CN116256686A - MRI equipment abnormality detection method and system based on Internet of things data - Google Patents
MRI equipment abnormality detection method and system based on Internet of things data Download PDFInfo
- Publication number
- CN116256686A CN116256686A CN202310103018.3A CN202310103018A CN116256686A CN 116256686 A CN116256686 A CN 116256686A CN 202310103018 A CN202310103018 A CN 202310103018A CN 116256686 A CN116256686 A CN 116256686A
- Authority
- CN
- China
- Prior art keywords
- data
- pressure
- time
- sliding window
- internet
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005856 abnormality Effects 0.000 title claims abstract description 44
- 238000001514 detection method Methods 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 claims abstract description 40
- 230000002159 abnormal effect Effects 0.000 claims abstract description 36
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000002372 labelling Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 12
- 238000007637 random forest analysis Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 230000003247 decreasing effect Effects 0.000 claims description 5
- 238000009413 insulation Methods 0.000 claims description 5
- 230000000087 stabilizing effect Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000006855 networking Effects 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 3
- 238000010276 construction Methods 0.000 abstract 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 34
- 239000001307 helium Substances 0.000 description 19
- 229910052734 helium Inorganic materials 0.000 description 19
- SWQJXJOGLNCZEY-UHFFFAOYSA-N helium atom Chemical compound [He] SWQJXJOGLNCZEY-UHFFFAOYSA-N 0.000 description 19
- 239000007788 liquid Substances 0.000 description 10
- 230000000694 effects Effects 0.000 description 7
- 238000005057 refrigeration Methods 0.000 description 7
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 6
- 238000003066 decision tree Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000007423 decrease Effects 0.000 description 4
- 238000001816 cooling Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 229910052757 nitrogen Inorganic materials 0.000 description 3
- 238000010791 quenching Methods 0.000 description 3
- 230000000171 quenching effect Effects 0.000 description 3
- 239000002826 coolant Substances 0.000 description 2
- 238000001704 evaporation Methods 0.000 description 2
- 230000008020 evaporation Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 229910052739 hydrogen Inorganic materials 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000013450 outlier detection Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 238000005481 NMR spectroscopy Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000012809 cooling fluid Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000002887 superconductor Substances 0.000 description 1
- 230000002459 sustained effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R35/00—Testing or calibrating of apparatus covered by the other groups of this subclass
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- General Business, Economics & Management (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention belongs to the technical field of medical equipment maintenance, and particularly relates to an MRI equipment abnormality detection method and system based on Internet of things data. The method comprises the following steps: step 1, collecting time sequence data in the running process of MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO; and 2, carrying out real-time prediction by adopting a sliding window algorithm, obtaining the characteristics in the sliding window based on the time sequence data, inputting the characteristics into an abnormality detection model, and obtaining the probability of equipment abnormality at the time point after the sliding window. The invention provides the labeling method and the characteristic construction method of the abnormal sequence, realizes the function of predicting the abnormality of the MRI equipment by utilizing the time sequence data acquired by the Internet of things, and has good application prospect.
Description
Technical Field
The invention belongs to the technical field of medical equipment maintenance, and particularly relates to an MRI equipment abnormality detection method and system based on Internet of things data.
Background
Magnetic Resonance Imaging (MRI) is commonly used to visualize the structure and function of an object. By detecting a small difference in magnetic field, it can display the internal structure of the human body in 2D or 3D. Compared with common CT, MRI can better judge the tissue position of complex areas such as human joints.
MRI imaging uses signals from hydrogen nuclei (1H) to generate images. The current is conducted through the coil and wire, creating a strong magnetic field. To maintain such a strong magnetic field while reducing energy loss, superconductors are required. In order to maintain superconductivity, the coils in the magnetic cavity need to reach superconducting temperatures. The entire main magnet is immersed in the cryogenic coolant. The preferred coolant for MRI machines is liquid helium. In order to maintain a desired temperature, a refrigeration subsystem is required. Normally, a refrigeration system consists of a coldhead (the core of the cooling system), a helium compressor (to cool the coldhead), a shroud (the medium that conducts temperature through the cavity), and a chiller (to let cold water flow through to reduce the heat of all systems).
