A machine learning-based medical device failure warning method

Abstract

The application discloses a kind of medical equipment failure early warning method based on machine learning, it is related to abnormal detection technical field.Method includes the following steps: obtaining temperature, voltage and load disturbance data in electromagnetic coil cycle, extract first and last section trend to generate feature set, identify magnet vibration and current peak segment, classify disturbance event, track cooling pump efficiency and signal-to-noise ratio change to classify drift behavior, count the frequency of consistent combination in direction within cycle, mark synchronous offset and disturbance coexistence cycle to generate failure warning result.In the application, by segmenting temperature, voltage and load disturbance data and comparing trends, the feature extraction accuracy is improved, the disturbance recognition ability is enhanced by vibration frequency and current peak time sequence classification, combined with cooling efficiency and signal-to-noise ratio dynamic tracking, sensitive judgment of operating state change is realized, consistent combination statistics in fluctuation sequence reveals synergistic relationship, and failure prediction effect is strengthened.

Description

Technical Field

[0001] This invention relates to the field of anomaly detection technology, and in particular to a machine learning-based method for early warning of medical device malfunctions. Background Technology

[0002] The field of anomaly detection technology encompasses machine learning and data analysis techniques to identify irregular patterns or behaviors in data. These irregular patterns typically indicate deviations in system performance from normal operating conditions. The core of this field involves various algorithms and models, including statistical methods, clustering techniques, and neural networks, designed to automatically detect anomalies from large datasets. The entire technology field continuously improves the accuracy and efficiency of anomaly detection through algorithm optimization and model training to adapt to the application needs of different industries.

[0003] Among them, the machine learning-based medical equipment fault early warning method refers to using machine learning technology to predict and identify potential faults in medical equipment. This patent topic covers using specific machine learning models to analyze the operational data of medical equipment and identify abnormal patterns that may occur during equipment operation through training data. Specifically, it involves collecting operational data from medical equipment and then using a trained model to analyze the data in real time in order to identify signals that may lead to equipment failure at an early stage.

[0004] Traditional technologies rely on model analysis of overall cyclical data, lacking attention to details within critical time periods, thus failing to accurately capture early signals of anomalies. When handling disturbance identification, existing methods often rely on single indicators, failing to fully utilize the dynamic relationships between multiple variables, which limits the model's sensitivity and accuracy in handling complex changes. The lack of effective trend change and behavioral evolution analysis leads to overly simplistic interpretations of cyclical changes, easily resulting in misjudgments. For example, failure to respond promptly to long-term, subtle changes may lead to equipment instability or unexpected malfunctions. Furthermore, existing technologies perform poorly in adapting to novel disturbance patterns, especially under complex conditions such as equipment aging or environmental changes, where their early warning capabilities significantly decrease. These limitations stem from the failure to effectively analyze data and predict behavior from multiple dimensions and dynamic perspectives, posing challenges in the fault prevention and diagnosis of high-precision medical equipment. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention aims to provide a machine learning-based method for early warning of medical device malfunctions, the technical solution of which is as follows:

[0006] A machine learning-based method for early warning of medical device malfunctions includes the following steps:

[0007] S1: Acquire temperature, voltage and load disturbance data of electromagnetic coils used for medical imaging during the operating cycle, compare the trends at the beginning and end of the cycle in a directional manner, integrate the change trajectory, and obtain the cycle feature dataset.

[0008] S2: Based on the fluctuation cycle number in the periodic feature dataset, extract the magnet vibration frequency and current peak record, mark the start and end segments of the disturbance, identify the reversal and continuous peak segments, classify and summarize them in chronological order, and obtain a list of disturbance event segments;

[0009] S3: Based on the period corresponding item in the list of disturbance event segments, extract the heat exchange efficiency and signal-to-noise ratio data of the cooling pump, track the direction of continuous periodic change, and combine the time information to classify it into the period corresponding category to obtain the drift behavior classification table.

[0010] S4: Based on the period number of the drift behavior classification record, retrieve the temperature and voltage fluctuation sequence within the period, identify the combination content with the same direction, sort by period and count the combination frequency, and generate a behavior covariance correlation record table.

[0011] S5: Based on the prominent cycle numbers in the behavior covariance correlation record table, call the cycle features and disturbance content, analyze the synchronous offset disturbance state of multiple indicators, mark them as potential fault cycles, and generate equipment fault early warning results.

[0012] As a further embodiment of the present invention, the periodic feature dataset includes stage distribution information, temperature direction identifier, voltage duration segment status, and load disturbance frequency label; the disturbance event segment list includes vibration reversal segment, current peak change sequence, and time sequence index; the drift behavior classification table includes trend classification category, period number correspondence, and signal change direction attribute; the behavior covariance correlation record table includes combination direction matching result, intra-period change consistency label, and cross-period combination frequency value; and the equipment fault early warning result includes abnormal period number, intra-period state combination, and period label structure item.

[0013] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0014] S101: Acquire the temperature change, voltage fluctuation and load disturbance performance of the electromagnetic coil assembly for medical imaging during continuous operation cycles, extract the original records of the beginning, middle and end segments of each cycle, divide and organize them into stage recording units of the cycle according to the segments, and obtain the original data group of the operation stage.

[0015] S102: Based on the original data set of the operation phase, analyze the direction of temperature change between the first and last segments, mark the continuous segments of voltage in multiple segments, record the location of load disturbances within the time period, summarize the change characteristics in the cycle, and obtain the set of changes descriptions within the segment.

[0016] S103: Based on the set of changes within the segment, integrate the temperature, voltage and disturbance changes corresponding to each cycle, generate corresponding marked entries by associating the cycle number, and organize all entries into a unified record structure in chronological order to obtain the cycle feature dataset.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the fluctuation cycle number identified in the periodic feature dataset, obtain the vibration frequency and current peak value record of the magnet vibration control component in the corresponding cycle, extract the continuous signal segment in the cycle and divide it into segmented sequences according to time to obtain the periodic signal segmented sequence set.

[0019] S202: Based on the vibration frequency change trajectory of each segment in the periodic signal segment sequence set, identify the position of direction reversal and mark the start and end points, record the continuous change of current peak value in the corresponding segment, calculate the number of direction reversals and the number of current peak value changes in the segment, and obtain a set of statistical indicators of disturbance behavior.

[0020] S203: Call the segment numbers with continuous reversal and multiple current fluctuations in the set of disturbance behavior statistics indicators, classify and organize them according to time sequence and period, and archive them uniformly as periodic segment records with disturbance characteristics to obtain a list of disturbance event segments.

