Gradient system fault early warning method based on multi-parameter double-distance degree magnitude cascade
By using a multi-parameter dual-distance metric cascade method, combining Mahalanobis distance and Euclidean distance, the problem of multi-parameter correlation and early fault identification in magnetic resonance gradient system fault monitoring is solved, achieving efficient and accurate fault warning and location, and improving the reliability and safety of magnetic resonance equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies for fault monitoring of magnetic resonance gradient systems fail to adequately utilize multi-parameter correlation, have poor ability to identify weak early fault signals, and lack model generalization ability, leading to frequent false alarms and missed alarms, making it difficult to achieve accurate early warning of faults.
A method based on multi-parameter dual-distance metric cascade is adopted, which combines Mahalanobis distance and Euclidean distance, along with residual analysis and trend judgment, to integrate multi-dimensional monitoring data, capture the correlation between parameters, identify early abnormal patterns, and provide fault warning through risk probability ranking.
It improves the accuracy and timeliness of fault early warning, reduces false alarm and false alarm rates, provides clear fault location and risk quantification, and enhances the applicability and engineering practicality of the method.
Smart Images
Figure CN121656930A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of magnetic resonance technology, specifically relating to a gradient system fault early warning method based on multi-parameter dual distance metric cascade. Background Technology
[0002] The stable operation of magnetic resonance (MR) imaging equipment is highly dependent on the reliability of its gradient system. The gradient system, through precise control of the switching and intensity of the gradient field, directly affects the spatial coding quality and scanning speed of the image. A failure in the gradient system can lead to equipment downtime, interruption of clinical diagnosis, and even endanger patient safety. Therefore, achieving accurate and early warning of gradient system failures has significant engineering and clinical value for improving equipment availability and ensuring medical safety.
[0003] Currently, fault monitoring and early warning for MR gradient systems mainly rely on the following methods: The first is traditional monitoring methods based on fixed thresholds, which set safety upper and lower limits for key parameters such as gradient field strength, coil temperature, and cooling flow rate to trigger alarms. While simple and intuitive, this method ignores the dynamic coupling relationship between multiple parameters, making it prone to false alarms due to instantaneous fluctuations in a single parameter or missed alarms due to the failure of multiple parameters to exceed the threshold during coordinated degradation. It is also insensitive to early, slowly changing faults. The second is diagnostic methods based on signal analysis, such as performing spectral analysis (e.g., Fourier transform, wavelet transform) on vibration or current signals to extract fault characteristic frequencies. These methods have some ability to locate certain typical faults, but their feature extraction rules often rely on expert experience, making them poorly adaptable to scenarios with weak early fault signals and indistinct features, and difficult to extend to different models and operating conditions. The third is methods based on classical machine learning, such as using support vector machines (SVM) or neural networks to classify and model historical fault data. However, MR gradient systems monitor parameters with high dimensionality and strong nonlinearity. Such methods usually require a large number of balanced fault samples for training. In reality, when fault samples are scarce, the model is prone to overfitting, has insufficient generalization ability, and limited early warning accuracy.
[0004] In summary, existing technologies have significant shortcomings in utilizing multi-parameter correlations, identifying early weak fault signals, and model generalization capabilities. To overcome these limitations, there is an urgent need for an intelligent early warning method that can effectively integrate high-dimensional monitoring data, capture correlations between parameters, and be highly sensitive to early abnormal patterns, in order to achieve "early detection and accurate assessment" of faults in MR gradient systems. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a gradient system fault early warning method based on a multi-parameter dual-distance metric cascade. It applies Mahalanobis distance based on multi-dimensional collected data to detect abnormal data and uses the maximum residual method to determine the abnormal dimension. Based on the fault type corresponding to the abnormal dimension, it uses Euclidean distance and trend prediction methods to confirm the fault type. Finally, it ranks the risk probabilities of various possible faults and issues early warnings.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] A gradient system fault early warning method based on multi-parameter dual-distance metric cascade includes:
[0008] Step 1: Establish a gradient system fault database, which includes various marked fault types and corresponding historical fault data and historical normal data;
[0009] Step 2: Obtain the current gradient system test data to be detected, calculate its Mahalanobis distance with the historical normal data, and compare it with a preset distance threshold calculated based on the historical normal data to determine whether the current gradient system test data to be detected is abnormal.
[0010] Step 3: If the test data is determined to be abnormal, calculate the residual values of the test data in each data dimension, and determine the dimension corresponding to the maximum residual value as the warning variable.
