on the accuracy of classification when each optimal K value is used, and using the K nearest neighbor algorithm to classify the samples monitored in real time, so that the fault classification results of the medical device are more accurate.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a digital medical device quality data intelligent management platform. Background Art

[0002] With the rapid development of medical technology and the growing demand for healthcare, medical devices play a vital role in medical diagnosis and treatment. However, traditional medical device management methods face numerous challenges, such as incomplete device information records, difficult traceability, high maintenance costs, and data silos. These issues not only affect the efficiency and safety of medical devices but also bring significant inconvenience to the management of medical institutions. To address these issues, a digital medical device quality data intelligent management platform has emerged. Leveraging modern information technology, it provides comprehensive, intelligent management of medical devices throughout their entire lifecycle, from procurement, warehousing, use, maintenance, to disposal.

[0003] Digital medical device quality data intelligent management platforms usually use PHM algorithms to monitor various parameters of medical devices, analyze the quality of medical devices, and predict the health status and failure problems of medical devices, so that medical staff can promptly detect quality problems of medical devices and facilitate the repair or replacement of medical devices.

[0004] To better identify medical device failures and take timely action, existing digital medical device quality data intelligent management platforms construct a failure database based on historical failure data. Using the K-nearest neighbor algorithm, they retrieve the K fault data closest to the real-time monitoring data from the failure database. These K fault data then identify the fault type with the most fault data. This identifies the fault type for the medical device being monitored in real time, and based on the classification results, they take appropriate action. However, because the K value in the K-nearest neighbor algorithm is artificially defined, the choice of K value can significantly impact the algorithm's results. A small K value can cause the classification results to be affected by a small amount of data, leading to inaccurate classification. A large K value can result in an excessively large amount of data, making effective differentiation impossible and leading to inaccurate classification of medical device failures.

[0005] Therefore, how to obtain the optimal K value in the K-nearest neighbor algorithm to make the fault classification of medical devices more accurate has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, an embodiment of the present invention provides a digital medical device quality data intelligent management platform to solve the problem of how to obtain the optimal K value in the K nearest neighbor algorithm to make the fault classification of medical devices more accurate.

[0007] An embodiment of the present invention provides a digital medical device quality data intelligent management platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the following method is implemented:

[0008] For any type of medical device, the multi-dimensional historical parameter data of each medical device at each sampling moment is taken as a sample to obtain a sample set;

[0009] Classifying the samples in the sample set to obtain a fault sample set and a fault-free sample set, and dividing the fault sample set into an unknown fault sample set and a known fault sample set according to the similarity between the samples in the fault sample set;

[0010] Record any sample in the unknown fault sample set as an unknown fault sample, and obtain the rationality of each K value within a preset K value range when classifying the unknown fault sample using the K nearest neighbor algorithm based on the distance difference between the unknown fault sample and the samples in the known fault sample set, obtain the rationality of each K value corresponding to each unknown fault sample, and update the unknown fault sample set and the known fault sample set to obtain a new unknown fault sample set and a new known fault sample set;

[0011] Obtaining the rationality of each K value corresponding to each sample based on the distance difference between each sample in the new unknown fault sample set and the samples in the new known fault sample set, and obtaining the optimal K value for each sample based on the rationality of each K value corresponding to each sample;

[0012] The optimal K value is obtained based on the accuracy of classification for each optimal K value, and each sample monitored in real time is classified using the optimal K value as the value of K in the K nearest neighbor algorithm based on the fault-free sample set and the new known fault sample set.

[0013] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0014] The present invention is directed to any type of medical device, and takes the multidimensional historical parameter data of each medical device at each sampling moment as a sample to obtain a sample set; classifies the samples in the sample set to obtain a fault sample set and a fault-free sample set; and divides the fault sample set into an unknown fault sample set and a known fault sample set according to the similarity between the samples in the fault sample set; records any sample in the unknown fault sample set as an unknown fault sample, and obtains the rationality of each K value within a preset K value range when classifying the unknown fault sample using a K nearest neighbor algorithm, and obtains the probability of each unknown fault sample being classified as an unknown fault sample. According to the rationality of each K value corresponding to the fault sample, the unknown fault sample set and the known fault sample set are updated to obtain a new unknown fault sample set and a new known fault sample set; according to the distance difference between each sample in the new unknown fault sample set and the sample in the new known fault sample set, the rationality of each K value corresponding to each sample is obtained respectively, and according to the rationality of each K value corresponding to each sample, the optimal K value of each sample is obtained respectively; according to the accuracy of classification when each optimal K value is used, the optimal K value is obtained, and according to the fault-free sample set and the new known fault sample set, the optimal K value is used as the value of K in the K nearest neighbor algorithm to classify each sample monitored in real time. According to the rationality of each K value corresponding to each sample, the unknown fault sample set and the fault sample set are updated so that the classification of the fault samples in the new known fault sample set is more accurate; then the optimal K value of each sample is obtained, and according to the accuracy of classification when each optimal K value is used, the optimal K value that best meets the requirements of the sample under this type of medical device is obtained, so that the results of classifying the samples monitored in real time using the optimal K value using the K nearest neighbor algorithm are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 This is a flow chart of a method for intelligent management of digital medical device quality data provided by the first embodiment of the present invention;

[0017] Figure 2 This is a scatter plot provided by Example 1 of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0019] It should be noted that the terms "first," "second," and the like in the specification of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0020] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0021] The embodiment of the present invention provides a digital medical device quality data intelligent management platform, including a processor and a memory, wherein the processor executes a computer program in the memory to implement a digital medical device quality data intelligent management method, such as Figure 1 As shown, the method comprises the following steps:

[0022] Step S101 : for any type of medical device, multi-dimensional historical parameter data of each medical device at each sampling moment is taken as a sample to obtain a sample set.

[0023] The digital medical device quality data intelligent management platform leverages modern information technology to provide comprehensive, intelligent management of medical devices throughout their entire lifecycle, from procurement, warehousing, use, maintenance, to disposal. To better identify medical device failures, a fault database is typically constructed based on historical failure data. A K-nearest neighbor algorithm is then used to identify the type of failures in real-time monitored medical devices. This requires collecting multidimensional historical parameter data from medical devices to create a fault database.