[0021] As a further aspect of the present invention, the calculation formula for the number of direction reversals and the number of current peak value changes within the segment is specifically as follows:

[0022] ;

[0023] in, This represents the number of direction reversals and the number of times the peak current changes within a segment. This represents the peak current in the nth period signal segment. This represents the peak current value in the selected reference segment of the nth cycle signal segment. This represents the vibration frequency in the nth periodic signal segment. This represents the vibration frequency in the reference segment matched by the nth periodic signal segment. This represents the sum of squared deviations of all current peak value changes in the current paragraph. This represents the total number of direction reversals identified within paragraph x. This represents the mean of all vibration frequencies within the x-th segment. This represents the median of the current peak sequence within the x-th segment. This represents the sum of squares of the differences in peak current variations within the x-th segment. This represents the total number of periodic signal segments in the current paragraph.

[0024] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0025] S301: Based on the corresponding period number in the list of disturbance event segments, extract the heat exchange efficiency and electrical signal-to-noise ratio records of the cooling pump module within the period, sort them by time and organize the data structure according to the period, and obtain the period signal archive data group.

[0026] S302: Call the heat exchange efficiency sequence of each period in the periodic signal archive data group, identify the direction of numerical change in the segment, count the total number of consecutive upward segments in the sequence, calculate the proportion of consecutive upward segments, and obtain the trend upward feature set.

[0027] S303: Based on the change characteristics corresponding to the upward trend feature set, and combined with the attribution cycle number and the time position of the disturbance event, the trend structure and cycle information are combined and classified to obtain a drift behavior classification table.

[0028] As a further aspect of the present invention, the formula for calculating the proportion of the continuous upward paragraphs is as follows:

[0029] ;

[0030] in, This represents the proportion of consecutive upward-moving paragraphs. Represents all that satisfy The sum of the differences in heat exchange efficiency under different conditions. This represents the sum of the absolute values ​​of the heat exchange efficiency differences across all cycles. This represents the average completeness rate of all upstream sampling points. Represents the maximum number of consecutive upward cycles. This represents the average of the extreme fluctuations in heat exchange efficiency measured across all upward segments. This represents the average magnitude of the difference in heat exchange efficiency across different cycles. This represents the average difference between the current archiving period and the value set in the previous period. This represents the average frequency of environmental disturbances per unit time in the current data set.

[0031] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0032] S401: Call the records in the drift behavior classification table that are classified as upward or downward, extract the temperature offset data and voltage fluctuation sequence within the corresponding period, organize them into periodic records of two channels in chronological order, and mark the direction of change for each segment to obtain the periodic change direction data group.

[0033] S402: Based on the temperature and voltage change direction labeling content in the periodic change direction data group, filter out the combination segments with the same direction in the same period, calculate the total number of times the combination segments with the same direction appear in all periods, and store them in association with the period number to obtain the frequency value of the combination segments with the same direction.

[0034] S403: Based on the periodic distribution of the frequency values ​​corresponding to the direction-consistent combination, the data is categorized and organized into a set structure with periodic numbers, combination features and frequency indicators, and uniformly archived as periodic behavior coordination information to obtain a behavior covariance correlation record table.

[0035] As a further aspect of the present invention, the formula for calculating the total number of occurrences of the direction-consistent combination segment in the entire cycle is as follows:

[0036] ;

[0037] in, This represents the total number of times that a combination of elements with the same direction appears throughout the entire cycle. Representing the Temperature change within the period Representing the Voltage change during the period Representing the The dimensionless ratio parameter of the temperature-electricity variation amplitude within the period. Represents the smallest positive number approaching zero. This represents the total number of cycles.

[0038] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0039] S501: Based on the prominent cycle numbers in the behavior covariance correlation record table, call the corresponding content of the cycle in the cycle feature dataset and the list of disturbance event segments, extract the disturbance trajectory information and state change content of the cycle, and organize them into corresponding entries according to the cycle number to obtain the cycle state correlation record set.

[0040] S502: Based on each periodic record in the periodic state association record set, screen for the co-occurrence of continuous disturbance segments and multiple synchronous state changes, mark the period numbers with combination characteristics, and summarize them into the abnormal state list in chronological order to obtain the abnormal periodic state list.

[0041] S503: Call the cycle number in the abnormal cycle status list, combine the temperature change, voltage fluctuation and vibration performance status within the cycle, identify abnormal features and classify them, and obtain equipment fault warning results.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0043] In this embodiment of the invention, the accuracy of feature extraction is improved by collecting temperature, voltage and load disturbance data in segments and comparing trends. The disturbance identification capability is enhanced by classifying vibration frequency and current peak value in time sequence. Combined with dynamic tracking of cooling efficiency and signal-to-noise ratio, sensitive judgment of changes in operating status is achieved. The statistical analysis of consistent combinations in the fluctuation sequence reveals the synergistic relationship, which strengthens the fault prediction effect. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0046] Please see Figure 1 A machine learning-based method for early warning of medical device malfunctions includes the following steps:

[0047] S1: Acquire the temperature changes, voltage fluctuations and load disturbances of the electromagnetic coil assembly for medical imaging during continuous operation cycles, extract the recorded data of the beginning, middle and end segments of each cycle, compare the temperature shift direction between the first and last segments, extract the changing trends of the voltage maintenance interval length and the number of load disturbances between each segment, integrate the changing direction information of the monitored content in the cycle, and obtain the cycle feature dataset.

[0048] S2: Based on the period number of the periodic feature data with significant fluctuations, extract the vibration frequency and current peak value records of the magnet vibration control component in the corresponding period, mark the start and end positions of continuous vibration changes, identify the vibration reversal action and the continuous rise process of current peak value between differentiated segments, filter segments with continuous disturbance characteristics, classify and summarize them in chronological order, and obtain a list of disturbance event segments.

[0049] S3: Based on the period correspondence in the list of disturbance event segments, extract the heat exchange efficiency and electrical signal-to-noise ratio recording sequence of the cooling pump module in the corresponding period, track the change of numerical direction in continuous periods, classify it as continuous upward, stable or continuous downward, and combine it with time information to classify it into the corresponding period category to obtain the drift behavior classification table.

[0050] S4: Call the records that have been classified into the upward or downward categories in the drift behavior classification table, retrieve the temperature offset performance and voltage fluctuation sequence in the corresponding period, compare the change direction of the data in the period, identify the combination content with the same direction, sort out the behavior combination list in order, and count the frequency of the class combination in all periods to obtain the behavior covariance correlation record table.

[0051] S5: Based on the period number that stands out in the behavior covariance correlation record table, the contents of the period in the period feature dataset and the disturbance event fragment list are called, and the feature records that have both continuous disturbance trajectory and multiple synchronous changes are screened. The periods that meet the conditions are marked as abnormal periods. The structured output is generated by combining the number and the monitoring status to obtain the equipment fault warning result.