Among the faults that occur in MRI, refrigeration system faults account for a significant proportion. Prolonged refrigeration system failure will lead to elevated He pressure and lowered helium level and ultimately to a very serious consequence, magnetic quenching. According to an interview study conducted at a hospital for nine months at the university of sudan, errors in the magnet account for about 95% of the total portion when checking for error problems occurring in the MRI system. According to his team's study, a failure of the refrigeration system will result in an increase in He pressure and a decrease in helium level. In both coldhead shut-down conditions, the helium pressure exceeded the standard threshold of about 0.78PSI, with helium levels below 60% of the normal range, respectively. Another study also shows He pressure versus refrigeration system downtime. In both cases, he pressure exceeds the threshold (4.1 PSI) by 0.11PSI and 0.76PSI. Thus, the method is applicable to a variety of applications. It is necessary to monitor the refrigeration system periodically to avoid the He pressure from continuously increasing and causing magnetic quenching.
When the temperature in the entire central magnet bore exceeds the superconducting temperature due to a failure of the cooling system or excessive heat received from another system over a period of time, rapid evaporation of the cooling fluid may cause volume expansion. Liquid helium expands 7 times when it evaporates, which brings about a huge pressure on the cavity. The cavity will be directly connected to the outside to pump out helium, i.e. magnetically quenched. This situation can cause significant loss to hospitals. According to some examples in hospital maintenance reports, 300L of liquid helium may be lost after nuclear magnetic resonance quenching. The value of these liquid helium adds to the cost of subsequent cuts in the cavity totaling 100 tens of thousands of yuan. During this time, the examination of about 1300 patients was deferred. In addition to capital losses, the resource reserves of liquid nitrogen are also a problem related to national policies. Currently, most of the world's liquid nitrogen comes from natural gas collection. The widely used cooling resource liquid nitrogen is greatly limited due to source limitations and monopolies of countries in the supply chain. Each institute should develop new methods to prevent evaporation of liquid helium into the air.
In order to find faults in time, most hospital engineers periodically go to a machine room to check the MRI equipment and observe whether parameters such as helium pressure exceed a threshold value. The disadvantage of this inspection method is obvious that long-term continuous monitoring and immediate effective monitoring cannot be achieved. And cannot track the fluctuation of the liquid helium pressure which is too high in a short time and the like.
The internet of things (IoMT) data is time series data obtained by acquiring equipment in real time through the internet of things, and comprises state information of equipment operation. Currently, there is no research on how to identify anomalies in MRI devices from internet of things data, and how to further predict the risk that future MRI devices may fail using internet of things data. Therefore, the technical scheme for predicting the abnormality of the MRI equipment through the data of the Internet of things still lacks in the prior art.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an MRI equipment abnormality detection method and system based on internet of things data, and aims to predict the possible fault risk of the MRI equipment by using the internet of things data.
An MRI equipment abnormality detection method based on internet of things data comprises the following steps:
step 1, collecting time sequence data in the running process of MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO;
step 2, carrying out real-time prediction by adopting a sliding window algorithm, obtaining characteristics in a sliding window based on the time sequence data, and inputting the characteristics into an abnormality detection model to obtain the probability of equipment abnormality at a time point after the sliding window;
in the training process of the anomaly detection model, the marking method of training data comprises the following steps:
step a, dividing original time sequence data into a plurality of subsequences by using a sliding window;
step b, selecting an abnormal index for analysis and calculating a subsequence; the calculation mode of the abnormal index is as follows: calculating the slope between any two adjacent data points in the subsequence, and calculating the radius of the confidence interval;
and c, marking the subsequence as an abnormal sequence by using an index deviating from a normal value.
Preferably, the equipment anomaly comprises at least one of the following three cases:
1) The He pressure changes suddenly and rapidly rise, and rise again after stabilizing for a period of time;
2) He pressure is continuous, nonvolatile, rapidly increasing or decreasing;
3) He pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again.