[0011] Step 4: Based on the warning variables, associate one or more candidate fault types;
[0012] Step 5: For each candidate fault type, calculate the Euclidean distance between the test data and the mean of the historical fault dataset and the mean of the historical normal dataset corresponding to that candidate fault type. Combine the distance change trend of multiple sets of continuously collected test data to comprehensively determine whether the current state is highly abnormal, suspected abnormal, or normal, and output the warning information.
[0013] Furthermore, in step 1, both historical fault data and historical normal data contain multiple parameter dimensions that reflect the operating status of the gradient system. These parameters include at least one or more of the following: gradient system performance parameters such as gradient coil temperature rise and gradient amplifier heat dissipation device temperature, hardware operating parameters, and status parameters. The data of each dimension are normalized before distance calculation.
[0014] Furthermore, in step 2, the Mahalanobis distance between each group of data and all other data in the historical normal dataset is calculated, and the calculated maximum Mahalanobis distance is used as the distance threshold.
[0015] Furthermore, in step 3, the specific method for calculating the residual values of the test data in each data dimension is as follows: calculate the difference between the value of the test data in each data dimension and the average value of the historical normal dataset in that dimension, and use the absolute value of the difference as the residual value in the corresponding data dimension.
[0016] Furthermore, in step 4, by traversing all the fault types marked in the fault database, the warning variable is matched with the dimensions contained in the historical fault data corresponding to each fault type, and all fault types containing the warning variable are taken as the candidate fault types.
[0017] Furthermore, in step 5, the comprehensive judgment logic for a candidate fault type includes: if the Euclidean distance between the test data and the mean of the normal dataset of the corresponding fault type is greater than or equal to the Euclidean distance between the test data and the mean of the fault dataset of the corresponding fault type, it is judged as highly abnormal; if the Euclidean distance between the test data and the mean of the normal dataset is less than the Euclidean distance between the test data and the mean of the fault dataset, then a trend judgment is also required based on multiple continuously collected test data. Only when the Euclidean distance between the test data and the mean of the normal dataset shows a continuous increasing trend is it judged as suspected abnormal; otherwise, it is judged as normal.
[0018] Furthermore, for candidate fault types that are determined to be highly abnormal or suspected abnormal, the risk probability of the fault type is calculated based on the Euclidean distance between the test data and the mean of the normal dataset of the corresponding fault type, and the Euclidean distance between the test data and the mean of the fault dataset of the corresponding fault type.
[0019] Furthermore, the step of outputting early warning information includes: when multiple candidate fault types are determined to be abnormal, sorting them according to the risk probability of each fault type, and generating an early warning prompt containing fault type, early warning status and priority based on the sorting result.
[0020] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned gradient system fault early warning method based on a multi-parameter dual-distance metric cascade.
[0021] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned gradient system fault early warning method based on a multi-parameter dual-distance metric cascade.
[0022] The beneficial effects of this invention are as follows:
[0023] By integrating multi-parameter correlation and trend tracking, the accuracy of early warning is improved: Compared with traditional methods based on a single threshold or simple multiple thresholds, this invention utilizes Mahalanobis distance to fully consider the correlation between multiple monitoring parameters, effectively avoiding false alarms caused by normal fluctuations of a single parameter. Simultaneously, by combining subsequent Euclidean distance discrimination and trend analysis of multiple sets of data, it can keenly capture early, weak, complex anomaly patterns, significantly reducing the risk of missed alarms.
[0024] Reducing reliance on fault samples and enhancing method applicability: Unlike traditional machine learning models (such as SVM and neural networks) that require massive amounts of fault data for training, the method in this invention mainly relies on sufficient normal system state data, requiring only a small amount of typical fault data as a reference. This feature solves the problem of scarce fault samples and high acquisition costs in practical engineering, making the method easier to deploy and apply on MR gradient systems of different models and configurations.
[0025] It provides clear anomaly localization and risk quantification: By using the maximum residual method, it quickly identifies the dominant dimensional variable causing the anomaly, providing direct clues for fault tracing. Furthermore, by calculating and ranking the risk probabilities of different fault types, it provides maintenance personnel with clear fault likelihood and handling priorities, assisting them in making accurate decisions and preventative maintenance, thereby improving maintenance efficiency.