[0052] The periodic feature dataset includes stage distribution information, temperature direction identifier, voltage duration segment status, and load disturbance frequency label. The disturbance event segment list includes vibration reversal segments, current peak change sequences, and time sequence indexes. The drift behavior classification table includes trend classification categories, cycle number correspondence, and signal change direction attributes. The behavior covariance correlation record table includes combination direction matching results, intra-cycle change consistency annotations, and cross-cycle combination frequency values. The equipment fault warning results include abnormal cycle numbers, intra-cycle state combinations, and cycle annotation structure items.

[0053] The specific steps of S1 are as follows:

[0054] S101: Acquire the temperature change, voltage fluctuation and load disturbance performance of the electromagnetic coil assembly for medical imaging during continuous operation cycles, extract the original records of the beginning, middle and end segments of each cycle, divide and organize them into stage recording units of the cycle according to the segments, and obtain the original data group of the operation stage.

[0055] First, clarify the cycle division standard. Typically, the operating control cycle set by the internal clock control module of the equipment is used as the benchmark. Each start-up to the predetermined shutdown interval is defined as a complete cycle. For example, if the cycle is set to 180 seconds, then every 30 seconds is divided into a start segment, a middle segment, and a finish segment. The monitoring content within each segment is extracted accordingly. Temperature changes can be recorded using thermistor detection values. The sampling frequency is set to 1Hz, meaning one temperature reading is acquired per second, resulting in 180 temperature values ​​within a 180-second cycle. Voltage fluctuations are recorded based on the input and output terminal voltage changes collected in the electromagnetic excitation control circuit. For example, if the initial voltage is 220V, and the detected voltage is 210V at the 25th second, this is a downward fluctuation. Load disturbance can be indirectly deduced from the current changes in the coil combined with the electromagnetic force distribution. If a continuous current amplitude jump occurs between the 60th and 75th seconds, it is recorded as a load disturbance segment. After extracting the above data within each segment, the data in each segment needs to be independently identified and a mapping structure established. The segment is named D1, the middle segment D2, and the end segment D3, and all segments are uniformly associated with the corresponding cycle numbers P1, P2...Pn, forming a two-layer data structure of cycle and stage. In actual operation, if there is a continuous rise in temperature, stable voltage, and severe load disturbance in segment D1 corresponding to operating cycle P3, while there is a drop in temperature, frequent voltage fluctuations, and intermittent load disturbances in segment D3, then the complete data entries of these three segments are recorded in the original numerical form, and the corresponding timestamp index of the stage is retained. At the same time, data loss, jumps, and other phenomena in different segments are marked. For example, a temperature sampling interval greater than 3 seconds is considered a recording interruption, a voltage fluctuation exceeding ±15V is considered a boundary deviation, and a load disturbance record exceeding 3 times can be classified as a strong disturbance segment. The data in each segment are stored in a multi-dimensional structure in tabular form for subsequent retrieval. Finally, the information of each segment is summarized according to the cycle sequence and stage order to establish a combined data set containing P number, D segment, temperature record, voltage record, and disturbance identifier, thus obtaining the original data group of the operating stage.

[0056] S102: Based on the original data set of the operation phase, analyze the direction of temperature change between the first and last segments, mark the continuous segments of voltage in multiple segments, record the location of load disturbances within the time period, summarize the change characteristics in the cycle, and obtain the set of changes descriptions within the segment.

[0057] First, the structure of the three segments in each cycle is reconstructed. The direction of temperature change trajectories at the beginning and end of the segment is determined. If the temperature value in the beginning segment shows an upward trend and the temperature value in the end segment shows a downward trend, it is determined to be downward. If the temperature continues to rise between the two segments, it is recorded as upward. The determination of the temperature change trend is based on the direction of change between the first and last sampling points within the segment. For example, if the temperature of the beginning segment is 37.2℃ and the temperature of the end segment is 39.1℃, the temperature direction of this cycle is determined to be upward. If the temperature of the beginning segment is 38.5℃ and the temperature of the end segment is 37.6℃, it is downward. If the temperature difference is less than 0.3℃, it is considered a stable fluctuation segment. In actual scenarios, the temperature stability judgment interval can be set to ±0.3℃, which is the reference value for the temperature direction interval. Then, the performance of voltage fluctuation in multiple segments is labeled. The increasing and decreasing trends of the continuous sampling data of each voltage record are identified, and the start and end points of the maintenance segment are recorded. For example, if the continuous reading values ​​in a certain cycle are 218V and 219V respectively... If the voltage is 219V, 220V, 220V, 220V, and 219V, and the duration of 220V exceeds 10 seconds, then 220V is marked as the maintenance segment value. At the same time, the voltage stability duration T is recorded in this segment, and it is determined whether T exceeds the set voltage stability judgment threshold, usually 8 seconds as the judgment benchmark. If it meets the threshold, the segment is marked as a voltage maintenance segment. Then, the location of the load disturbance is recorded. The identification benchmark of the disturbance event is set according to the amplitude of the load disturbance signal. For example, the current jump segment with a fluctuation amplitude of more than 5A per second and lasting for more than 3 seconds is defined as the disturbance occurrence point. For example, the continuous amplitude abnormal current jump that occurs from the 22nd second to the 26th second is the disturbance segment. The start point and end point are recorded in this segment and associated with the corresponding cycle number to establish a mapping. Finally, the complete labeling information of temperature direction, voltage maintenance segment, and load disturbance segment is sorted out in each cycle structure, generating three types of change and their time interval labeling records for each cycle, and obtaining the segment change description set.

[0058] S103: Based on the intra-segment change description set, integrate the temperature, voltage and disturbance changes corresponding to each cycle, generate corresponding marked entries by associating cycle numbers, and organize all entries into a unified record structure in chronological order to obtain the cycle feature dataset;

[0059] First, the direction of temperature change is extracted from each record and categorized into three types: "rising," "falling," and "stable." For example, if the starting temperature in period P4 is 36.8℃ and the ending temperature is 37.9℃, the temperature change for that period is marked as "rising." If the ending temperature is 36.2℃, it is marked as "falling." When the difference is within ±0.3℃, it can be classified as "stable." This labeling result is written to the temperature change field. Next, the duration of voltage fluctuations within the same period is extracted. The behavior of the entire segment is classified according to the start and end times of the marked voltage maintenance segment. For example, if the voltage value in the middle of period P4 remains above 220V without significant fluctuations and lasts for more than 10 seconds, this voltage segment is marked as "continuous." If there are intermittent fluctuations, it is recorded as "non-continuous." The original voltage sequence index is retained for comparison. At the same time, the occurrence of load disturbances within the time period is extracted. The current location is determined, and the number of events is counted. If more than two disturbance segments appear in a period (a disturbance segment is defined by ≥3 jumps and amplitude ≥5A), the disturbance intensity of that period is marked as "high". If only one disturbance occurs or the disturbance intensity is between 2A and 4A, it is marked as "medium". If there is no disturbance, it is marked as "low". All three types of labels are bound to the period number to generate a single-period state combination entry. For example, the number P4 corresponds to the state "temperature rise - voltage continuity - high disturbance". Then, all entries are arranged and summarized according to the time sequence of the period number to generate a standard table format. Each row contains the period number, three state classification labels and the original record index position. For example: P4, temperature rise, voltage continuity, high disturbance, sequence [34–58]. Finally, all period entries are uniformly included in the structure record table to obtain the period feature dataset.