Preferably, the time series data includes He pressure and Shield Si410;
or, the time series data includes He pressure, shield Si410, and He level.
Preferably, the features include an average value of He pressure within the sliding window, a standard deviation of He pressure within the sliding window, an original value of He pressure, and a standard deviation of insulation layer temperature.
Preferably, in step 1, the time series data is collected and then preprocessed, and the preprocessing step includes compressing a plurality of values that are continuously unchanged into a single value.
Preferably, the anomaly detection model adopts a random forest algorithm.
Preferably, the time series data is that one data point is acquired every minute, and the window length of the sliding window takes 8-9 data points.
The invention also provides an MRI equipment abnormality detection system based on the data of the Internet of things, which comprises:
the data acquisition and storage module is used for acquiring and storing time sequence data in the operation of the MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO;
the abnormality detection module is used for carrying out real-time prediction by using a sliding window algorithm, obtaining characteristics in a sliding window based on the time sequence data, inputting the characteristics into an abnormality detection model, and obtaining the probability of equipment abnormality at a time point after the sliding window;
in the training process of the anomaly detection model, the method for labeling training data comprises the following steps:
step a, dividing original time sequence data into a plurality of subsequences by using a sliding window;
step b, selecting an abnormal index for analysis and calculating a subsequence; the calculation mode of the abnormal index is as follows: calculating the slope between any two adjacent data points in the subsequence, and calculating the radius of the confidence interval;
and c, marking the subsequence as an abnormal sequence by using an index deviating from a normal value.
Preferably, the equipment anomaly comprises at least one of the following three cases:
1) The He pressure changes suddenly and rapidly rise, and rise again after stabilizing for a period of time;
2) He pressure is continuous, nonvolatile, rapidly increasing or decreasing;
3) He pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again.
The invention also provides a computer readable storage medium, on which a computer program for implementing the above-mentioned method for detecting the abnormality of the MRI apparatus based on the internet of things data or a computer program for implementing the above-mentioned system for detecting the abnormality of the MRI apparatus based on the internet of things data is stored.
In the present invention, the raw data such as "He pressure" is a time series, indexes are 1 to n, each index represents one sample (instance), including the input data value and class label. At this time, the raw data is called an original value, such as "He pressure original value". The "average value of He pressure" is calculated by using a sliding window based on the "original value of He pressure". For example, the "He pressure original values" corresponding to the indexes of k-3, k-2, k-1 and k are averaged to calculate an "average value of He pressure"
Aiming at the characteristics of the data of the Internet of things of the MRI equipment, the method for labeling the abnormal sequence and how to construct the characteristics for predicting the abnormality of the MRI equipment are determined, so that the aim of accurately predicting the abnormality of the MRI equipment is fulfilled. The invention is used for real-time monitoring of MRI equipment, can be prepared before the actual fault occurs, so as to avoid serious faults, and has good application prospect.
It should be apparent that, in light of the foregoing, various modifications, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
The above-described aspects of the present invention will be described in further detail below with reference to specific embodiments in the form of examples. It should not be understood that the scope of the above subject matter of the present invention is limited to the following examples only. All techniques implemented based on the above description of the invention are within the scope of the invention.
Drawings
FIG. 1 is a typical He pressure profile;
FIG. 2 is a variation of interval He pressure;
FIG. 3 is a graph of observed outlier quantity as a function of window length;
FIG. 4 is a graph of observed outlier quantity variation as a function of window length;
FIG. 5 is a time series of changes in He pressure after a first order difference;
FIG. 6 is a schematic diagram of a sliding window;
FIG. 7 is a ROC curve of a support vector machine model;
FIG. 8 is a ROC curve of a decision tree model;
FIG. 9 is a graph showing the effect of different lead times on random forest model performance;
FIG. 10 is a graph showing the effect of different feature combinations on random forest model performance;
FIG. 11 is a confusion matrix for the random forest model when the feature combinations are HS and HSL;
fig. 12 is a ROC curve for a random forest model with the feature combination HS and HSL.