[0024] The digital medical device quality data intelligent management platform collects multi-dimensional historical parameter data of each medical device at each sampling moment. Medical devices include but are not limited to ultrasonic therapy devices, gravity infusion devices, digital photography x-ray machines, ultrasonic Doppler fetal monitors, and high-frequency electrocautery therapy devices. Multi-dimensional parameters include but are not limited to transmission power, operating frequency, and sound beam uniformity. The multi-dimensional historical parameter data of each medical device at each sampling moment is used as a sample. Since samples are collected to obtain a fault database, in order to make the fault types contained in the fault database more comprehensive, it is necessary to collect enough samples to cover a more comprehensive range of fault types. Therefore, for any type of medical device, the collection frequency is set to once per minute, and samples of each medical device within 30 days are collected to form a sample set. There is no restriction here, and it can be set according to the specific implementation scenario.

[0025] Step S102 , classifying the samples in the sample set to obtain a fault sample set and a non-fault sample set, and dividing the fault sample set into an unknown fault sample set and a known fault sample set according to the similarity between the samples in the fault sample set.

[0026] In order to better identify the fault problems of medical devices, a fault database is usually constructed based on historical fault data. The K nearest neighbor algorithm is used to obtain the K fault data closest to the real-time monitoring data in the fault database, and the fault type with the most fault data in the K fault data is obtained to obtain the fault identification type result of the medical device monitored in real time.

[0027] Therefore, it is necessary to classify the sample set based on the acquired sample set to obtain a fault database. Since the samples in the sample set are not classified, the sample set includes faulty samples and non-faulty samples, and the fault database is obtained by classifying faulty samples. Therefore, it is necessary to first classify the sample set to obtain a faulty sample set and a non-faulty sample set.

[0028] Since there are multiple parameter data in a sample, under normal circumstances, each parameter data of the medical device should be within the normal parameter range, and the parameter data will not change significantly within a certain period of time. Therefore, based on the difference between the parameter data and its corresponding parameter range, as well as the fluctuation characteristics of the parameter data within a certain time range, it is possible to judge whether the parameter data is faulty parameter data, and then obtain a faulty sample set and a fault-free sample set.

[0029] Any sample in the sample set is recorded as a target sample. Samples belonging to the same medical device as the target sample are obtained from the sample set to form a reference sample set. Based on the difference between the parameter data in the target sample and the corresponding parameter range, as well as the fluctuation characteristics of the parameter data in the parameter sample set, it is determined whether each parameter data in the target sample is fault parameter data, thereby determining whether the target sample is a fault sample. The specific determination method is as follows:

[0030] (1) According to the fluctuation characteristics of parameter data within a certain time range, the change difference value of each parameter data in the target sample is obtained.

[0031] Specifically, for any parameter data in the target sample, parameter data belonging to the same parameter as the any parameter data is obtained from the reference sample set, and a parameter data sequence is formed together with the any parameter data. In the parameter data sequence, a target window of a preset size is constructed with the any parameter data as the center, and the preset size is used as a sliding step. The target window is slid to the left to obtain a first window, and the target window is slid to the right to obtain a second window.

[0032] Obtain the parameter data variances in the target window, the first window, and the second window respectively, obtain the mean of the parameter data variances between the first window and the second window, obtain the absolute value of the difference between the parameter data variance of the target window and the mean, obtain the degree of fluctuation of any parameter data, obtain the addition result of the fluctuation degree and a constant 1, and obtain the change difference value of any parameter data according to the difference between constant 1 and the reciprocal of the addition result.

[0033] In one embodiment, taking the i-th parameter data in the target sample as an example, the formula for calculating the change difference value of the i-th parameter data is:

[0034]

[0035] Among them, θ is the change difference value of the i-th parameter data; S is the variance of the parameter data in the target window; S1 is the variance of the parameter data in the first window; S2 is the variance of the parameter data in the second window; || is the absolute value sign; 1 is a constant.

[0036] It should be noted that the default size of the target window is 3, which is not limited here and can be set according to the specific implementation scenario. Indicates the degree of fluctuation of the i-th parameter data, The larger the value is, the greater the fluctuation of the i-th parameter data within a certain time range, the more consistent the i-th parameter data is with the fault parameter data, and the greater the change difference value of the i-th parameter data; in the parameter data sequence, if the i-th parameter data is the first parameter data, the i-th parameter data and the two consecutive parameter data after it form the target window, becomes |S-S2)|; if the i-th parameter data is the last parameter data, the i-th parameter data and the two consecutive parameter data before it form the target window, becomes |S-S1)|; if the parameter data before or after the target window is less than 3, then Becomes |S-S2)| or |S-S1)|.

[0037] (2) According to the change difference value of each parameter data in the target sample, and the difference between the parameter data and its corresponding parameter range, it is determined whether the target sample is a non-faulty sample.

[0038] Specifically, obtain the parameter range of the i-th parameter data, set the change difference threshold to 0.1, which is not limited here and can be set according to the specific implementation scenario. If the i-th parameter data is not within the parameter range, or the change difference value of the i-th parameter data is greater than or equal to 0.1, then the i-th parameter data is confirmed to be faulty parameter data.

[0039] Furthermore, each parameter data in the target sample is traversed. Specifically, according to the judgment method of the i-th parameter data, the fault parameter data in the target sample data is detected. If there is at least one fault parameter data in the target sample, the target sample is confirmed to be a fault sample.

[0040] According to the fault sample judgment method, each sample in the sample set is judged, all fault samples in the sample set are obtained to form a fault sample set, and non-fault samples in the sample set are formed into a non-fault sample set.

[0041] After obtaining the fault sample set, it is necessary to classify the fault samples in the fault sample set by fault type. This allows samples of different fault types to be organized into a fault database, which facilitates fault classification using the K-nearest neighbor algorithm. Because fault samples of the same fault type have a certain degree of similarity, the fault samples in the fault sample set can be classified based on their similarity to form a fault database.

[0042] For any two fault samples in the fault samples, when the fault degrees of the two fault samples are the same, it means that the two fault samples may be of the same fault type. Therefore, the similarity between the fault samples can be obtained based on the difference in the fault degrees between the fault samples.

[0043] For any fault sample in the fault sample set, when the abnormal value of the change of the parameter data in the fault sample is large, or the parameter data deviates greatly from its corresponding parameter range, it means that the fault degree of this fault sample is more serious. Therefore, any parameter data in any fault sample can be recorded as the target parameter data. According to the change difference value of the fault parameter data in the fault sample and the degree of deviation of the fault parameter data from its corresponding parameter range, the overall difference degree of the fault sample is obtained, and then the similarity degree between the fault samples is obtained according to the overall difference degree of the fault sample.