[0060] The specific steps of S2 are as follows:

[0061] S201: Based on the wave cycle number identified in the periodic feature dataset, obtain the vibration frequency and current peak record of the magnet vibration control component in the corresponding cycle, extract the continuous signal segment in the cycle and divide it into segmented sequences according to time to obtain the periodic signal segmented sequence set.

[0062] First, extract the operating data of the magnet vibration control component corresponding to each number. Vibration frequency recording can be achieved through a frequency sampling device connected to an accelerometer, typically sampling at 100Hz. Approximately 6000 frequency sample values ​​can be obtained within each operating cycle. Current peak value recording is achieved through a current transformer, recording the maximum instantaneous current value once per second to form a current peak value sequence. After acquisition, a raw data index table needs to be established for each cycle, corresponding to the data streams of the two signal channels. Then, continuous signal segments are divided based on the operating time. In practice, each segment duration can be set to 5 seconds, meaning that out of the 6000 frequency samples, every 300 samples constitute a segment. The division results are numbered sequentially; for example, the frequency segments in cycle number P1 are divided into P1-F1, P1-F2…P1-F20. The beginning and end time points and segment identifiers are retained within each segment, and a correspondence with the current peak values ​​is established. The system synchronously numbers each current peak segment within the cycle according to the same time interval, forming P1-I1, P1-I2...P1-I20. Channel data docking is performed for each pair of segments to ensure that each frequency segment and its corresponding current segment form a synchronous signal unit. For example, in cycle P1, P1-F4 represents the frequency data segment between the 16th and 20th seconds of operation, and P1-I4 represents the current peak segment in the same time period. Within the segment, it can be observed that the frequency gradually increases from 65Hz to 75Hz, and the current peak increases from 3.2A to 4.1A. This segment is determined to be an upward trend segment. All segmentation results retain the original sequence content, and index markers are set at the beginning and end of the segment for subsequent retrieval or review. After processing, the frequency segments and current peak segments of all cycles are uniformly recorded as a structured set, forming a dual-index data table of cycle number and segment number, and retaining the original sampled values ​​and timestamp structure to obtain the cycle signal segmentation sequence set.

[0063] S202: Based on the vibration frequency change trajectory of each segment in the periodic signal segmented sequence set, identify the position of direction reversal and mark the start and end points, record the continuous change of current peak value in the corresponding segment, calculate the number of direction reversals and the number of current peak value changes in the segment, and obtain a set of statistical indicators of disturbance behavior.

[0064] Formulas for the number of direction reversals and the number of current peak changes within a segment:

[0065] ;

[0066] in, This represents the number of direction reversals and the number of times the peak current changes within a segment. This represents the peak current in the nth period signal segment. This represents the peak current value in the selected reference segment of the nth cycle signal segment. This represents the vibration frequency in the nth periodic signal segment. This represents the vibration frequency in the reference segment matched by the nth periodic signal segment. This represents the sum of squared deviations of all current peak value changes in the current paragraph. This represents the total number of direction reversals identified within paragraph x. This represents the mean of all vibration frequencies within the x-th segment. This represents the median of the current peak sequence within the x-th segment. This represents the sum of squares of the differences in peak current variations within the x-th segment. This represents the total number of periodic signal segments in the current paragraph.

[0067] First, reasonable measured data needs to be set for calculation;

[0068] Numerical settings and data sources:

[0069] Suppose we have a set of peak current data:

[0070] Monitoring ampere;

[0071] Peak segment current ampere;

[0072] hertz;

[0073] Reference frequency hertz;

[0074] Ampere², hertz, hertz, ampere, ;

[0075] Formula calculation process:

[0076] Calculate the absolute value of the peak current difference and its product:

[0077] ;

[0078] Calculate the denominator:

[0079] ;

[0080] Calculate the final formula value:

[0081] ;

[0082] This result demonstrates the combined difference between the variations in peak current and vibration frequency and their corresponding reference values. This numerical result indicates that the changes in current and frequency, after standardization of the denominator, represent the average rate of change within the current segment. This data is used to further analyze a set of statistical indicators of disturbance behavior, reflecting the stability and volatility of current and vibration.

[0083] S203: Call the segment numbers with continuous reversal and multiple current fluctuations in the disturbance behavior statistical index set, classify and organize them according to time sequence and period, and archive them uniformly as periodic segment records with disturbance characteristics to obtain a list of disturbance event segments.

[0084] First, the disturbance index corresponding to each number needs to be extracted. The calculation process involves determining the input parameters, including the current peak record sequence within the period, the current difference sequence, the number of oscillation frequency direction reversals, the average frequency, and the number of segment samples. During execution, the current peak data is first extracted. For example, if the current sequence for a certain period is 3.2, 3.5, 3.1, 3.6, 3.0, the difference sequence is calculated, the absolute values ​​are taken, and the sum is obtained, resulting in a total current amplitude fluctuation of 1.1. The frequency sequence is 50, 52, 49, 53, with an average of 51Hz. The number of direction reversals is counted as 3. Combining this with the 4 segment sample points, the disturbance intensity is calculated using a pre-defined formula. The calculation yields a result of 0.1122 in the example. This value is compared with the preset disturbance detection threshold of 0.1. If the value is >0.1, the segment is identified as a disturbance segment and proceeds to the next screening step. Then, the vibration frequency change direction corresponding to the segment is traversed sequentially by number to confirm whether there are three or more trend reversals. For example, if the frequency rises from 50 to 52 (upward), then falls to 49 (downward), and then rises again to 53, constituting three consecutive reversals, it meets the reversal characteristic and enters the current fluctuation judgment process. The corresponding current peak sequence is then taken, and the presence of three or more peak points is determined by local maxima. The maximum value location is identified by comparing adjacent points. For example, if 3.5 is greater than 3.2 and 3.1, or 3.6 is greater than 3.1 and 3.0, the segment has two peak points and does not meet the fluctuation condition; therefore, this segment is not included in the final list. Conversely, if peaks such as 4.1, 4.3, and 4.6 are all detected, the condition is met and the segment enters the recording process. Finally, segments that meet the criteria are... Information on segments exceeding a set threshold and exhibiting both continuous reversal of vibration and multiple current fluctuations is collected. Recorded fields include period number, segment number, and... The values, start and end sampling points, number of reversals, and number of fluctuations are used to form a unified list of disturbance event fragments. These fragments are then sorted in ascending order according to their starting positions. This completes the entire data processing process of fragment identification, indicator quantification, and behavioral feature screening, resulting in the list of disturbance event fragments.