[0026] The method is robust and practical for engineering applications: Mahalanobis distance itself has the advantages of noise resistance and elimination of dimensional effects. Combined with normalization preprocessing, the algorithm's robustness to data noise and parameter unit differences is enhanced. The entire process has clear logic, controllable computational complexity, and good engineering feasibility and real-time processing potential. Attached Figure Description
[0027] Figure 1 This is a flowchart of a gradient system fault early warning method based on multi-parameter dual distance metric cascade according to the present invention. Detailed Implementation
[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0029] like Figure 1 As shown, this invention provides a gradient system fault early warning method based on multi-parameter dual distance metric cascade. By sequentially applying two distance metric methods, it achieves accurate classification of multi-dimensional gradient system data, enabling fault early warning and improving equipment operational reliability. The method specifically includes:
[0030] Step 1: Establish a gradient system fault database, which includes various marked fault types and corresponding historical fault data and historical normal data;
[0031] Establish a gradient system fault database and mark the fault type as follows: The corresponding fault data are as follows: Where n is the number of fault types. The dimension of the variable corresponding to each type of fault data is as follows: For example, fault type Corresponding fault dataset Fault dataset The dimension of the gradient system test data variables included is The test data variables refer to gradient-related parameters such as the temperature rise of the gradient coil and the temperature of the gradient amplifier's heat dissipation device. Similarly, the fault type... The corresponding normal gradient system data type is marked as The corresponding normal dataset is The test data variables it contains have the following dimensions. .
[0032] Preferably, if the data or It stores multiple sets of historical data, and at the same time, because or The units of the data variables in each dimension are different and the values vary greatly. The test data variables here are all normalized variables.
[0033] Step 2: Obtain the current gradient system test data to be detected, calculate its Mahalanobis distance with the historical normal data, and compare it with a preset distance threshold calculated based on the historical normal data to determine whether the current gradient system test data to be detected is abnormal.
[0034] Calculate the current gradient system test data Compared with normal datasets The Mahalanobis distance between them is calculated using the following formula:
[0035] ,
[0036] in, Normal dataset The mean vector reflects the center position of a normal dataset, and the superscript T indicates transpose; Normal dataset The covariance matrix reflects the degree of dispersion of each dimension of the data and the correlation between dimensions. for The inverse matrix.
[0037] Referring to the formula for calculating Mahalanobis distance, the normal dataset is calculated through a traversal operation. The maximum Mahalanobis distance between each data set and other data sets is denoted as . .
[0038] if It is determined that the current gradient system test data to be tested falls within the normal data range; if It is determined that the current gradient system test data to be tested belongs to the abnormal data range.
[0039] Step 3: If the test data is determined to be abnormal, calculate the residual values of the test data in each data dimension, and determine the dimension corresponding to the maximum residual value as the warning variable.
[0040] For outlier data, calculate the residuals for each dimension using the following formula:
[0041] ,
[0042] in, Fault type The corresponding fault variable number, For the first The residual values of the dimensional data, For test data The Dimensional data values, Normal dataset No. The average value of the dimension.
[0043] Among them, ( The maximum value of ) The corresponding dimension is marked as a warning variable.
[0044] Step 4: Based on the warning variables, associate one or more candidate fault types;
[0045] Repeat steps 2-3. Range of values Where n is the number of fault types. Iterate through all fault types, find all warning variables, and list all fault types related to these variables. For example, after iteration and calculation, two fault types are obtained. and ( ).
[0046] Step 5: For each candidate fault type, calculate the Euclidean distance between the test data and the mean of the historical fault dataset and the mean of the historical normal dataset corresponding to that candidate fault type. Combine the distance change trend of multiple sets of continuously collected test data to comprehensively determine whether the current state is highly abnormal, suspected abnormal, or normal, and output the warning information.
[0047] According to the type of fault Calculate the normal dataset mean vector and data collection The Euclidean distance is given by the following formula:
[0048] ,
[0049] Calculate fault data mean vector and data collection The Euclidean distance is given by the following formula:
[0050] ,
[0051] (1) When Error warning displayed, error type Highly abnormal (marked as state 1);
[0052] (2) When Then, collect two sets of data in chronological order. and And calculate and Only when When an error warning is displayed (error type), an error alert will be displayed. If the condition is suspected to be abnormal, it is marked as status 2. All other conditions indicate that the status is normal (marked as status 3).
[0053] Similarly, regarding the type of fault Perform the same processing steps as described above to determine whether the collected data is normal or faulty.
[0054] Based on the magnitude of risk probability and The system is sorted and early warnings are issued. The formulas for calculating the probability of failure risk are as follows:
[0055] , ,
[0056] Regarding the aforementioned fault types or (and) The prompts will be given in three different ways:
[0057] Status 3: Data is normal, no risk;
[0058] Status 1: Data anomaly, issue an alert, error type or (and) Highly abnormal;
[0059] Status 2: Data anomaly, issue an alert, error type or (and) This appears to be an anomaly.