[0044] The method for obtaining the overall difference degree of the fault samples is as follows, based on the change difference value of the fault parameter data in the fault samples and the degree of deviation of the fault parameter data from its corresponding parameter range:

[0045] (1) According to the degree of deviation between the target parameter data and its corresponding parameter range, the numerical difference value of the target parameter data is obtained.

[0046] Specifically, obtaining a parameter range of the target parameter data, obtaining a maximum value and a minimum value within the parameter range of the target parameter data, and obtaining a second difference between the maximum value and the minimum value;

[0047] When the target parameter data is less than the minimum value, obtaining a first difference between the minimum value and the target parameter data, and obtaining a numerical difference value of the target parameter data according to a ratio of the first difference to the second difference;

[0048] When the target parameter data is greater than the maximum value, obtaining a third difference between the target parameter data and the maximum value, and obtaining a numerical difference value of the target parameter according to a ratio of the third difference to the second difference;

[0049] When the target parameter data is within the parameter range of the target parameter data, the numerical difference value of the target parameter is set to a preset value.

[0050] In one embodiment, the calculation formula of the numerical difference value of the target parameter data is:

[0051]

[0052] Where f is the numerical difference of the target parameter data; x is the target parameter data; x min is the minimum value within the parameter range of the target parameter data; x max It is the maximum value within the parameter range of the target parameter data; 0 is the default value.

[0053] It should be noted that the greater the difference between the target parameter data and the maximum or minimum value within its parameter range, the more the target parameter data deviates from its parameter range, and the greater the numerical difference value of the target parameter data; when the target parameter data is within its parameter range, it means that the target parameter data has not deviated from its parameter range, and therefore, the numerical difference value of the target parameter data is set to 0.

[0054] (2) Obtaining the abnormality degree of the target parameter data based on the average value between the numerical difference value and the change difference value of the target parameter data.

[0055] In one embodiment, according to the above-mentioned method for obtaining the change difference value of the i-th parameter data, the change difference value of the target parameter data is obtained, and combined with the numerical difference value of the target parameter data, the formula for calculating the abnormality degree of the target parameter data is:

[0056]

[0057] Among them, β is the abnormality degree of the target parameter data; f is the numerical difference value of the target parameter data; θ is the change difference value of the target parameter data.

[0058] It should be noted that the larger the numerical difference value of the target parameter data, the more the target parameter data deviates from its parameter range, the more likely the target parameter data is to be abnormal, and the greater the degree of abnormality of the target parameter data; the larger the change difference value of the target parameter data, the greater the fluctuation degree of the target parameter data, the more likely the target parameter data is to be abnormal, and the greater the degree of abnormality of the target parameter data.

[0059] (3) According to the above-mentioned method for obtaining the abnormality degree of the target parameter data, the abnormality degree of each parameter data in any one of the fault samples is obtained respectively, and then the sum of the abnormality degrees of each parameter data in any one of the fault samples is used as the overall abnormality degree of any one of the fault samples.

[0060] Furthermore, if the overall abnormality degree of two fault samples is the same, but the parameters corresponding to the fault parameter data in the two fault samples are different, it cannot be said that the two fault samples are similar. When the parameters corresponding to the fault parameter data in the two fault samples are the same, it means that the two fault samples are similar. Therefore, the similarity degree between the fault samples can be obtained based on the similarity of the parameters corresponding to the fault parameter data between the fault samples and the overall difference degree between the fault samples.

[0061] Because the parameters corresponding to the fault parameter data in the same fault sample may be correlated, that is, when one parameter fails, it may cause other parameters to fail as well. Therefore, the degree of parameter correlation in the fault sample can be obtained based on the degree of correlation between the parameters in the fault sample. When the parameters corresponding to the fault parameter data in two fault samples are the same, the degree of parameter correlation between the two fault samples is also the same. Therefore, the degree of parameter correlation between two fault samples can be used to reflect the similarity of the parameters corresponding to the fault parameter data in the two fault samples.

[0062] Among them, according to the correlation degree between each parameter in the fault sample, the method for obtaining the parameter correlation degree in the fault sample is as follows:

[0063] (1) For any dimensional parameter of any type of medical device, obtain the fault sample whose parameter data corresponding to the any dimensional parameter is the fault parameter data in the fault sample set as the fault sample mapping set of the any dimensional parameter, and obtain the association probability of each dimensional parameter with other dimensional parameters according to the intersection between the fault sample mapping sets of all dimensional parameters, and obtain the association probability sequence corresponding to each dimensional parameter.

[0064] For example: any dimensional parameter under any type of medical device is recorded as a Class I parameter, and the fault samples whose parameter data corresponding to the Class I parameter is the fault parameter data are obtained in the fault sample set. All fault samples whose parameter data corresponding to the Class I parameter is the fault parameter data are composed of the fault sample mapping set of the Class I parameter. Similarly, the fault sample mapping set of the Class II parameter can be obtained; the number of samples in the fault sample mapping set of the Class I parameter is used as the denominator to obtain the intersection of the fault sample mapping sets of the Class I parameter and the Class II parameter, the number of samples in the intersection of the fault sample mapping sets of the Class I parameter and the Class II parameter is used as the numerator, and the ratio of the numerator to the denominator is used as the association probability of the Class I parameter and the Class II parameter. Similarly, the association probability of the Class I parameter and each other type of parameter can be obtained, and the association probability of the Class I parameter and each other type of parameter is composed of the association probability sequence of the Class I parameter. Similarly, the association probability sequence corresponding to each type of parameter is obtained.

[0065] (2) If there are at least two fault parameter data in any of the fault samples, the dimension parameters corresponding to the fault parameter data in any of the fault samples are recorded as association parameters, and the parameter association degree of any of the fault samples is obtained based on the association probability between the association parameters of any of the fault samples.

[0066] Specifically, according to the number of associated parameters in any fault sample, a proportion of associated parameters in any fault sample is obtained;

[0067] In the association probability sequence corresponding to each dimensional parameter of any fault sample, the association probability of any two association parameters in any fault sample is obtained, and the corresponding association probability cumulative value is obtained;

[0068] The parameter correlation degree of any fault sample is obtained according to the product between the correlation parameter proportion and the cumulative value of the correlation probability.

[0069] In one embodiment, taking the t-th fault sample in the fault sample set as an example, the association probabilities of every two associated parameters in the t-th fault sample are combined into an associated parameter probability sequence of the t-th fault sample, and the calculation formula for calculating the parameter association degree of the t-th fault sample is:

[0070]

[0071] Where γ is the parameter correlation degree of the t-th fault sample; n is the number of correlated parameters in the t-th fault sample; N is the number of all parameters in the t-th fault sample; P j is the jth associated probability in the probability sequence of associated parameters of the tth fault sample; j is the sequence number of the associated probability in the probability sequence of associated parameters of the tth fault sample; is the number of associated probabilities in the associated parameter probability sequence of the t-th fault sample.