[0085] The specific steps for S3 are as follows:

[0086] S301: Based on the corresponding cycle number in the list of disturbance event segments, extract the heat exchange efficiency and electrical signal-to-noise ratio records of the cooling pump module within the cycle, sort them by time and organize the data structure according to the cycle to obtain the cycle signal archive data group.

[0087] First, the specific cycle number to which each disturbance segment belongs is located, and this number is used as the retrieval condition to extract the corresponding operational monitoring data of the cooling pump module within that cycle. Heat exchange efficiency data can be collected by monitoring the temperature difference and flow rate of the cooling water inlet and outlet. For example, temperature sensors embedded in the cooling water inlet and outlet pipes are used to record the temperature difference value, while the real-time flow rate of the flow sensor is read. If the cycle number is C07, and the recorded inlet temperature is 18.6℃ and the outlet temperature is 26.4℃, then the heat exchange temperature difference for this segment is 7.8℃. Combined with a flow rate of 32L / min, this segment's heat exchange efficiency is recorded as a high value. The electrical signal-to-noise ratio (SNR) data is obtained by acquiring the ratio between the voltage interference amplitude generated during equipment operation and the original signal amplitude using an embedded electromagnetic interference monitor. Within the same cycle, if the average original signal amplitude is 1.2V and the interference voltage is 0.09V, then the SNR for this segment is... The value is 13.3dB. Based on the calculation results, it is classified into the mid-to-high range. After extracting the heat exchange efficiency and signal-to-noise ratio, the records collected within the same period are sorted by timestamp. Using the unified sampling time of the equipment as the benchmark, all data are arranged in ascending order. For example, in the C07 period, the heat exchange efficiency is recorded once every 10 seconds, and the signal-to-noise ratio is recorded once every 5 seconds. After sorting, two aligned time series record segments are generated, forming dual-channel operating status information within the period. Then, the heat exchange and signal-to-noise ratio sequences are classified into the corresponding period structures, and a mapping index table is established. Record structure fields are uniformly output under each period, including period number, timestamp start and end, heat exchange reading, signal-to-noise ratio sequence ID, and sampling point index, etc. A complete record matrix is ​​generated through structure aggregation. Finally, all record segments of each period are summarized and sorted according to the period number to obtain the period signal archive data group.

[0088] S302: Call the heat exchange efficiency sequence of each cycle in the periodic signal archive data group, identify the direction of numerical change in the segment, count the total number of consecutive upward segments in the sequence, calculate the proportion of consecutive upward segments, and obtain the trend upward feature set.

[0089] The specific formula for calculating the proportion of consecutive upward paragraphs is as follows:

[0090] ;

[0091] in, This represents the proportion of consecutive upward-moving paragraphs. Represents all that satisfy The sum of the differences in heat exchange efficiency under different conditions. This represents the sum of the absolute values ​​of the heat exchange efficiency differences across all cycles. This represents the average completeness rate of all upstream sampling points. Represents the maximum number of consecutive upward cycles. This represents the average of the extreme fluctuations in heat exchange efficiency measured across all upward segments. This represents the average magnitude of the difference in heat exchange efficiency across different cycles. This represents the average difference between the current archiving period and the value set in the previous period. This represents the average frequency of environmental disturbances per unit time in the current data set.

[0092] Assume the collected values ​​are as follows:

[0093] =200 joules, =0.95 joules, =5 joules, =30 joules, =1000 joules, =20 joules, =5 joules, = 2 times / hour;

[0094] Formula calculation derivation process:

[0095] First, substitute the specific values ​​mentioned above into the formula to calculate:

[0096] ;

[0097] Next, perform the calculation:

[0098] ;

[0099] ;

[0100] ;

[0101] The results show that the proportion of continuously rising segments is 13.45%, which means that only 13.45% of the segments in the monitored period show a trend of continuous increase in heat exchange efficiency, reflecting that the current efficiency is in a state of slow change or fluctuation. This proportion is used to measure the frequency of performance improvement of the equipment in the operating cycle, and provides basic data support for judging the effect of operation adjustment, optimizing control indicators and identifying efficiency trends.

[0102] S303: Based on the change characteristics corresponding to the upward trend feature set, combined with the attribution cycle number and the time position of the disturbance event, the trend structure and cycle information are combined and classified to obtain the drift behavior classification table.

[0103] First, extract the number of consecutive upward segments marked in each period and their corresponding segment numbers. Then, compare them one by one with the archived start and end positions in the disturbance event segment records. Perform a bidirectional overlap judgment between the segment time interval marked in the trend features and the sampling index of the disturbance segment, identifying the start and end points. In this judgment operation, first extract the start index of the trend segment. With Endpoint Index Then extract the starting index of the corresponding perturbation segment. Determine the endpoint index b2. Is it less than or equal to? and Is it greater than or equal to? If the condition is met, it is marked as the trend and disturbance occurring at the same time, and the overlap relationship flag is set to 1. If the condition is not met, it is set to 0. After performing this judgment logic on the relationship between all trend segments and disturbance segments, the results are collected into a binary matching vector according to the period number. For the period number with a matching value of 1, the number of consecutive upward segments recorded in the corresponding upward trend feature is extracted, and combined with the ΔJ value in the disturbance segment, that is, the disturbance behavior identification index value, for joint recording. The ΔJ value is calculated by combining the current fluctuation amplitude and the number of reversals of the vibration frequency in each period, for example, the current peak sequence. The values ​​are 3.1, 3.5, 3.3, 3.8, and 4.0, and the vibration frequency sequence is 58, 62, 59, 64, and 61. First, the current fluctuation range is calculated to be 0.9. The vibration frequency direction reverses 3 times. ΔJ is taken as the fluctuation amplitude multiplied by the number of reversals, resulting in 2.7. The number of trend segments is set to 4, recorded by period number as C08. After performing this joint process on all period numbers with overlapping trend disturbance times, a triplet record set is constructed. Each set of data includes the period number, the number of trend segments, and the disturbance index value ΔJ. Then, the number of trend segments is further processed... The system categorizes periods into three segments: four or more segments are classified as continuous upward trends, two to three segments as short-term upward trends, and one or fewer segments as fluctuating trends. The threshold is set based on the frequency distribution of the number of trend segments across all periods, with the upper quartile serving as the reference line for continuous upward trends. For example, if four segments are located at the 75th percentile after sorting all periods, this is set as the dividing line. Trend level labels are written to the record field and linked to the period number. Further analysis is performed on the ΔJ indicator in each record, with a significant ΔJ disturbance benchmark set at 2.5. If ΔJ is greater than 2.5, a disturbance is indicated. Stronger disturbances are marked as weaker disturbances. This benchmark is set by adding one standard deviation to the mean ΔJ value of historical operating data. For example, if the mean ΔJ value is 1.7 and the standard deviation is 0.8, then 2.5 is the limit. After labeling, all records are sorted in ascending order by period number and summarized into a joint data table of trend and disturbance. The trend level and disturbance intensity in each record are combined and coded. For example, "continuous rise + strong disturbance" is marked as type A, and "short-term rise + weak disturbance" is marked as type C. Finally, a complete combination of trend structure and period information is generated to obtain a drift behavior classification table.