Detailed Description
It should be noted that, in the embodiments, algorithms of steps such as data acquisition, transmission, storage, and processing, which are not specifically described, and hardware structures, circuit connections, and the like, which are not specifically described may be implemented through the disclosure of the prior art.
Embodiment 1 MRI (magnetic resonance imaging) equipment abnormality detection method and system based on Internet of things data
The embodiment provides an MRI equipment abnormality detection method and system based on internet of things data, wherein the system comprises:
the data acquisition and storage module is used for acquiring and storing time sequence data in the operation of the MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO;
the anomaly detection module is used for carrying out real-time prediction by using a sliding window algorithm, obtaining characteristics in a sliding window based on the time sequence data, inputting the characteristics into an anomaly detection model, and obtaining the probability of equipment anomaly at the time point after the sliding window.
The method for detecting the abnormality of the MRI equipment by adopting the system comprises the following steps:
step 1, collecting time sequence data in the running process of MRI equipment; the time series data includes at least one of He pressure, he level, water flow, water temperature, insulation layer temperature, shield Si410, or coldhead RuO. The time series data is collected and then preprocessed, and the preprocessing step comprises compressing a plurality of values which are continuously unchanged into a single value.
And 2, carrying out real-time prediction by adopting a sliding window algorithm, obtaining the characteristics in the sliding window based on the time sequence data, inputting the characteristics into an abnormality detection model, and obtaining the probability of equipment abnormality at the time point after the sliding window.
In the training process of the anomaly detection model, the marking method of training data comprises the following steps:
step a, dividing original time sequence data into a plurality of subsequences by using a sliding window;
step b, selecting an abnormal index for analysis and calculating a subsequence; the calculation mode of the abnormal index is as follows: calculating the slope between any two adjacent data points in the subsequence, and calculating the radius of the confidence interval;
and c, marking the subsequence as an abnormal sequence by using an index deviating from a normal value (marking the subsequence as abnormal if the confidence interval radius is larger than a threshold value).
After labeling according to the method, the equipment abnormality comprises at least one of the following three situations:
1) The He pressure changes suddenly and rapidly rise, and rise again after stabilizing for a period of time; as shown in fig. 1 c;
2) He pressure is continuous, nonvolatile, rapidly increasing or decreasing; as shown in fig. 1d and 1 e;
3) He pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again as shown in fig. 1 f.
In contrast, fig. 1a and 1b also provide a variation of He pressure under normal conditions.
The following shows how to label the abnormal sequence of the data (time sequence data) of the Internet of things through the processing process of the specific data, so that the subsequent modeling process is realized.
1. Description of data
The experimental data is from MRI model GEMRI 382 from 10, 30, 2018 to 11, 19, 2020. He pressure was chosen as a feature of MRI. The data sampling frequency was once per minute. In this time series, there were a total of 1048575 data, with 30000 missing values. In general, the data values approximate to some extent the periodically varying internal fluctuations. Fig. 2 shows the variation of He pressure during interception.
2. Data preprocessing
Since the data of this experimental example is relatively large, the He pressure is constant throughout a certain time, and suddenly rises or falls in a certain minute, the data is compressed by compressing the time series, which is unchanged, into a single value. Since the value of He pressure always rises or falls in a certain minute, t in the new time series j+i -t j+i-1 Always 1 minute. Finally, a new sequence of length 258055, average 3.677 and standard deviation 0.481 was obtained.
3. Outlier detection
First, the effect of window length on the algorithm was studied. In the experiment, the step r is equal to 1. Through normal distribution inspection, the threshold is set to six standards of CIR, then different lengths are set, and the change of the number of abnormal values is observed.