[0072] It should be noted that is the proportion of associated parameters in the t-th fault sample, The larger it is, the more associated parameters there are in the t-th fault sample, and the greater the degree of parameter correlation of the t-th fault sample; is the cumulative value of the association probability between the associated parameters in the t-th fault sample, The larger it is, the greater the association probability between the associated parameters in the t-th fault sample is, and the greater the degree of parameter correlation of the t-th fault sample is.

[0073] (3) If there are less than two fault parameter data in the tth fault sample, it means that the fault parameter data in the tth fault sample will not affect other parameter data, then there is no associated parameter in the tth fault sample, that is, the number of associated parameters in the tth fault sample is 0, the proportion of associated parameters in the tth fault sample is 0, and the parameter correlation degree of the tth fault sample is also 0.

[0074] Furthermore, the overall abnormality degree and parameter correlation degree of each fault sample in the fault sample set are obtained, and the fault samples in the fault sample set other than any one fault sample are taken as other fault samples. Based on the difference in the overall abnormality degree and parameter correlation degree between any one fault sample and each other fault sample, the similarity degree between any one fault sample and each other fault sample is obtained. Then, based on the similarity between the fault samples, the fault samples in the fault sample set are classified, and a fault database is obtained based on the classification results.

[0075] Wherein, according to the overall abnormality difference and parameter correlation difference between any fault sample and each other fault sample, the method for obtaining the similarity between any fault sample and each other fault sample is as follows:

[0076] For any other fault sample, obtain the absolute value of the difference between the overall abnormality degree of the any fault sample and the any other fault sample to obtain the overall abnormality degree difference, obtain the inverse of the sum of the overall abnormality degree difference and a constant 1, and obtain the overall abnormality degree similarity value between the any fault sample and the any other fault sample;

[0077] Obtaining an absolute value of a difference between the parameter correlation degrees of any one fault sample and any other fault sample to obtain a parameter correlation degree difference, obtaining a reciprocal of a sum of the parameter correlation degree difference and a constant 1 to obtain a parameter correlation degree similarity value between any one fault sample and any other fault sample;

[0078] The similarity between the fault sample and any other fault sample is obtained according to the average of the overall abnormality similarity value and the parameter correlation similarity value between the fault sample and any other fault sample.

[0079] In one embodiment, taking the t-th fault sample and the u-th fault sample in the fault sample set as an example, the formula for calculating the similarity between the t-th fault sample and the u-th fault sample is:

[0080]

[0081] Among them, σ tu is the similarity between the t-th fault sample and the u-th fault sample; β' t is the overall abnormality degree of the t-th fault sample; β' u is the overall abnormality degree of the u-th fault sample; γ t is the parameter correlation degree of the t-th fault sample; γ u is the parameter correlation degree of the u-th fault sample; || is the absolute value sign; 1 is a constant.

[0082] It should be noted that is the overall abnormality similarity value between the t-th fault sample and the u-th fault sample, The larger it is, the more similar the overall abnormality of the t-th fault sample and the u-th fault sample is, and the greater the similarity between the t-th fault sample and the u-th fault sample is; is the similarity value of the parameter correlation degree between the t-th fault sample and the u-th fault sample, The larger is, the more similar the parameter correlation between the tth fault sample and the uth fault sample is, and the greater the similarity between the tth fault sample and the uth fault sample is.

[0083] According to the above-mentioned method for obtaining the similarity between the t-th fault sample and the u-th fault sample, the similarity between every two fault samples in the fault sample set is obtained, and the fault samples in the fault sample set are classified. In order to ensure the correctness of the fault sample classification, a high similarity threshold needs to be set. The similarity threshold is set to 0.9 based on the results obtained from experimental statistical calculations. There is no limitation here and it can be set according to the specific implementation scenario. The other fault samples corresponding to the similarity greater than or equal to 0.9 and the any fault sample are combined into an initial fault sample subset corresponding to the any fault sample. The initial fault sample subset corresponding to each fault sample in the fault sample set is obtained, and the two initial fault sample subsets with intersection are divided into one category. All categories are merged using a recursive algorithm to obtain at least one fault sample subset. One fault sample subset corresponds to one fault type. All fault sample subsets are combined into a known fault sample set, and the fault samples in the fault sample set that do not belong to the known fault sample set are combined into an unknown fault sample set.

[0084] For example: Assume that the fault sample set is {A, B, C, D, E, F, G, H}, where A, B, C, D, E, F, G, and H are fault samples. For fault sample A, if the similarity between A and B is greater than 0.9, and the similarity between A and C is greater than 0.9, then A, B, and C are combined into the initial fault sample subset corresponding to A, namely {A, B, C}. Similarly, the initial fault sample subset corresponding to B is {A, B}, the initial fault sample subset corresponding to C is {A, C, F}, the initial fault sample subset corresponding to D is {D, F}, the initial fault sample subset corresponding to E is {D, E}, the initial fault sample subset corresponding to F is {F, C, G}, and the initial fault sample subset corresponding to G is {F, G}. H has no corresponding initial fault sample subset. Since the initial fault sample subset corresponding to B is {A, B} and the initial fault sample subset corresponding to C is {A, C, F}, the initial fault sample subset corresponding to D is {D, F}. The initial fault sample subset {A, C, F} all contain A, so {A, B} and {A, C, F} are divided into one category, and {A, B} and {A, C, F} form the set {A, B, C, F}. Since the initial fault sample subset {F, C, G} corresponding to the sets {A, B, C, F} and F all contain C and F, {A, B, C, F} and {F, C, G} are divided into one category, and {A, B, C, F} and {F, C, G} form the set {A, B, C, F, G}, obtaining two fault sample subsets {A, B, C, F, G} and {D, E}. The set {A, B, C, F, G} corresponds to fault type 1, and the set {D, E} corresponds to fault type 2. The fault sample subsets {A, B, C, F, G} and {D, E} form a known fault sample set, and G forms an unknown fault sample set.

[0085] At this point, an unknown fault sample set and a known fault sample set are obtained, and the known fault sample set is used as a fault database.