[0104] The specific steps of S4 are as follows:

[0105] S401: Call the records in the drift behavior classification table that are classified as upward or downward, extract the temperature offset data and voltage fluctuation sequence within the corresponding period, organize them into periodic records of two channels in chronological order, and mark the direction of change for each segment to obtain the periodic change direction data group.

[0106] First, the cycle number already classified as either upward or downward in the drift behavior classification table is used as the target cycle index. Temperature change data and voltage fluctuation data within each cycle are extracted sequentially. The temperature data comes from the temperature sequence formed by the thermal sensor sampling once per second during device operation. The voltage data is collected in real-time by the device's internal voltage monitoring module and arranged in ascending order by timestamp to form a voltage sequence. To further identify the synchronicity trend of temperature and voltage changes within a cycle, the two channels are divided into fixed-length continuous segments according to time, with each segment being a 10-second unit interval by default. The division results are uniformly named according to the cycle number. Examples of temperature channel segment numbers are C12–T1, C12–T2…C12–T18, and voltage channel segment numbers are C12–V1, C12–V2…C12–V18. The direction of change within each segment is determined by comparing the first and last values. If the last value is higher than the first value, it is labeled "upward"; if the last value is lower than the first value, it is labeled "downward". If the absolute value of the difference between the first and last values ​​is less than a preset stability threshold (e.g., temperature 0.3℃, voltage 2V), it is labeled "stable". For example, the temperature of segment C12–T7 starts at 36.5℃ and ends at 37.4℃, so it is labeled "upward"; the voltage of segment C12–V7 starts at 221V and ends at 217V, so it is labeled "downward". The direction labeling results for all segments are written to a structure record. Each record includes the period number, segment number, channel type, start and end sampling time, and direction label. To ensure the correspondence between the temperature and voltage channel labeling results within the same period, the labeling content of the two channels is aligned by time and then merged to obtain the periodic change direction data group.

[0107] S402: Based on the temperature and voltage change direction annotations in the periodic change direction data group, filter out the combination segments with the same direction within the same period, calculate the total number of times the combination segments with the same direction appear in all periods, and store them in association with the period number to obtain the frequency value of the combination segments with the same direction.

[0108] The formula for calculating the total number of occurrences of directional combination segments throughout the entire cycle is as follows:

[0109] ;

[0110] in, This represents the total number of times that a combination of elements with the same direction appears throughout the entire cycle. Representing the Temperature change within the period Representing the Voltage change during the period Representing the The dimensionless ratio parameter of the temperature-electricity variation amplitude within a period. Represents the smallest positive number approaching zero. Represents the total number of cycles;

[0111] Assumption:

[0112] Total number of cycles The sampling period is 3;

[0113] Period 1: Starting temperature is 298.1K, ending temperature is 301.4K, therefore... ;

[0114] Period 2: From 300.5K to 303.0K, we get ;

[0115] Period 3: From 297.0K to 300.0K, we get ;

[0116] Voltage change :

[0117] Cycle 1: Starting voltage 12.05V, ending voltage 12.40V. ;

[0118] Cycle 2: 12.20V to 12.50V, ;

[0119] Cycle 3: 11.95V to 12.25V, ;

[0120] Amplitude ratio parameter :

[0121] Period 1: ;

[0122] Period 2: ;

[0123] Period 3: ;

[0124] Normalization correction constant Set as ;

[0125] Actual numerical substitution and calculation process:

[0126] Calculation of Period 1:

[0127] molecular: ;

[0128] Denominator: ;

[0129] result: ;

[0130] Calculation of period 2:

[0131] molecular: ;

[0132] Denominator: ;

[0133] result: ;

[0134] Calculation of period 3:

[0135] molecular: ;

[0136] Denominator: ;

[0137] result: ;

[0138] Total weighted frequency value:

[0139] ;

[0140] The results indicate that the temperature and voltage changes show a good synchronous trend within the three defined periods, with a calculated weighted directional consistency frequency of 2.8726. This is higher than the set baseline of 1.8, suggesting that the current data segment contains multiple periodic segments with highly matched amplitude directions.

[0141] S403: Based on the periodic distribution of the frequency values ​​of the direction-consistent combination, the data is classified and organized into a set structure with periodic numbers, combination features and frequency indicators, and uniformly archived as periodic behavior coordination information to obtain a behavior covariance correlation record table.

[0142] First, it needs to be clarified that "directionally consistent combination" refers to the consistent direction of change exhibited by different monitoring channels, such as temperature and voltage channels, within the same time slice within a cycle. For example, if temperature segment T5 shows an upward trend within a certain cycle, and voltage segment V5 also shows an upward trend, this combination is counted as one instance of directional consistency. The frequency value of directionally consistent combinations is thus accumulated. Based on the frequency of directional consistency in each cycle from the statistical results, a cycle number is extracted and mapped one-to-one with the segment combination relationship to which this number belongs. The segment group numbers of directional consistency in each record are collected and processed. At the same time, clear classification labels are assigned to different types of combination characteristics. For example, simultaneous upward movement of temperature and voltage is marked as "ascending combination," and a decrease in temperature and a decrease in voltage is marked as "decreasing combination." If one increases and the other decreases, it is marked as "dissimilar combination." All combination relationships are classified through classification labels to form a cycle. A preliminary association table between the number and the combination features is created. Then, the cumulative frequency of different types of combination features in each period is summarized. For example, if there are 3 instances of "ascending co-co ...

[0143] The specific steps of S5 are as follows:

[0144] S501: Based on the period number that stands out in the behavior covariance correlation record table, call the corresponding content of the period in the period feature dataset and the list of disturbance event segments, extract the disturbance trajectory information and state change content of the period, and organize them into corresponding entries according to the period number to obtain the period state correlation record set.