The confidence interval distance radius calculation method comprises the following steps: first, calculating the slope, k, between any two adjacent data points in the subsequence j ＝(y(t j+i )-y(t j+i-1 ))/(t j+i -t j+i-1 ) (1. Ltoreq.i. Ltoreq.l-1) and then calculate the confidence interval
It can then pass->
Obtaining a radius, wherein>
j θIs the upper and lower confidence interval limits of the relative. By combining the equations above, the confidence interval radius can be found to be equal to +.>
There is also a new sequence,/A->
CIR is a sequence of confidence inner radii. Assume that for time series y= (Y (t) 1 ),y(t 2 ),…,y(t n ) For n variables, there is a confidence interval distance radius dj with slope for the jth subsequence Y (j) (1. Ltoreq.j. Ltoreq.n-l)/r+1)>Gamma, where gamma is the threshold. Then the jth subsequence Y, Y (j) in the time sequence Y is an abnormal subsequence. Statistically, if the data distribution is approximately normal, then about 68% of the data values are within one standard deviation before and after the average, about 95% of the data values are within both standard deviations before and after the average, and about 99.7% of the data values are within three standard deviations before and after the averageWithin the deviation. To make outliers more pronounced, six standard deviations were used to select them. This means that γ is equal to six standard deviations of the sequence CIR.
Fig. 3 and 4 show that as the window length increases, the number of outliers gradually decreases and tends to stabilize, while the number of outlier changes increases and then decreases. When the window length is smaller, the number of adjacent data points is smaller, and accurate subsequence features are difficult to obtain according to the data slope change information. Thus, although many abnormal values can be detected, the detection accuracy is low. When the window length is large, there are too many adjacent data points, so that the outliers are balanced by the normal data, resulting in low detection accuracy. When the window length is equal to 8, the variation of the outliers is largest, and then the number of outliers becomes more and more stable. In the next step, it sets the sliding window length equal to 9.CIR is a normal distribution sequence with an average value of 0.003 and a standard deviation of 0.003.
If the window length is equal to 9, 277 abnormal subsequences are obtained. However, many abnormal subsequences are contiguous, and contiguous abnormal subsequences need to be combined. After combining adjacent abnormal subsequences into some larger and discontinuous subsequences, 55 abnormal subsequences are obtained, containing 717 data values in total. The differences between the normal and abnormal subsequences are compared. Some representative outliers and normals were chosen by me below. Fig. 1 (a) and 1 (b) are typical normal numbers, and fig. 1 (c) to 1 (f) are typical abnormal numbers. Under normal conditions, the He pressure steadily rises or falls over time. However, when an abnormality occurs, it can be classified into three cases. The first case is that, as in the first case, the pressure change of He is normal and suddenly rises sharply. It will then stabilize for a period of time and rise again as shown in fig. 1 (c). Under normal conditions, he pressure changes to wave up or down, but is relatively smooth. The second anomaly is a sustained, non-volatile, rapid increase or decrease in helium pressure, as shown in fig. 1 (d) and 1 (e). The last one is fig. 1 (f), he pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again. Fig. 5 shows a time series of He pressure changes after the first-order difference. Blue bars represent normal values and yellow bars represent outliers. The graph after the first order difference makes the abnormality more visual.
The technical scheme of the invention is further described through experiments.
Experimental example 1 comparison of different machine learning models
The experimental example compares the prediction performance of three machine learning models, wherein the used data and the abnormal labeling process are the same as that of experimental example 1.
After detecting an anomaly, a new sequence will be added as y for the prediction. If abnormal, y is equal to 1, otherwise y is equal to 0. Our prediction principle is shown in figure 6. The Lead time is the number of previous data of the detected y-point, and the data of the Lead time is used to predict whether the y-point is abnormal. There are six parameters in the raw data, namely He level, he pressure, water flow, water temperature, shield Si410 and coldhead RuO. The cold head RuO remains unchanged and has no effect on the prediction, so the effective parameter is 5. Since the anomaly is detected by the He pressure, the effect of the other four parameters is virtually unknown, and the data x used in the lead time is the He pressure or a set of He pressures and other several parameters. Everything is determined by the level of recall scores and precision scores of the predicted outcomes to determine which parameters to use. Recall score means the probability that an actual outlier is detected as an outlier, while precision score means the prediction is an outlier, which is in fact the likelihood of an outlier, as shown in fig. 6.