[0086] Step S103: record any sample in the unknown fault sample set as an unknown fault sample, and obtain the rationality of each K value within the preset K value range when classifying the unknown fault sample using the K nearest neighbor algorithm based on the distance difference between the unknown fault sample and the samples in the known fault sample set. Obtain the rationality of each K value corresponding to each unknown fault sample, update the unknown fault sample set and the known fault sample set, and obtain a new unknown fault sample set and a new known fault sample set.

[0087] Since the fault database is obtained based on a high degree of similarity, there may be some fault samples whose similarity is not as high as the similarity threshold but belong to the same type of fault and are not classified into the fault database. Therefore, it may not be reasonable to use the K-nearest neighbor algorithm to classify directly according to the fault database.

[0088] In order to solve the above problems, it is necessary to determine whether the fault database is reasonable and to modify it if it is unreasonable, so that the optimal classification result can be obtained when the K-nearest neighbor algorithm is used for classification based on the fault database.

[0089] Since when the fault database is unreasonable, no matter what the value of K in the K-nearest neighbor algorithm is, the final classification result will be unreasonable. Therefore, any sample in the unknown fault sample set can be recorded as an unknown fault sample. The rationality of the classification result when the unknown fault sample is classified by using the K-nearest neighbor algorithm can reflect the rationality of each K value when the K-nearest neighbor algorithm is used to classify the unknown fault sample. Then, according to the rationality of each K value corresponding to each unknown fault sample, whether the fault database is reasonable is judged. When the fault database is unreasonable, the fault database is modified.

[0090] Since the classification result of the K-nearest neighbor algorithm is affected by the selection of the K value, when the K value is selected, the classification result of the K-nearest neighbor algorithm is related to the number of fault samples of each fault type within the range corresponding to the K value, as well as the distance to the fault samples of each fault type. Therefore, based on the distance difference between the unknown fault sample and the samples in the known fault sample set, the rationality of each K value within the preset K value range when using the K-nearest neighbor algorithm to classify the unknown fault samples can be obtained.

[0091] Specifically, a scatter plot is constructed based on the overall abnormality degree and parameter correlation degree of all fault samples in the fault sample set, such as Figure 2 As shown, the horizontal axis of the scatter plot is the parameter correlation degree, and the vertical axis is the overall abnormality degree. Figure 2 The points of different shapes are samples of different fault types. The samples in the known fault sample set are recorded as known fault samples. In the scatter plot, the Euclidean distance between the unknown fault sample and each known fault sample is calculated, and all Euclidean distances are sorted in ascending order to obtain a distance sequence. The Euclidean distance belongs to the prior art and will not be described in detail here.

[0092] For any K value within the preset K value range, obtain the Kth distance in the distance sequence, and construct a first circular area in the scatter plot with the unknown fault sample as the center and the Kth distance as the radius;

[0093] According to the number of known fault samples in the first circular area, a proportion of known fault samples in the first circular area is obtained; the number of fault samples under each fault type in the first circular area is counted, and all fault samples under the fault type with the largest number are formed into a target set;

[0094] Obtaining the number of fault samples in the target set, calculating a fourth difference between a constant 1 and the reciprocal of the number of fault samples in the target set, respectively obtaining the Euclidean distance between the unknown fault sample and each fault sample in the target set, obtaining a corresponding distance mean, calculating the average of the fourth difference and the reciprocal of the distance mean, and obtaining a classification rationality coefficient for the unknown fault sample in the first circular area;

[0095] The rationality of any K value is obtained based on the average value between the proportion of known fault samples in the first circular area and the classification rationality coefficient.

[0096] In one embodiment, since the K-nearest neighbor algorithm follows the principle of majority rule in classification judgment, the value of K in the K-nearest neighbor algorithm is generally an odd number. The number of fault samples in each fault sample subset is counted, and the K value range corresponding to the unknown fault sample is set to (3, 5, 7, 9, ..., Q), where Q is less than the number of fault samples in the fault sample subset with the largest number of fault samples in the known fault set. Taking the mth K value corresponding to the unknown fault sample as an example, the calculation formula for the rationality of the mth K value is:

[0097]

[0098] Among them, μ m is the rationality of the mth K value; d' is the number of known fault samples in the first circular area; d is the number of all samples in the first circular area; max(d') is the number of fault samples in the target set; V max(d')r is the Euclidean distance between the unknown fault sample and the rth fault sample in the target set; r is the sequence number of the fault sample in the target set; 1 is a constant.

[0099] It should be noted that is the proportion of known fault samples in the first circular area, The smaller it is, the fewer known fault samples there are in the first circular area, and the more unknown fault samples there are. The more unreasonable the classification of unknown fault samples in the first circular area is, the smaller the rationality of the mth K value corresponding to the unknown fault samples is. Represents the classification rationality coefficient of unknown fault samples, The smaller it is, the farther the distance between the known fault samples and the unknown fault samples in the target set within the first circular area is, the more unreasonable the classification of the unknown fault samples in the first circular area is, and the less reasonable the mth K value corresponding to the unknown fault samples is.

[0100] According to the method for obtaining the rationality of the mth K value corresponding to the unknown fault sample, the rationality of each K value corresponding to the unknown fault sample is obtained. The rationality threshold is set to 0.4 according to the results obtained by experimental statistical calculation. There is no restriction here and it can be set according to the specific implementation scenario. When there are at least 3 K values ​​in the unknown fault sample whose rationality is less than or equal to 0.4, it means that it is unreasonable to use the K nearest neighbor algorithm to classify the unknown fault sample according to the fault database, where 3 is the first preset number, which is obtained according to experimental statistical calculation. There is no restriction this time and it can be set according to the specific implementation scenario. Further, the rationality of each K value corresponding to each unknown fault sample in the unknown fault sample set is obtained.

[0101] When there are no at least three unknown fault samples in the unknown fault sample set whose rationality of K values ​​is less than or equal to 0.4, it means that the classification of the unknown fault samples using the K nearest neighbor algorithm according to the fault database is reasonable, that is, the fault database is reasonable, so there is no need to modify the fault database, and the unknown fault sample set is used as a new unknown fault sample set, and the known fault sample set is used as a new known fault sample set.

[0102] When the classification of unknown fault samples in the unknown fault sample set is unreasonable, it means that the fault database is unreasonable. In this case, the fault database needs to be modified.