[0060] In summary, this invention proposes a gradient system fault early warning method based on a cascaded multi-parameter dual-distance metric. By introducing a cascaded metric mechanism of Mahalanobis and Euclidean distance, combined with residual analysis and trend judgment, it effectively solves the problems of insufficient utilization of multi-parameter correlation, low sensitivity to early faults, and strong dependence on massive fault samples in existing technologies. This method not only improves the accuracy and timeliness of fault early warning and reduces false alarm and false negative rates, but also provides a clear basis for operation and maintenance decisions through risk probability ranking, enhancing the method's engineering applicability and scalability. The implementation of this invention will provide efficient and reliable technical support for the condition monitoring and preventive maintenance of gradient systems in magnetic resonance imaging equipment.
[0061] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned gradient system fault early warning method based on a multi-parameter dual-distance metric cascade.
[0062] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned gradient system fault early warning method based on a multi-parameter dual-distance metric cascade.
[0063] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A gradient system fault early warning method based on multi-parameter dual-distance metric cascade, characterized in that, include: Step 1: Establish a gradient system fault database, which includes various marked fault types and corresponding historical fault data and historical normal data; Step 2: Obtain the current gradient system test data to be detected, calculate its Mahalanobis distance with the historical normal data, and compare it with a preset distance threshold calculated based on the historical normal data to determine whether the current gradient system test data to be detected is abnormal. Step 3: If the test data is determined to be abnormal, calculate the residual values of the test data in each data dimension, and determine the dimension corresponding to the maximum residual value as the warning variable. Step 4: Based on the warning variables, associate one or more candidate fault types; Step 5: For each candidate fault type, calculate the Euclidean distance between the test data and the mean of the historical fault dataset and the mean of the historical normal dataset corresponding to that candidate fault type. Combine the distance change trend of multiple sets of continuously collected test data to comprehensively determine whether the current state is highly abnormal, suspected abnormal, or normal, and output the warning information.
2. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 1, characterized in that, In step 1, both historical fault data and historical normal data contain multiple parameter dimensions that reflect the operating status of the gradient system. The parameters include at least one or more of the following: gradient coil temperature rise, gradient amplifier heat dissipation device temperature, hardware operating parameters, and status parameters. The data of each dimension are normalized before distance calculation.
3. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 1, characterized in that, In step 2, the Mahalanobis distance between each group of data and all other data in the historical normal dataset is calculated, and the calculated maximum Mahalanobis distance is used as the distance threshold.
4. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 1, characterized in that, In step 3, the specific method for calculating the residual values of the test data in each data dimension is as follows: calculate the difference between the value of the test data in each data dimension and the average value of the historical normal dataset in that dimension, and take the absolute value of the difference as the residual value in the corresponding data dimension.
5. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 1, characterized in that, In step 4, by traversing all the fault types marked in the fault database, the warning variable is matched with the dimensions contained in the historical fault data corresponding to each fault type, and all fault types containing the warning variable are taken as the candidate fault types.
6. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 1, characterized in that, In step 5, the comprehensive judgment logic for a candidate fault type includes: if the Euclidean distance between the test data and the mean of the normal dataset of the corresponding fault type is greater than or equal to the Euclidean distance between the test data and the mean of the fault dataset of the corresponding fault type, it is judged as highly abnormal; if the Euclidean distance between the test data and the mean of the normal dataset is less than the Euclidean distance between the test data and the mean of the fault dataset, it is also necessary to perform trend judgment based on multiple continuously collected test data. Only when the Euclidean distance between the test data and the mean of the normal dataset shows a continuous increasing trend is it judged as suspected abnormal; otherwise, it is judged as normal.
7. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 6, characterized in that, For candidate fault types that are determined to be highly abnormal or suspected to be abnormal, the risk probability of the fault type is calculated based on the Euclidean distance between the test data and the mean of the normal dataset of the corresponding fault type, and the Euclidean distance between the test data and the mean of the fault dataset of the corresponding fault type.
8. The gradient system fault early warning method based on multi-parameter dual-distance metric cascade as described in claim 7, characterized in that, The steps for outputting early warning information include: when multiple candidate fault types are determined to be abnormal, sorting them according to the risk probability of each fault type, and generating an early warning prompt containing the fault type, early warning status and priority based on the sorting result.
9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the gradient system fault early warning method based on multi-parameter dual distance metric cascade as described in any one of claims 1-8.
10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the gradient system fault early warning method based on multi-parameter dual distance metric cascade as described in any one of claims 1-8.