1. Support vector machine
The training set randomly decimates 80% of the entire sample size. The raw data includes 5 features including He pressure, he level, water temperature, water flow and insulation temperature, and one label Y. A sliding window is created to improve the accuracy of the model. The sliding window is 5 data long. The final input data is selected as an average value of He pressure in the sliding window, a standard deviation of He pressure in the sliding window, an original value of He pressure, and a standard deviation of insulation layer temperature. Programming is based on Python, with a cost index C of 4 and a gamma value of 300. Model results are shown in the confusion matrix and ROC curve as shown in table 1.
TABLE 1
According to the confusion matrix:
the ROC curve is shown in fig. 7.
2. Decision tree
And taking the abnormal helium pressure detection result as a dependent variable, and taking water temperature, water flow, helium liquid level, helium pressure, cold pressure head and shielding data as independent variables, so that a CART model is established. Wherein, when distinguishing between normal points and abnormal points, the collected data are divided at fixed intervals, and 51591 sets of data are obtained in total, wherein 48510 sets of data are regarded as normal fluctuations in helium pressure, and the remaining 3081 sets of data are regarded as abnormal fluctuations in helium pressure. Because of the non-uniform data distribution between the two categories, to avoid sample skew, it determines the appropriate parameters by calculating the loss function, thus balancing the weights between the categories. To prevent overfitting, it limits the depth of the decision tree and the number of leaf nodes, and finally, 41271 sets of data are selected as training sample sets and 10320 sets of data are selected as test sample sets.
The classification results obtained by this method are shown in table 2 and fig. 8:
TABLE 2
Through the confusion matrix and the ROC curve, the accuracy and the reliability of the CART method in the detection of the normal state of the MRI model are verified. However, for CART method, the accuracy of outlier detection is not ideal.
3. Random Forest (RF)
There are three problems to consider in this prediction. One is the effect of the algorithm parameters on the outcome, e.g., the number, depth, or whether the decision tree in RF will be overfitted. Second is how the predicted outcome will change with increasing lead time. A third problem is how the accuracy of the model varies with different combinations of parameters.
First, the study explores the parameters of the model itself, and it is found that the parameters of the model have little effect on accuracy due to the large data volume, and do not exceed the model because the depth is not limited. Thus, there is no need to analyze too much of the model parameters.
Secondly, the influence of different advance periods on model precision is discussed, and sensitivity analysis is carried out on the advance time in a reasonable interval, so that as shown in fig. 9, the overall precision is improved along with the increase of the advance time, and the optimal advance time value range is 400-500.
For the third problem, fig. 10 shows model accuracy for different parameter combinations in horizontal coordinates (lead time=100), where H represents He pressure, S represents Shield Si410, L represents He level, F represents water flow, and T represents water temperature. HS represents a combination of H and S, i.e., a combination of parameters He pressure and Shield Si410, and so on. The results show that the best performance model accuracy is a two parameter combination of He pressure and Shield Si410 or a three parameter combination of He pressure, shield Si410 and He level. The accuracy score and recall score for HS were 0.9 and 0.5, respectively, while the accuracy and recall score for HSL were 0.51 and 0.88, respectively. Fig. 11 is a confusion matrix for HS and HSL. Fig. 12 is the ROC curves for HS and HSL.
Comparing these three prediction methods, it is apparent that the RF model is superior to the simple decision tree model and the SVM model in predicting outliers. For decision tree models, the prediction results are not ideal. Although the prediction accuracy of the SVM model exceeds 80%, the recall rate is only about 20%. Meanwhile, for the RF model, based on a comparison of different parameter combinations at the same lead period, two sets of parameter combinations are finally selected: a combination of He pressure and Shield Si410 (HS), and a combination of He pressure, shield Si410, and He level (HSL). The HS accuracy score is better than the HSL score, which recall score is higher than the HS score, but the difference is not significant.
According to the embodiment and experimental example, the method for labeling the abnormal sequence and the method for constructing the characteristics are provided, the function of predicting the abnormality of the MRI equipment by using the time sequence data acquired by the Internet of things is realized, and the method has a good application prospect.