[0103] Since some fault samples whose similarity is not as high as the similarity threshold but belong to the same type of fault are not classified into the fault database, the classification of unknown fault samples using the K nearest neighbor algorithm based on the fault database is unreasonable. Therefore, in the unknown fault sample set, according to the rationality of each K value corresponding to each unknown fault sample, fault samples belonging to the same fault type can be obtained in the unknown fault sample set, and the unknown fault sample set and the known fault sample set are updated to obtain a new unknown fault sample set and a new known fault sample set. The specific method for updating the unknown fault sample set and the known fault sample set is as follows:

[0104] For any unknown fault sample in the unknown fault sample set, when the classification of any unknown fault sample is unreasonable, record the any unknown fault sample as a target unknown fault sample, and obtain all target unknown fault samples in the unknown fault sample set;

[0105] For any target unknown fault sample in the unknown fault sample set, obtain the minimum distance in the distance sequence, and construct a second circular area in the scatter plot with the any target unknown fault sample as the center and the minimum distance as the radius;

[0106] For any three unknown fault samples in the second circular area, if the difference in similarity between one of the unknown fault samples and the other two unknown fault samples is less than or equal to a preset similarity difference threshold, the any three unknown fault samples are formed into a new initial fault sample subset, all new initial fault sample subsets in the second circular area are obtained, two new initial fault sample subsets with intersection are divided into one category, and all categories are merged using a recursive algorithm to obtain at least one new fault sample subset, and one new fault sample subset corresponds to one fault type.

[0107] For example: Taking the wth target unknown fault sample as an example, the second circular area is constructed. The unknown fault samples in the second circular area form a set {I, J, K, L, M, N, O, P}, where I, J, K, L, M, N, O, P are unknown fault samples. For the three unknown fault samples I, J, and K, if the difference between the similarity between I and J and the similarity between I and K is less than or equal to 0.2, then the three unknown fault samples I, J, and K are combined into a new initial fault sample subset {I, J, K}. Similarly, we get To the new initial fault sample subsets {J, K, L} and {L, M, N}, the intersection of {I, J, K} and {J, K, L} is merged to obtain {I, J, K, L}, and all categories are merged using the recursive algorithm to obtain the new fault sample subset {I, J, K, L, M, N}. The new fault sample subset {I, J, K, L, M, N} corresponds to a new fault type, where 0.2 is the similarity difference threshold, which is calculated based on experimental statistics. There is no restriction here and it can be set according to the specific implementation scenario.

[0108] Furthermore, all new fault sample subsets are combined into a new known fault sample subset corresponding to any target unknown fault sample, a new known fault sample subset corresponding to each target unknown fault sample is obtained, all new known fault sample subsets are added to the known fault sample set to obtain an initial known fault sample set, and the fault samples in the unknown fault sample set that are the same as all new known fault sample subsets are eliminated to obtain an initial unknown fault sample set.

[0109] When there is no target unknown fault sample in the initial unknown fault sample set, the initial known fault sample set is used as a new known fault sample set, and the initial unknown fault sample set is used as a new unknown fault sample set.

[0110] When there is a target unknown fault sample in the initial unknown fault sample set, any target unknown fault sample in the initial fault sample set is recorded as a new target unknown fault sample, and the third circular area corresponding to the new target unknown fault sample is obtained. According to the method for obtaining a new fault sample subset, a new known fault sample subset corresponding to the new target unknown fault sample is obtained in the third circular area, and a new known fault sample subset corresponding to each new target unknown fault sample in the initial fault sample set is obtained. All new known fault sample subsets are added to the initial known fault sample set to obtain a new initial known fault sample set, and the fault samples in the initial unknown fault sample set that are the same as all new known fault sample subsets are eliminated to obtain a new initial unknown fault sample set. The method for obtaining a new known fault sample subset is repeated until there is no target unknown fault sample in the initial unknown fault sample set, and a new unknown fault sample set and a new known fault sample set are obtained.

[0111] At this point, a new unknown fault sample set and a new known fault sample set are obtained, and the new known fault sample set is used as a new fault database.

[0112] Step S104, according to the distance difference between each sample in the new unknown fault sample set and the samples in the new known fault sample set, respectively obtain the rationality of each K value corresponding to each sample, and according to the rationality of each K value corresponding to each sample, respectively obtain the optimal K value of each sample.

[0113] Since the K value in the K-nearest neighbor algorithm is artificially defined, the choice of K value will have a significant impact on the results of the K-nearest neighbor algorithm. If the K value is too small, the classification results may be affected by a small amount of data, resulting in inaccurate classification. If the K value is too large, the data volume may be too large and effective distinction cannot be made, resulting in inaccurate classification of medical device failure problems. Therefore, it is necessary to obtain an optimal K value for classifying real-time samples.

[0114] Because each K value has its rationality, the rationality can reflect whether the current K value meets the optimal result of classification. Therefore, for any sample in the new unknown fault sample set, according to the rationality of each K value corresponding to any sample, the K value with the largest rationality is taken as the optimal K value of any sample, and then the optimal K value is obtained based on the optimal K value of each sample.

[0115] At this point, the optimal K value for each sample in the new unknown fault sample set is obtained.

[0116] Step S105, obtain the optimal K value according to the accuracy of classification for each optimal K value, use the optimal K value as the value of K in the K nearest neighbor algorithm according to the fault-free sample set and the new known fault sample set, and classify each sample monitored in real time.

[0117] Since each sample in the new unknown fault sample set has its corresponding optimal K value, it is necessary to select an optimal K value from all the optimal K values ​​as the optimal K value corresponding to this type of medical device to classify subsequent real-time samples of this type of medical device.

[0118] Since the fault samples in the new known fault sample set all have their corresponding fault types, each optimal K value can be used to classify the fault samples of different fault types in the new known fault sample set to obtain the classification accuracy. According to the accuracy corresponding to each optimal K value, the optimal K value is obtained for classifying real-time samples.

[0119] Since there may be a large number of fault samples of a certain fault type, which affects the accuracy of classification for each optimal K value, 10 fault samples are selected from the fault samples corresponding to each fault type to form a comparison fault sample set. 10 is a second preset number, which is obtained based on experimental statistical calculations and is not limited here. It can be set according to the specific implementation scenario. The fault type of each fault sample in the comparison fault sample set is recorded as the initial fault type, and the optimal K value of each sample in the new unknown fault sample set is used to form an optimal K value set.

[0120] Further, for any optimal K value in the optimal K value set, any optimal K value is recorded as a target K value. When the result of classification of any fault sample in the comparison fault sample set using the target K value is the same as its initial fault type, it is confirmed that the result of classification of any sample in the comparison fault sample set using the target K value is accurate. In the comparison fault sample set, the number of samples with accurate results when classified using the target K value is obtained, recorded as the number of correct samples. The ratio of the number of correct samples to the number of all fault samples in the comparison fault sample set is calculated to obtain the accuracy rate when each fault sample in the comparison fault sample set is classified using the target K value.