[0145] First, the frequency values ​​of consistent directional combinations recorded in the behavioral covariance correlation record table need to be sorted and screened. Several period numbers with significantly higher-than-average frequency of consistent directional combinations in the overall period set are selected as target periods. These are used as indexes to retrieve the corresponding period's state description entries from the period feature dataset. The three state elements of temperature change trend, voltage trend performance, and load disturbance level are extracted and combined to generate a unified period state structure record. For example, when the period number is P11, if the three corresponding identifiers in the period feature dataset are "temperature rising", "voltage continuing", and "disturbance strong", then the period state combination can be expressed as "rising-stable-strong". Subsequently, based on the same period number, the disturbance event segment list is searched to extract the main disturbance segment information identified under that period. The content includes segment number, number of vibration reversals, number of current fluctuation segments, and segment start and end time index. For example, the segment number corresponding to P11 is F05, recorded as 3 vibration reversals and 4 current fluctuation segments, with a segment time range of 62 to 76 seconds. The aforementioned periodic state information and disturbance index content will be merged to generate a complete periodic comprehensive record entry. The entry structure includes a period number, three-state combination labels, disturbance segment number, disturbance index item, and corresponding time interval. An example record format is "P11, Rise-Stable-Strong, F05, Reversal 3, Fluctuation 4, 62s-76s". All periodic entries must undergo bidirectional cross-validation using the period number to ensure data consistency and structural field integrity. During the processing, entries are arranged in ascending order by period number. The field structure uniformly includes a period index, state description, disturbance code, and behavioral details, and sequence identifiers are set to support batch retrieval and analysis operations. Finally, all entries are merged and summarized to obtain a periodic state associated record set.

[0146] S502: Based on each periodic record in the periodic state association record set, screen for the co-occurrence of continuous disturbance segments and multiple synchronous changes in state content, mark the period numbers with combination characteristics, and summarize them into the abnormal state list in chronological order to obtain the abnormal periodic state list.

[0147] First, extract the disturbance segment structure and synchronization state description sequentially from the recorded content. The disturbance segment is identified by the number of vibration reversals and the number of current fluctuation segments. For example, in period P14, there is segment F06 with 4 reversals and 5 fluctuation segments, which can be identified as a continuous disturbance segment. Then, extract the state label for this period, such as "upward" temperature trend, "continuous" voltage change, and "strong" load disturbance level. This is a combination structure under multiple state synchronization label conditions. It is necessary to identify whether each record has the combination of these two types of characteristics appearing at the same time. The operation method is to traverse the period state field and the disturbance segment field. When both satisfy "continuous disturbance segment + all three state items have the same trend or strong expression" in the same period, mark the period number as the target period. The period is assigned an anomaly label. In the example, P14 satisfies the combination characteristics of "continuous disturbance + upward - persistent - strong" and is included in the preliminary record set of the anomaly cycle. All cycle state records are processed in this way, and all cycle numbers with consistent structural combination characteristics are filtered and labeled. After labeling, these cycle numbers are collected into the anomaly classification list and then arranged in ascending order according to the start timestamp corresponding to the number to form a time series structure. For example, after sorting P08, P12, and P14, the records are 08:01, 12:05, 14:02, etc. A standard table entry structure is constructed with cycle number, combined state label and time position information, and a complete data list is organized. Finally, a centralized record set containing multiple anomaly cycle numbers is output, which is the anomaly cycle state list.

[0148] S503: Call the cycle number in the abnormal cycle status list, combine the temperature change, voltage fluctuation and vibration performance status within the cycle, identify abnormal features and classify them, and obtain equipment fault warning results;

[0149] First, iterate through all the cycle numbers already selected in the abnormal state list, and retrieve the temperature change, voltage fluctuation, and corresponding vibration performance data in the cycle feature dataset for each cycle, as well as the corresponding vibration performance data in the disturbance event segment list. The temperature state is judged based on the difference between the temperature values ​​measured at the beginning and end of the cycle. If the temperature at the end of the cycle is significantly higher than that at the beginning of the cycle and the difference exceeds the set threshold of ±0.3℃, it is marked as "rising". If the difference is within ±0.3℃, it is marked as "stable". Otherwise, it is marked as "falling". The voltage state is judged based on the fluctuation range formed by the maximum and minimum values ​​of the voltage records within the cycle. If the voltage change range exceeds 5V, it is considered "fluctuating". If the change is less than 2V, it is marked as "stable". The vibration state is judged by calling the segment records under the corresponding cycle number in the disturbance event segment list, extracting the number of frequency reversals and the number of peaks and troughs in the segment as the basis for judgment. If the number of frequency reversals in a segment is greater than 3, or the average frequency change amplitude exceeds 5Hz, it is judged as "high disturbance segment". Otherwise, it is classified as "medium disturbance" or "low disturbance" according to the specific values. After the three status channels are classified and labeled, they are integrated into a unified periodic status group entry. The record structure includes three items: period number, temperature status, voltage status, and vibration status. For example, the period number C06 corresponds to the status "Temperature: Upward, Voltage: Fluctuating, Vibration: High Disturbance". This information entry will serve as the direct input data for generating equipment fault warnings. Subsequently, the system categorizes and integrates all labeled periodic status records in numerical order, sets a unified field structure and sorting logic to ensure that the result data has structural consistency and traceability. Finally, a structured table containing the period number and the combined status of the three monitoring channels is formed, resulting in the equipment fault warning result.

[0150] In summary, this invention improves feature extraction accuracy by collecting temperature, voltage, and load disturbance data in segments and comparing trends; enhances disturbance identification capability by classifying vibration frequency and current peak values ​​in a time sequence; achieves sensitive judgment of changes in operating status by combining dynamic tracking of cooling efficiency and signal-to-noise ratio; and reveals synergistic relationships through consistent combination statistics in the fluctuation sequence, thereby strengthening the fault prediction effect.

[0151] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A machine learning-based method for early warning of medical device malfunctions, characterized in that, Includes the following steps: S1: Acquire temperature, voltage and load disturbance data of electromagnetic coils used for medical imaging during the operating cycle, compare the trends at the beginning and end of the cycle in a directional manner, integrate the change trajectory, and obtain the cycle feature dataset. S2: Based on the fluctuation cycle number in the periodic feature dataset, extract the magnet vibration frequency and current peak record, mark the start and end segments of the disturbance, identify the reversal and continuous peak segments, classify and summarize them in chronological order, and obtain a list of disturbance event segments; S3: Based on the period corresponding item in the list of disturbance event segments, extract the heat exchange efficiency and signal-to-noise ratio data of the cooling pump, track the direction of continuous periodic change, and combine the time information to classify it into the period corresponding category to obtain the drift behavior classification table. S4: Based on the period number of the drift behavior classification record, retrieve the temperature and voltage fluctuation sequence within the period, identify the combination content with the same direction, sort by period and count the combination frequency, and generate a behavior covariance correlation record table. S5: Based on the prominent cycle numbers in the behavior covariance correlation record table, call the cycle features and disturbance content, analyze the synchronous offset disturbance state of multiple indicators, mark them as potential fault cycles, and generate equipment fault early warning results.

2. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The periodic feature dataset includes stage distribution information, temperature direction identifier, voltage duration segment status, and load disturbance frequency label. The disturbance event segment list includes vibration reversal segments, current peak change sequences, and time sequence indexes. The drift behavior classification table includes trend classification categories, cycle number correspondence, and signal change direction attributes. The behavior covariance correlation record table includes combination direction matching results, intra-cycle change consistency annotations, and cross-cycle combination frequency values. The equipment fault warning results include abnormal cycle numbers, intra-cycle state combinations, and cycle annotation structure items.

3. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire the temperature change, voltage fluctuation and load disturbance performance of the electromagnetic coil assembly for medical imaging during continuous operation cycles, extract the original records of the beginning, middle and end segments of each cycle, divide and organize them into stage recording units of the cycle according to the segments, and obtain the original data group of the operation stage. S102: Based on the original data set of the operation phase, analyze the direction of temperature change between the first and last segments, mark the continuous segments of voltage in multiple segments, record the location of load disturbances within the time period, summarize the change characteristics in the cycle, and obtain the set of changes descriptions within the segment. S103: Based on the set of changes within the segment, integrate the temperature, voltage and disturbance changes corresponding to each cycle, generate corresponding marked entries by associating the cycle number, and organize all entries into a unified record structure in chronological order to obtain the cycle feature dataset.

4. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the fluctuation cycle number identified in the periodic feature dataset, obtain the vibration frequency and current peak value record of the magnet vibration control component in the corresponding cycle, extract the continuous signal segment in the cycle and divide it into segmented sequences according to time to obtain the periodic signal segmented sequence set. S202: Based on the vibration frequency change trajectory of each segment in the periodic signal segment sequence set, identify the position of direction reversal and mark the start and end points, record the continuous change of current peak value in the corresponding segment, calculate the number of direction reversals and the number of current peak value changes in the segment, and obtain a set of statistical indicators of disturbance behavior. S203: Call the segment numbers with continuous reversal and multiple current fluctuations in the set of disturbance behavior statistics indicators, classify and organize them according to time sequence and period, and archive them uniformly as periodic segment records with disturbance characteristics to obtain a list of disturbance event segments.

5. The machine learning-based medical equipment fault early warning method according to claim 4, characterized in that, The specific formulas for calculating the number of direction reversals and the number of current peak value changes within the segment are as follows: ； in, This represents the number of direction reversals and the number of times the peak current changes within a segment. This represents the peak current in the nth period signal segment. This represents the peak current value in the selected reference segment of the nth cycle signal segment. This represents the vibration frequency in the nth periodic signal segment. This represents the vibration frequency in the reference segment matched by the nth periodic signal segment. This represents the sum of squared deviations of all current peak value changes in the current paragraph. This represents the total number of direction reversals identified within paragraph x. This represents the mean of all vibration frequencies within the x-th segment. This represents the median of the current peak sequence within the x-th segment. This represents the sum of squares of the differences in peak current variations within the x-th segment. This represents the total number of periodic signal segments in the current paragraph.

6. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the corresponding period number in the list of disturbance event segments, extract the heat exchange efficiency and electrical signal-to-noise ratio records of the cooling pump module within the period, sort them by time and organize the data structure according to the period, and obtain the period signal archive data group. S302: Call the heat exchange efficiency sequence of each period in the periodic signal archive data group, identify the direction of numerical change in the segment, count the total number of consecutive upward segments in the sequence, calculate the proportion of consecutive upward segments, and obtain the trend upward feature set. S303: Based on the change characteristics corresponding to the upward trend feature set, and combined with the attribution cycle number and the time position of the disturbance event, the trend structure and cycle information are combined and classified to obtain a drift behavior classification table.

7. The machine learning-based medical equipment fault early warning method according to claim 6, characterized in that, The specific formula for calculating the proportion of the continuous upward paragraphs is as follows: ; in, This represents the proportion of consecutive upward-moving paragraphs. Represents all that satisfy The sum of the differences in heat exchange efficiency under different conditions. This represents the sum of the absolute values ​​of the heat exchange efficiency differences across all cycles. This represents the average completeness rate of all upstream sampling points. Represents the maximum number of consecutive upward cycles. This represents the average of the extreme fluctuations in heat exchange efficiency measured across all upward segments. This represents the average magnitude of the difference in heat exchange efficiency across different cycles. This represents the average difference between the current archiving period and the value set in the previous period. This represents the average frequency of environmental disturbances per unit time in the current data set.

8. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Call the records in the drift behavior classification table that are classified as upward or downward, extract the temperature offset data and voltage fluctuation sequence within the corresponding period, organize them into periodic records of two channels in chronological order, and mark the direction of change for each segment to obtain the periodic change direction data group. S402: Based on the temperature and voltage change direction labeling content in the periodic change direction data group, filter out the combination segments with the same direction in the same period, calculate the total number of times the combination segments with the same direction appear in all periods, and store them in association with the period number to obtain the frequency value of the combination segments with the same direction. S403: Based on the periodic distribution of the frequency values ​​corresponding to the direction-consistent combination, the data is categorized and organized into a set structure with periodic numbers, combination features and frequency indicators, and uniformly archived as periodic behavior coordination information to obtain a behavior covariance correlation record table.

9. The machine learning-based medical equipment fault early warning method according to claim 8, characterized in that, The formula for calculating the total number of occurrences of the direction-consistent combination segment in all periods is as follows: ； in, This represents the total number of times that a combination of elements with the same direction appears throughout the entire cycle. Representing the Temperature change within the period Representing the Voltage change during the period Representing the The dimensionless ratio parameter of the temperature-electricity variation amplitude within a period. Represents the smallest positive number approaching zero. This represents the total number of cycles.

10. The machine learning-based medical equipment fault early warning method according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the prominent cycle numbers in the behavior covariance correlation record table, call the corresponding content of the cycle in the cycle feature dataset and the list of disturbance event segments, extract the disturbance trajectory information and state change content of the cycle, and organize them into corresponding entries according to the cycle number to obtain the cycle state correlation record set. S502: Based on each periodic record in the periodic state association record set, screen for the co-occurrence of continuous disturbance segments and multiple synchronous state changes, mark the period numbers with combination characteristics, and summarize them into the abnormal state list in chronological order to obtain the abnormal periodic state list. S503: Call the cycle number in the abnormal cycle status list, combine the temperature change, voltage fluctuation and vibration performance status within the cycle, identify abnormal features and classify them, and obtain equipment fault warning results.

Images