Claims (10)
1. An MRI equipment abnormality detection method based on internet of things data is characterized by comprising the following steps:
step 1, collecting time sequence data in the running process of MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO;
step 2, carrying out real-time prediction by adopting a sliding window algorithm, obtaining characteristics in a sliding window based on the time sequence data, and inputting the characteristics into an abnormality detection model to obtain the probability of equipment abnormality at a time point after the sliding window;
in the training process of the anomaly detection model, the marking method of training data comprises the following steps:
step a, dividing original time sequence data into a plurality of subsequences by using a sliding window;
step b, selecting an abnormal index for analysis and calculating a subsequence; the calculation mode of the abnormal index is as follows: calculating the slope between any two adjacent data points in the subsequence, and calculating the radius of the confidence interval;
and c, marking the subsequence as an abnormal sequence by using an index deviating from a normal value.
2. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: the device anomaly includes at least one of the following three situations:
1) The He pressure changes suddenly and rapidly rise, and rise again after stabilizing for a period of time;
2) He pressure is continuous, nonvolatile, rapidly increasing or decreasing;
3) He pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again.
3. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: the time series data includes He pressure and Shield Si410;
or, the time series data includes He pressure, shield Si410, and He level.
4. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: the characteristics include an average value of He pressure in the sliding window, a standard deviation of He pressure in the sliding window, an original value of He pressure, and a standard deviation of insulation layer temperature.
5. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: in step 1, the time series data is collected and then preprocessed, and the preprocessing step comprises compressing a plurality of values which are continuously unchanged into a single value.
6. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: the anomaly detection model adopts a random forest algorithm.
7. The method for detecting abnormality of MRI equipment based on internet of things data according to claim 1, characterized by: the time series data is that one data point is acquired every minute, and the window length of the sliding window takes 8-9 data points.
8. An MRI equipment anomaly detection system based on thing networking data, characterized by comprising:
the data acquisition and storage module is used for acquiring and storing time sequence data in the operation of the MRI equipment; the time series data comprises at least one of He pressure, he level, water flow, water temperature, insulating layer temperature, shield Si410 or cold head RuO;
the abnormality detection module is used for carrying out real-time prediction by using a sliding window algorithm, obtaining characteristics in a sliding window based on the time sequence data, inputting the characteristics into an abnormality detection model, and obtaining the probability of equipment abnormality at a time point after the sliding window;
in the training process of the anomaly detection model, the method for labeling training data comprises the following steps:
step a, dividing original time sequence data into a plurality of subsequences by using a sliding window;
step b, selecting an abnormal index for analysis and calculating a subsequence; the calculation mode of the abnormal index is as follows: calculating the slope between any two adjacent data points in the subsequence, and calculating the radius of the confidence interval;
and c, marking the subsequence as an abnormal sequence by using an index deviating from a normal value.
9. The system for detecting abnormality of MRI apparatus based on internet of things data according to claim 8, wherein: the device anomaly includes at least one of the following three situations:
1) The He pressure changes suddenly and rapidly rise, and rise again after stabilizing for a period of time;
2) He pressure is continuous, nonvolatile, rapidly increasing or decreasing;
3) He pressure will remain stable for a relatively long period of time, then rise or fall rapidly, and then stabilize again.