[0121] According to the accuracy of classifying each fault sample in the comparison fault sample set according to the target K value, the accuracy corresponding to each optimal K value in the optimal K value set is obtained respectively, and the optimal K value corresponding to the maximum accuracy is used as the optimal K value.

[0122] After obtaining the optimal K value, the real-time samples of the same type of medical devices corresponding to the optimal K value can be classified. Since the real-time samples may not have faults, the set of fault-free samples and the set of new known fault samples are combined into a reference database. According to the reference database, the optimal K value is used as the K value in the K nearest neighbor algorithm to classify the real-time samples of the same type of medical devices to obtain accurate classification results. When the classification result of the real-time sample is a fault, corresponding measures are taken according to the corresponding fault type, such as repair or replacement.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A digital medical device quality data intelligent management platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the following method is implemented: For any type of medical device, the multi-dimensional historical parameter data of each medical device at each sampling moment is taken as a sample to obtain a sample set; Classifying the samples in the sample set to obtain a fault sample set and a fault-free sample set, and dividing the fault sample set into an unknown fault sample set and a known fault sample set according to the similarity between the samples in the fault sample set; Record any sample in the unknown fault sample set as an unknown fault sample, and obtain the rationality of each K value within a preset K value range when classifying the unknown fault sample using the K nearest neighbor algorithm based on the distance difference between the unknown fault sample and the samples in the known fault sample set, obtain the rationality of each K value corresponding to each unknown fault sample, and update the unknown fault sample set and the known fault sample set to obtain a new unknown fault sample set and a new known fault sample set; Obtaining the rationality of each K value corresponding to each sample based on the distance difference between each sample in the new unknown fault sample set and the samples in the new known fault sample set, and obtaining the optimal K value for each sample based on the rationality of each K value corresponding to each sample; Obtaining an optimal K value based on the accuracy of classification for each optimal K value, and using the optimal K value as the value of K in the K-nearest neighbor algorithm to classify each sample monitored in real time based on the set of fault-free samples and the set of newly known faulty samples; The classifying the samples in the sample set to obtain a fault sample set and a non-fault sample set includes: Record any sample in the sample set as a target sample, and obtain samples from the sample set that belong to the same medical device as the target sample to form a reference sample set; For any parameter data in the target sample, parameter data belonging to the same parameter as the any parameter data is obtained from the reference sample set, and the parameter data and the any parameter data form a parameter data sequence. In the parameter data sequence, a target window of a preset size is constructed with the any parameter data as the center. The preset size is used as a sliding step, and the target window is slid to the left to obtain a first window, and the target window is slid to the right to obtain a second window. Obtaining the variances of the parameter data in the target window, the first window, and the second window respectively, obtaining the mean of the variances of the parameter data between the first window and the second window, obtaining the absolute value of the difference between the variance of the parameter data in the target window and the mean, obtaining the degree of fluctuation of any parameter data, obtaining the sum of the degree of fluctuation and a constant 1, and obtaining a change difference value of any parameter data based on the difference between the constant 1 and the reciprocal of the sum; Obtaining a parameter range of any parameter data; if any parameter data is not within the parameter range, or a change difference value of any parameter data is greater than or equal to a preset change difference threshold, determining that any parameter data is faulty parameter data; traversing each parameter data in the target sample; and if at least one faulty parameter data exists in the target sample, determining that the target sample is a faulty sample; All fault samples in the sample set are acquired to form a fault sample set, and non-fault samples in the sample set are acquired to form a non-fault sample set.

2. A digital medical device quality data intelligent management platform according to claim 1, characterized in that: The step of dividing the fault sample set into an unknown fault sample set and a known fault sample set according to the similarity between samples in the fault sample set includes: For any fault sample in the fault sample set, based on the differences and fluctuation characteristics between the parameter data of the any fault sample, the overall abnormality degree of the any fault sample is obtained; based on the correlation between the parameter data of the any fault sample, the parameter correlation degree of the any fault sample is obtained; Obtaining the overall abnormality degree and parameter correlation degree of each fault sample in the fault sample set, taking the fault samples in the fault sample set other than the any one fault sample as other fault samples, and obtaining the similarity between the any one fault sample and each other fault sample based on the difference in the overall abnormality degree and the difference in the parameter correlation degree between the any one fault sample and each other fault sample; The other fault samples corresponding to the similarity greater than or equal to the preset similarity threshold and the any fault sample form an initial fault sample subset corresponding to the any fault sample, obtaining an initial fault sample subset corresponding to each fault sample in the fault sample set, dividing two initial fault sample subsets with an intersection into one category, and merging all categories using a recursive algorithm to obtain at least one fault sample subset, where one fault sample subset corresponds to one fault type; All fault sample subsets are grouped into a known fault sample set, and fault samples in the fault sample set that do not belong to the known fault sample set are grouped into an unknown fault sample set.

3. A digital medical device quality data intelligent management platform according to claim 2, characterized in that: The step of obtaining the overall abnormality degree of any fault sample according to the differences and fluctuation characteristics between the parameter data of any fault sample includes: Recording any parameter data in any fault sample as target parameter data, obtaining a parameter range of the target parameter data, obtaining a maximum value and a minimum value within the parameter range of the target parameter data, and obtaining a second difference between the maximum value and the minimum value; When the target parameter data is less than the minimum value, obtaining a first difference between the minimum value and the target parameter data, and obtaining a numerical difference value of the target parameter data according to a ratio of the first difference to the second difference; When the target parameter data is greater than the maximum value, obtaining a third difference between the target parameter data and the maximum value, and obtaining a numerical difference value of the target parameter according to a ratio of the third difference to the second difference; When the target parameter data is within the parameter range of the target parameter data, setting the numerical difference value of the target parameter to a preset value; The abnormality degree of the target parameter data is obtained based on the average value between the numerical difference value and the change difference value of the target parameter data, and the overall abnormality degree of any fault sample is obtained based on the sum of the abnormality degrees of each parameter data in any fault sample.

4. A digital medical device quality data intelligent management platform according to claim 2, characterized in that: The obtaining of the parameter correlation degree of any fault sample according to the correlation degree between the parameter data of any fault sample includes: For any dimensional parameter of any type of medical device, obtain a fault sample whose parameter data corresponding to the any dimensional parameter is the fault parameter data in the fault sample set as the fault sample mapping set of the any dimensional parameter; obtain the association probability of each dimensional parameter with other dimensional parameters according to the intersection of the fault sample mapping sets of all dimensional parameters, and obtain the association probability sequence corresponding to each dimensional parameter; If there are at least two fault parameter data in any fault sample, the dimension parameters corresponding to the fault parameter data in any fault sample are recorded as associated parameters, and the proportion of associated parameters in any fault sample is obtained according to the number of associated parameters in the any fault sample; If there are less than two fault parameter data in any fault sample, the proportion of associated parameters in any fault sample is 0; In the association probability sequence corresponding to each dimensional parameter of any fault sample, the association probability of any two association parameters in any fault sample is obtained, and the corresponding association probability cumulative value is obtained; The parameter correlation degree of any fault sample is obtained according to the product between the correlation parameter proportion and the cumulative value of the correlation probability.

5. A digital medical device quality data intelligent management platform according to claim 2, characterized in that: The obtaining of the similarity between any fault sample and each other fault sample according to the overall abnormality difference and parameter correlation difference between any fault sample and each other fault sample includes: For any other fault sample, obtain the absolute value of the difference between the overall abnormality degree of the any fault sample and the any other fault sample to obtain the overall abnormality degree difference, obtain the inverse of the sum of the overall abnormality degree difference and a constant 1, and obtain the overall abnormality degree similarity value between the any fault sample and the any other fault sample; Obtaining an absolute value of a difference between the parameter correlation degrees of any one fault sample and any other fault sample to obtain a parameter correlation degree difference, obtaining a reciprocal of a sum of the parameter correlation degree difference and a constant 1 to obtain a parameter correlation degree similarity value between any one fault sample and any other fault sample; The similarity between the fault sample and any other fault sample is obtained according to the average of the overall abnormality similarity value and the parameter correlation similarity value between the fault sample and any other fault sample.

6. A digital medical device quality data intelligent management platform according to claim 2, characterized in that: The obtaining, based on the distance difference between the unknown fault sample and the samples in the known fault sample set, the rationality of each K value within a preset K value range when classifying the unknown fault sample using the K nearest neighbor algorithm, includes: Constructing a scatter plot based on the overall abnormality degree and parameter correlation degree of all fault samples in the fault sample set, wherein the abscissa of the scatter plot is the parameter correlation degree and the ordinate is the overall abnormality degree. In the scatter plot, calculating the Euclidean distance between the unknown fault sample and each known fault sample, sorting all Euclidean distances in ascending order to obtain a distance sequence, wherein the known fault sample refers to a sample in the known fault sample set; For any K value within the preset K value range, obtain the Kth distance in the distance sequence, and construct a first circular area in the scatter plot with the unknown fault sample as the center and the Kth distance as the radius; According to the number of known fault samples in the first circular area, a proportion of known fault samples in the first circular area is obtained; the number of fault samples under each fault type in the first circular area is counted, and all fault samples under the fault type with the largest number are formed into a target set; Obtaining the number of fault samples in the target set, calculating a fourth difference between a constant 1 and the reciprocal of the number of fault samples in the target set, respectively obtaining the Euclidean distance between the unknown fault sample and each fault sample in the target set, obtaining a corresponding distance mean, calculating the average of the fourth difference and the reciprocal of the distance mean, and obtaining a classification rationality coefficient for the unknown fault sample in the first circular area; The rationality of any K value is obtained based on the average value between the proportion of known fault samples in the first circular area and the classification rationality coefficient.

7. A digital medical device quality data intelligent management platform according to claim 6, characterized in that: The updating of the unknown fault sample set and the known fault sample set to obtain a new unknown fault sample set and a new known fault sample set includes: For any unknown fault sample in the unknown fault sample set, when there are at least a first preset number of K values ​​in the any unknown fault sample whose rationality is less than or equal to a preset rationality threshold, the any unknown fault sample is recorded as a target unknown fault sample, and all target unknown fault samples in the unknown fault sample set are obtained; For any target unknown fault sample in the unknown fault sample set, obtain the minimum distance in the distance sequence, and construct a second circular area in the scatter plot with the any target unknown fault sample as the center and the minimum distance as the radius; For any three unknown fault samples in the second circular area, if the difference in similarity between one of the three unknown fault samples and the other two unknown fault samples is less than or equal to a preset similarity difference threshold, the three unknown fault samples are combined into a new initial fault sample subset, all new initial fault sample subsets in the second circular area are obtained, two new initial fault sample subsets with an intersection are divided into one category, and all categories are merged using a recursive algorithm to obtain at least one new fault sample subset, where each new fault sample subset corresponds to one fault type; All new fault sample subsets are combined into a new known fault sample subset corresponding to any target unknown fault sample, a new known fault sample subset corresponding to each target unknown fault sample is obtained, all new known fault sample subsets are added to the known fault sample set to obtain an initial known fault sample set, and the fault samples in the unknown fault sample set that are the same as those in all new known fault sample subsets are removed to obtain an initial unknown fault sample set; When there is no target unknown fault sample in the initial unknown fault sample set, taking the initial known fault sample set as a new known fault sample set and taking the initial unknown fault sample set as a new unknown fault sample set; When the target unknown fault sample exists in the initial unknown fault sample set, repeating the method for obtaining the new known fault sample subset until the target unknown fault sample does not exist in the initial unknown fault sample set, thereby obtaining a new unknown fault sample set and a new known fault sample set; If the target unknown fault sample does not exist in the unknown fault sample set, the unknown fault sample set is used as a new unknown fault sample set, and the known fault sample set is used as a new known fault sample set.

8. A digital medical device quality data intelligent management platform according to claim 1, characterized in that: Obtaining the optimal K value for each sample based on the rationality of each K value corresponding to each sample includes: For any sample in the new unknown fault sample set, based on the rationality of each K value corresponding to the any sample, the K value with the greatest rationality is taken as the optimal K value for the any sample.

9. A digital medical device quality data intelligent management platform according to claim 7, characterized in that: The method of obtaining the optimal K value according to the accuracy of classification performed on each optimal K value includes: Selecting a second preset number of fault samples from the fault samples corresponding to each fault type to form a comparison fault sample set, and forming an optimal K value set from the best K value of each sample in the new unknown fault sample set; For any optimal K value in the optimal K value set, obtaining an accuracy rate when classifying each fault sample in the comparison fault sample set using the any optimal K value; Obtain the accuracy corresponding to each optimal K value in the optimal K value set, and take the optimal K value corresponding to the maximum accuracy as the optimal K value.

Images