10. A computer-readable storage medium, characterized by: a computer program for implementing the method for detecting an abnormality of an MRI apparatus based on internet of things data according to any one of claims 1 to 7, or a computer program for implementing the system for detecting an abnormality of an MRI apparatus based on internet of things data according to claim 8 or 9, stored thereon.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310103018.3A CN116256686A (en) | 2023-02-03 | 2023-02-03 | MRI equipment abnormality detection method and system based on Internet of things data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310103018.3A CN116256686A (en) | 2023-02-03 | 2023-02-03 | MRI equipment abnormality detection method and system based on Internet of things data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116256686A true CN116256686A (en) | 2023-06-13 |
Family
ID=86680413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310103018.3A Pending CN116256686A (en) | 2023-02-03 | 2023-02-03 | MRI equipment abnormality detection method and system based on Internet of things data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116256686A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060122472A1 (en) * | 2004-09-28 | 2006-06-08 | Pullman Seth L | System and method for clinically assessing motor function |
| KR101615085B1 (en) * | 2015-08-20 | 2016-04-26 | 주식회사 베가스 | Alarm system and method for equipment abnormality |
| WO2018004580A1 (en) * | 2016-06-30 | 2018-01-04 | Intel Corporation | Device-based anomaly detection using random forest models |
| CN109947079A (en) * | 2019-03-20 | 2019-06-28 | 阿里巴巴集团控股有限公司 | Region method for detecting abnormality and edge calculations equipment based on edge calculations |
| US20220128983A1 (en) * | 2020-10-28 | 2022-04-28 | Boe Technology Group Co., Ltd. | Defect prediction methods, apparautses, electronic devices and storage media |
| CN115359852A (en) * | 2022-07-22 | 2022-11-18 | 北京科技大学 | Method for predicting local anode effect of large aluminum electrolytic cell |
-
2023
- 2023-02-03 CN CN202310103018.3A patent/CN116256686A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060122472A1 (en) * | 2004-09-28 | 2006-06-08 | Pullman Seth L | System and method for clinically assessing motor function |
| KR101615085B1 (en) * | 2015-08-20 | 2016-04-26 | 주식회사 베가스 | Alarm system and method for equipment abnormality |
| WO2018004580A1 (en) * | 2016-06-30 | 2018-01-04 | Intel Corporation | Device-based anomaly detection using random forest models |
| CN109947079A (en) * | 2019-03-20 | 2019-06-28 | 阿里巴巴集团控股有限公司 | Region method for detecting abnormality and edge calculations equipment based on edge calculations |
| US20220128983A1 (en) * | 2020-10-28 | 2022-04-28 | Boe Technology Group Co., Ltd. | Defect prediction methods, apparautses, electronic devices and storage media |
| CN115359852A (en) * | 2022-07-22 | 2022-11-18 | 北京科技大学 | Method for predicting local anode effect of large aluminum electrolytic cell |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | A two-stage data-driven-based prognostic approach for bearing degradation problem | |
| Costello et al. | Machine learning model for event-based prognostics in gas circulator condition monitoring | |
| CN109460574A (en) | A kind of prediction technique of aero-engine remaining life | |
| Yang et al. | A hybrid model-based fault detection strategy for air handling unit sensors | |
| CN110458039A (en) | A kind of construction method of industrial process fault diagnosis model and its application | |
| CN116069079A (en) | Intelligent heat dissipation control method and system for intelligent switch cabinet | |
| CN115358281B (en) | Machine learning-based cold and hot all-in-one machine monitoring control method and system | |
| CN109737045A (en) | Air compressor machine fault early warning method, system and relevant apparatus applied to chip production | |
| US20190086145A1 (en) | Freezer with remote management | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN116256686A (en) | MRI equipment abnormality detection method and system based on Internet of things data | |
| CN111192163B (en) | Generator reliability medium-short term prediction method based on wind turbine generator operating data | |
| CN111460027B (en) | Intelligent dynamic monitoring method and system suitable for energy Internet | |
| CN116259161A (en) | Power failure early warning system | |
| CN113536050B (en) | Distribution network monitoring system curve data query processing method | |
| JP2017117034A (en) | Diagnosis device and diagnostic method | |
| CN115795999A (en) | Performance abnormity early warning method for long-term service pumped storage unit | |
| CN112260211B (en) | Method and device for determining operating state of relay protection device | |
| CN116401604B (en) | Method for classifying and detecting cold head state and predicting service life | |
| KR102486463B1 (en) | Method and Apparatus for Real Time Fault Detection Using Time series data According to Degradation | |
| Katırcıoğlu et al. | Fault diagnosis for overcharge and undercharge conditions in refrigeration systems using infrared thermal images | |
| CN115795067B (en) | Storage method for temperature measurement data | |
| Song et al. | Health indicator construction and remaining useful life prediction for space Stirling cryocooler | |
| CN114528914B (en) | Method, terminal and storage medium for monitoring state of cold water host in loop | |
| CN117260053B (en) | Special steel welding control method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |