Multi-criteria fair queuing of alarms

Abstract

A method (100) of generating a ranked list (42) of alerts (30) from files (32) of electronic devices (12) includes applying a predictive model (34) to data of the files to generate maintenance alerts for the electronic devices; assigning a discrete value to each maintenance alert for a set of criteria (36), wherein the maintenance alerts have a share that is computed as a product of the assigned discrete values; performing a credit-based queuing process (38) to assign the maintenance alerts to time slots of a ranked list, wherein in the credit-based queuing process the maintenance alerts accumulate credit according to the share computed for the respective maintenance alert; and displaying the maintenance alerts as the ranked list on a display device (24) in operative connection with the at least one electronic processor according to the assignment of the maintenance alerts to the time slots of the ranked list.

Description

Technical Field

[0001] The following generally covers the fields of maintenance, medical equipment, medical equipment maintenance, weighted fair queuing, multi-criteria optimization, and related fields. Background Technology

[0002] Maintenance of medical imaging systems (and other medical devices such as patient monitoring systems) involves several types of maintenance activities. Maintenance can include planned activities where a Field Service Engineer (FSE) visits the hospital periodically (e.g., once or twice a year, or at a frequency determined by system usage, or dynamically scheduled based on the condition of the system being remotely monitored) to perform tasks such as lubrication, cleaning, and calibration. Additionally, there are corrective maintenance activities initiated in response to problems reported by the hospital. If the problem is severe, this can result in unplanned downtime of the system. The system may become inoperable until the problem is resolved remotely or through a field visit by the FSE. Unexpected downtime can result in significant costs for the hospital because examinations cannot be scheduled for a period of time. It can also lead to patient dissatisfaction, as examinations may have to be rescheduled to a later date.

[0003] In recent years, in addition to the maintenance activities mentioned above, predictive maintenance activities have been used more frequently to try to avoid unplanned downtime. Predictive models have been developed for various parts of medical imaging systems, designed to predict when a part might fail in the near future, allowing for preventative replacement before actual failure. These predictive models can be built using machine learning (ML) algorithms, which construct the predictive model based on a training set of historical cases. For a given subsystem or part... Such a predictive model will be applied to a given system and a given time window Estimation of subsystem or system Part The probability of failure within this time window These estimated probabilities can then be used to determine the precautionary replacement of the subsystem / part in the following week or more. Does it make sense? Predictive models use log event data generated by the system for remote analysis. Log event data can include sensor measurements as well as log events in the form of low-level error and warning messages.

[0004] Once the predictive model has been tested and found to perform at a sufficient level of performance (considering the probability and cost of false positives and false negatives), it can be deployed to monitor numerous systems in the field. The model can be run at regular intervals on each system's most recent log event data, e.g., hourly or daily, or dynamically triggered by the availability of new and relevant data. If for a system... It concluded If a certain threshold is exceeded, it can issue an alarm. For example, in... Exceeding the threshold in at least a certain number of consecutive time units Alternative strategies can also be used to trigger alerts.

[0005] Specialized Remote Monitoring Engineers (RMEs) are trained to review triggered alarms using computer programs that display a sorted list of recently triggered alarms. Priority Associated with type For each alert, all alerts triggered by multiple prediction models are sorted by priority, for example, the list of most recent alerts is sorted by non-increasing priority: at the top of the list, one or more alerts with the highest priority are shown, and the priority decreases as you go down the list. RME typically considers alerts in a top-down manner. Alert types Based on the prediction model, and similarly, prioritization. Based on the parameters of the prediction model, such as accuracy, false positive rate (FPR), and window size. .

[0006] The drawback of using only alarm priority to sort alarms in an alarm list is that it fails to consider many other factors that may play a role in the rational sorting of alarms. For example, in addition to priority, considering the remaining time the service organization must have to respond can be useful, so as not to risk having parts replacement still scheduled too late to avoid actual failure. Therefore, a deadline is attached in addition to priority. Furthermore, other aspects that may not be considered by an alarm system that uses prioritization, such as the type of service contract the hospital has for a given system, can exist.

[0007] To emphasize deadlines, alerts can be sorted in a list using algorithms such as earliest deadline first. However, this approach only considers deadlines and not priorities. Alternatively, various scheduling algorithms exist that prioritize events directly based on deadlines, allowing high-priority activities to have shorter deadlines by definition.

[0008] Another drawback of prioritizing alerts in an alert system is that some alerts may never be observed by the RME because they might be ranked so low that they never reach the top of the list. It's possible that some predictive models no longer function for the conditions they were originally designed for. For example, the data used might have changed after a new software release. Therefore, such models could start generating more alerts. However, if the predictive model has a low priority, this might only be detected later. Whenever the RME evaluates the correctness of alerts generated by the predictive models, it assesses whether the predictive models still function as expected during their construction. Therefore, it would be beneficial to periodically evaluate alerts generated by all models, assuming they are at different rates.

[0009] Another drawback is that model-priority-based ranking doesn't account for the fact that some hospitals may have multiple similar imaging systems, allowing them to reschedule examinations to another similar system, while other hospitals may only have one system of a given type, meaning they cannot perform these examinations due to a failure in that single system until the problem is resolved. Therefore, a high ranking of alarms related to an imaging system may not be optimal when a hospital has several other imaging systems of that modality. Similarly, a low ranking of alarms related to an imaging system (which is the only modality owned by the hospital in that modality) may also be suboptimal.

[0010] The following discloses some improvements to overcome these and other problems.

[0011] Document EP3379356A1 discloses a method for a lithography apparatus. This method determines causal relationships between events (AD) in a time series of multiple parameters associated with an industrial process, identifies at least a first event (A, B, D) associated with a parameter offset event; identifies at least a second event (C) associated with a failure event; and establishes the causal relationship between the first event (A, B, D) and the second event (C) by using a determined value of the transfer entropy between the first event (A, B, D) and the second event (C). Summary of the Invention

[0012] In one aspect, a non-transient computer-readable medium storing instructions executable by at least one electronic processor to perform a method for generating a sorted list of alarms from a file of an electronic device. The method includes: applying a predictive model to data in the file to generate maintenance alarms for the electronic device; assigning discrete values ​​to each maintenance alarm according to a set of criteria, wherein the maintenance alarm has a share calculated as a product of the assigned discrete values; performing a credit-based queuing process to allocate the maintenance alarms to time slots in the sorted list, wherein, in the credit-based queuing process, the maintenance alarms accumulate credit based on the share calculated for each respective maintenance alarm; and displaying the maintenance alarms as the sorted list on a display device operatively connected to the at least one electronic processor, according to the allocation of time slots from the maintenance alarms to the sorted list.

[0013] In another aspect, a non-transient computer-readable medium storing instructions capable of being executed by at least one electronic processor to perform a method of generating a sorted list of alarms from a file in an electronic device. The method includes: applying a predictive model to data in the file to generate maintenance alerts for the electronic device; assigning discrete values ​​to each maintenance alert according to a set of criteria, wherein the maintenance alert has a share calculated as the product of the assigned discrete values; performing a credit-based queuing process to assign the maintenance alerts to time slots in a sorted list, wherein, in the credit-based queuing process, the maintenance alerts accumulate credit based on the share calculated for each corresponding maintenance alert, wherein the credit-based queuing process is a credit-based fair queuing process, wherein each iteration of the credit-based fair queuing process includes: increasing the credit of each maintenance alert by an amount equal to the share calculated for the maintenance alert; assigning the maintenance alert with the highest accumulated credit to the current time slot of the sorted list; and after the assignment, decreasing the credit of the maintenance alert assigned to the current time slot by a predetermined amount; and displaying the maintenance alerts as part of the sorted list on a display device operatively connected to the at least one electronic processor according to the assignment of the maintenance alerts to the time slots of the sorted list.

[0014] In another aspect, a method for generating a sorted list of alarms from files of an electronic device includes: applying a predictive model to data of the files to generate maintenance alarms for the electronic device; assigning discrete values ​​to each maintenance alarm according to a set of criteria, wherein the maintenance alarms have a share calculated as a product of the assigned discrete values, wherein the set of criteria includes: (1) a predictive model criterion whose assigned values ​​for the maintenance alarms correspond to the predictive model that generated the maintenance alarms, and (2) a time range by which the predictive model assigns the maintenance alarms; performing a credit-based queuing process to assign the maintenance alarms to time slots in the sorted list, wherein, in the credit-based queuing process, the criterion... Maintenance alarms accumulate credit based on a share calculated for the corresponding maintenance alarm, wherein the credit-based queuing process is a credit-based fair queuing process, wherein each iteration of the credit-based fair queuing process includes: increasing the credit of each maintenance alarm by an amount equal to the share calculated for the maintenance alarm; assigning the maintenance alarm with the highest accumulated credit to the current time slot of the sorted list; and after the assignment, reducing the credit of the maintenance alarm assigned to the current time slot by a predetermined amount; and displaying the maintenance alarm as part of the sorted list on a display device operatively connected to the at least one electronic processor according to the assignment of the maintenance alarm to the time slot of the sorted list.

[0015] One advantage is that it reduces downtime for medical imaging equipment.

[0016] Another advantage is that it allows for the prioritization of medical imaging device alarms using criteria other than priority.

[0017] Another advantage is the ability to sort alarms from multiple modalities of medical imaging devices.

[0018] Another advantage is the ability to prioritize medical device alarms using multiple criteria.

[0019] Another advantage is that remote experts can review the sorted list of alerts and provide assistance to local operators of medical devices.

[0020] The given embodiments may provide zero, one, two, or all of the foregoing advantages, and / or may provide other advantages that will become apparent to those skilled in the art upon reading and understanding this disclosure. Attached Figure Description

[0021] This invention can take the form of various components and their arrangements, as well as various steps and their scheduling. The accompanying drawings are for illustrative purposes only and should not be construed as limiting this disclosure.

[0022] Figure 1An illustrative system for generating an ordered list of alerts from log files of medical electronic devices, according to the present disclosure, is shown schematically.

[0023] Figure 2 It shows Figure 1 An exemplary flowchart operation of the system. Detailed Implementation

[0024] The medical imaging equipment service department has RME (Real-Time Equipment Manager) staff who review automatically generated alerts based on machine logs of the imaging equipment assigned to them. For this purpose, predictive models are applied to the log data. For a given subsystem / part... Predictive model output subsystem / part In the time window Probability of failure ,in, This is the window width. In a typical scenario, an alert is generated if the probability exceeds a certain threshold (which can be different for different models). Currently, each alert ( ) are assigned by priority And present the probability to the RME. A sorted list of alarms for the assigned machines.

[0025] The problem with this approach is that the probability doesn't account for other relevant factors (e.g., deadlines for resolving alerts, the impact on the hospital or radiology lab if the section actually fails, etc.). Furthermore, priorities are typically fixed, so top-ranked alerts will remain top-ranked daily. Since a given RME might receive dozens of alerts of different priorities each day, this could cause the RME to miss lower-priority alerts that never even appear near the top of the list.

[0026] To address these problems, a class of wired or wireless queuing methods, known as fair queuing methods, is disclosed below to adapt to the alarm sequencing problem. Fair queuing methods allocate consecutive communication channel slots to data streams in such a way that each data stream is given a predetermined share of the channel due to that data stream, but this is done in such a way that the highest priority data stream does not receive all channel slots. For this purpose, a credit-based fair queuing method is described below, where each data stream has a maintained amount of credit (appropriately initially set to their respective shares). For each iteration (corresponding to a slot), the credit of each channel increases its share. The channel with the highest accumulated credit is allocated a slot, and its credit decreases by 1.00. Since the shares of all data streams sharing a channel are normalized to a sum of 1.00, this means that for each iteration, the total credit on all channels remains unchanged (because the total credit added in each iteration is 1.00, and this amount of 1.00 is precisely subtracted from the channel receiving the slot).

[0027] When making fair queuing adapt to the sorting of alarms, each alarm corresponds to a data stream, a channel corresponds to a sorted list, and each channel slot corresponds to a position in the sorted list. In one approach, the (normalized) priority of each alarm corresponds to the share of that alarm, and fair queuing thus ensures that the highest priority alarms are not repeatedly sorted at the top of the sorted alarm list.

[0028] However, this method does not take into account the various other factors that may be useful in sorting.

[0029] To consider multiple factors in the ranking process, an improved fair queuing method known as multi-criteria fair queuing is implemented in some embodiments disclosed herein. In this method, each criterion is assigned a set of assumed discrete criterion share values, the sum of which is 1.00 (i.e., the assumed criterion share value for each criterion is normalized). The criterion share value is represented as... ,in, Index the criteria, and To Indexing criteria are based on assumed values. Alert categories. Defined as the shared value of all criteria All possible combinations. As an example, if there are two criteria, one criterion has two possible criterion share values. and And another criterion has three possible criterion share values. , and Therefore, there are 2 × 3 = 6 alarm categories. The share of each alarm category is calculated as the product of the criterion share values ​​for that alarm category. Due to the normalization of the assumed criterion share values ​​for each criterion, this results in a total calculated share of 1 for each alarm category. Therefore, the resulting calculated shares can be used normally in the fair queuing algorithm. (Note that if a particular criterion share value does not appear in the list of alarms, that criterion share value is set to 0.0 and renormalization is performed).

[0030] Advantageously, as long as the updated criterion share value is renormalized, the criterion share value of a given criterion can be redistributed or completely redesigned without affecting any other criterions. Even if the value of any criterion cannot be determined, it can be effectively removed by setting its criterion share value to 1, since this has no effect on the product of the (remaining) criterion share values.

[0031] refer to Figure 1 This illustrates an apparatus or system 1 for generating an ordered list of alerts from a log file of a medical electronic device 2 (e.g., a medical imaging device). Figure 1 As shown, the local operator (LO) operating the medical imaging equipment (also known as an image acquisition device, imaging device, etc.) 2 is located in the medical imaging equipment compartment 3, and the remote monitoring engineer (RME) is located in a remote service location or center 4. The remote location 4 is typically a site owned, leased, or otherwise controlled by the imaging equipment supplier or other imaging equipment service provider.

[0032] A single image acquisition device 2 is illustrated, which can be a magnetic resonance (MR) image acquisition device, a computed tomography (CT) image acquisition device, a positron emission tomography (PET) image acquisition device, a single-photon emission computed tomography (SPECT) image acquisition device, an X-ray image acquisition device, an ultrasound (US) image acquisition device, or another modality of medical imaging device. Imaging device 2 can also be a hybrid imaging device, such as a PET / CT or SPECT / CT imaging system. Although in Figure 1The illustration shows a single image acquisition device 2, but more typically, a service location or center 4 will provide monitoring and maintenance services for a group of imaging devices, such as all or a subset of medical imaging devices sold by a supplier under a service contract between a customer (e.g., the owner or operator of the imaging chamber 3). The group served by the service location or center 4 can have dozens or hundreds or more imaging devices. A given customer (e.g., a medical imaging laboratory) will have one or (more typically) multiple image acquisition devices, which may have the same and / or different imaging modalities. For example, if a hospital performs many CT imaging examinations and relatively fewer MRI examinations and even fewer PET examinations, the hospital's imaging laboratory (sometimes called a "radiology laboratory" or some other similar nomenclature) may have three CT scanners, two MRI scanners, and only a single PET scanner. This is merely an example. Furthermore, a remote service center 4 will typically provide services to multiple hospitals or other customers.

[0033] In a typical scenario, the various imaging devices 2 served by service center 4 automatically generate machine logs that record operational information related to the imaging device 2, such as the scans performed, the hardware and parameters used in the scans, and changes in device configuration. As a non-limiting illustrative example, in the case of a CT scanner, the machine logs may store X-ray tube usage, including the time intervals at which the X-ray tube is energized and parameters such as X-ray tube voltage and current. These logs are automatically transmitted to service center 4 via communication link 14 (such as the Internet, possibly enhanced by a local area network at client 3 and / or service center 4). These logs may be transmitted, for example, daily or more frequently or less frequently, and are used as input data for predictive models running at service center 4 to generate alerts.

[0034] Figure 1 Also shown is a remote service center 4 comprising one or (typically) multiple workstations 12, such as electronic processing devices, smartphones, tablets, workstation computers, or more generally computers operated by an RME, operationally connected to receive data related to alarms from the medical imaging device 2, as described in more detail below. Alternatively or additionally, the remote workstation 12 may be implemented as a server computer or multiple server computers 16, multiple server computers 16 (e.g., interconnected to form a server cluster), cloud computing resources, etc. The workstation 12 includes typical components such as an electronic processor 20 (e.g., a microprocessor), at least one user input device (e.g., a mouse, keyboard, trackball, etc.) 22, and at least one display device 24 (e.g., an LCD display, a plasma display, a cathode ray tube display, etc.). In some embodiments, the display device 24 may be a separate component from the workstation 12. The display device 24 may also include two or more display devices.

[0035] Electronic processor 20 is operatively connected to one or more non-transient storage media 26. As a non-limiting illustrative example, non-transient storage media 26 may include one or more of the following: disk, RAID, or other magnetic storage media; solid-state drives, flash drives, electrically erasable read-only memory (EEROM), or other electronic storage; optical disks or other optical storage devices; various combinations thereof; and so on; and may be, for example, a network storage device, the internal hard disk drive of workstation 12, various combinations thereof, etc. It should be understood that any reference herein to one or more non-transient media 26 should be interpreted broadly to cover a single medium or multiple media of the same or different types. Similarly, electronic processor 20 may be embodied as a single electronic processor or two or more electronic processors. Non-transient storage media 26 stores instructions executable by at least one electronic processor 20. Instructions include instructions for generating a graphical user interface (GUI) 28 for display on a remote expert display device 24.

[0036] Service center 4 may be located in a single physical location or may be geographically distributed. Service center 4 includes server computer 16 (or server cluster, cloud computing resources including servers, etc.), which is programmed to perform functions such as applying predictive models to machine log data generated by imaging equipment 2 served by center 4 to generate alerts, and providing a user interface (UI) on workstation 12 equipped by RME. Server computer 16 is operatively connected to one or more non-transient storage media 18. As a non-limiting illustrative example, non-transient storage media 18 may include disks, RAID or other magnetic storage media; solid-state drives, flash drives, electrically erasable read-only memory (EEROM) or other electronic storage; optical discs or other optical storage devices; various combinations thereof; and one or more of the like; and may be, for example, network storage devices, internal hard drives of server computer 16, various combinations thereof, etc. It should be understood that any reference herein to one or more non-transient media 18 should be interpreted broadly to cover a single medium or multiple media of the same or different types. Similarly, server computer 16 may be embodied as a single electronic processor or two or more electronic processors. Non-transient storage medium 18 stores instructions that can be executed by at least one server computer 16.

[0037] Furthermore, as disclosed herein, server computer 16 executes a method or process 100 for generating an ordered list of alarms 30 from a file 32 representing the illustrative medical imaging device 2 stored in a non-transient computer-readable medium 31 of the medical device controller 10. File 32 may include, for example, machine log files, basic data (which specifies how many medical devices of a given type are owned by a given customer), contract data (i.e., between the customer and service provider), or any other suitable type of data. Additionally, the file may include data relating to the capabilities of the corresponding Remote Monitoring Engineer (RME). Server computer 16 is configured to retrieve file 32 from the medical device controller 10 via communication link 14, for example, daily or more frequently or less frequently. In this way, the Remote Monitoring Engineer (RME) can view the ordered list of alarms 30 and can take appropriate actions, such as attempting to remotely repair the alarms or contacting the customer to schedule an on-site service engineer (FSE) visit to handle the alarms on-site.

[0038] refer to Figure 2 And continue to refer to Figure 1 The server computer 16 is configured, as described above, to execute method 100 for generating a sorted list of alerts 30 from file 32 of medical imaging device 2. Non-transient storage medium 18 stores instructions that can be read and executed by the server computer 16 to perform the disclosed operations, including executing method or process 100. In some examples, method 100 may be executed at least partially via cloud processing.

[0039] At operation 102, one or more prediction models 34 are applied to the data in the retrieved file 32 to generate a maintenance alert 30 for the medical electronic device 2. The prediction models 34 are stored on a non-transient computer-readable medium 18 of the server computer 16. The prediction models 34 may include, for example, one or more statistical models that statistically predict the remaining X-ray tube lifespan based on historical data of past X-ray tube installations and failures, such as time-integrating tube current, time-integrating tube voltage, and / or other measurable parameters of the X-ray tube. The prediction models 34 may include one or more artificial intelligence (AI) models, for example, that predict the replacement time of the X-ray detector array (or array submodule) by applying a convolutional neural network (CNN) to an image (if available in data 32 on the server 16 or from another source) to detect image artifacts indicating impending performance degradation that the CNN has been trained on historical data to detect. Other types of AI, such as support vector machines (SVMs), may be trained to generate alerts based on training on historical data. The prediction models 34 may include less complex models, such as alerts based solely on time since installation, based on available component lifespan information provided by the component manufacturer. In this scenario, the predictive model reads the component's installation date from the machine logs and determines the alarm based on the time elapsed since that installation date. These are merely some non-limiting illustrative examples.

[0040] In operations 104 and 106, a credit-based queuing process 38 is applied to rank the alarms. At operation 104, discrete values ​​are assigned to each maintenance alarm 30 according to a set of criteria 36 (stored in a non-transient computer-readable medium 18 of the server computer 16). The set of criteria 36 may include, for example, a predictive model criterion whose assigned value for the maintenance alarm 30 corresponds to a predictive model 34 that generated the maintenance alarm, and a time range for which the predictive model 34 assigned the maintenance alarm. In some embodiments, the set of criteria 36 may further include criteria based on the maintenance contract for the electronic device 2 that generated the maintenance alarm 30, such as the terms of the maintenance contract, the nature of the contract on which the electronic device 2 is based, etc. In another example, the set of criteria 36 may include criteria based on the satisfaction level of the owner of the electronic device 2 that generated the maintenance alarm 30. For example, the satisfaction level may include criteria based on the potential downtime of the electronic device 2. For this purpose, the credit-based queuing process 38 may utilize machine learning algorithms to analyze the generated maintenance alarms 30 to predict possible downtime. Higher downtime may be due to the complexity of the service situation (i.e., more than one possible root cause or more than one set of contributing factors); available FSE resources in the area; part availability (generally or in the area), etc. In other embodiments, when the medical imaging device 2 includes different modalities (e.g., MR and CT), the set of criteria 36 may also include criteria based on the imaging modality of the medical imaging device 2 for which maintenance alarm 30 was generated (e.g., type of modality, number of medical imaging devices, etc.). In another example, the set of criteria 36 may include criteria based on the volume of examinations using the medical imaging device 2 relative to the number of medical imaging devices. This volume of examinations may be analyzed retrospectively and / or prospectively.

[0041] Maintenance alarm 30 has a share calculated as the product of assigned discrete values. For this purpose, each of the plurality of criteria 36 assumes a discrete value in the set of discrete values, where the sum of the discrete values ​​in the set of discrete values ​​is equal to 1 (e.g., 1.00). That is, each criterion may have an initial share value less than 1. In some examples, the set of discrete values ​​may be recalculated for a single criterion while ensuring that the sum of the discrete values ​​in the set of discrete values ​​remains equal to 1. Maintenance alarm 30 with zero or redundant values ​​may be discarded.

[0042] Further reference Figure 2At operation 106, a credit-based queuing process 38 is performed to assign maintenance alarms 30 to time slots 40 in a sorted list 42. In the credit-based queuing process 38, maintenance alarms 30 accumulate credit based on a share calculated for the corresponding maintenance alarm. In some embodiments, the credit-based queuing process 38 includes a credit-based fair queuing process. In each iteration of the credit-based fair queuing process 38, the credit of each maintenance alarm 30 increases by an amount equal to the share calculated for the maintenance alarm. The maintenance alarm 30 with the highest accumulated credit is assigned to the current time slot 40 in the sorted list 42. Once that maintenance alarm 30 is assigned to a time slot 40, the credit of the assigned maintenance alarm is reduced by a predetermined amount, such as 1.00. The credit-based fair queuing process 38 is repeated until all maintenance alarms 30 have been assigned to their corresponding time slots 40 in the sorted list 42.

[0043] Return to reference Figure 1 and 2 At operation 108, based on the time slot allocation from maintenance alarms to the sorted list, the sorted list 42 is displayed on the display device 24 of the remote workstation 12. The remote expert (RME) can review the list 42 and provide assistance or draw attention to the local operator (LO) of the corresponding medical imaging device 2 corresponding to the maintenance alarm 30 on the list 42.

[0044] Example

[0045] The credit-based queuing process 38 performed by method 100 is described in more detail below. The credit-based fair queuing process 38 is generalized as a multi-criteria variant. The credit-based fair queuing process 38 can be used if multiple processes must share scarce resources in a fair manner (e.g., remotely monitoring the attention of engineers (RMEs)). It is typically performed by processes numbered... of Each data stream shares a single communication channel. Used in the context of data flow. Obtain the channel Fraction of total capacity ,in, .

[0046] The communication channel is numbered as The system consists of 40 consecutive time slots, each of which is allocated in the following manner. One of the data streams. During the allocation of time slot 40 to the data stream, for each data stream Maintaining credit . Represents data stream In the time slot Initial credit. All data streams begin with zero credit, meaning that for all... , At the start of each time slot 40, each data stream Credit increase data stream The authorized share. Next, time slot 40 is allocated to the data stream with the highest credit, and the credit of that data stream is reduced by 1. Therefore, during each time slot 40, by reducing the credit of the data stream receiving the current time slot by 1, the sum of all credits increases. And decrease by 1 again. Therefore, when Then, for all , Therefore, each data stream receives an average of 40 of its fair share of time slots, and the credit of each data stream remains bounded at the beginning of each time slot. Thus, only those with small shares (e.g., small shares) are eligible for certain benefits. The data stream still receives an average of a small fraction of the time slots equal to that share. Fair queuing algorithm 38 is slightly more complex if one of the data streams has no data packets to send.

[0047] As disclosed herein, this principle can be used to rank alarms 30 from multiple predictive models 34. This can be performed for a single criterion (as illustrated in the example) as follows. There exists a criterion numbered... of There are 34 prediction models, of which 34 are derived from the model. The priority of alarm 30 is determined by Given, where, for each , ,and Now, at any given time point, consecutive time slots 40 in the sorted list 42 can be assigned to the pattern using fair queuing. The alert, where each model's share is determined by priority. Given. Each model, on average, ranks within the top 42 of the sorted list. Each individual's fair share. However, in this case, only the priority criterion is considered.

[0048] Several more or less independent criteria can now be considered, including, for example: the predictive model 34 that generates the alert (formerly known as priority), the deadline for alert 30, the type of customer contract, "customer temperature," the number of similar systems the customer has, the modalities with which the current RME has expertise, etc. As used herein, the term "customer temperature" refers to a measure of how satisfied the customer (i.e., the hospital) is with the current maintenance services they have experienced so far this year. This can be measured, for example, by a score of unplanned downtime. Since the maintenance service contract between the hospital and the maintenance service provider can be defined, for example, by an upper limit on the allowable score of unplanned downtime, customer temperature can be measured by the total unplanned downtime the system has experienced so far during the current year.

[0049] In addition, regarding the (For example, 6 guidelines 36) give each maintenance alarm 30 (represented as an alarm) a Calculate the multi-criteria share. For each criterion, a relatively small number of values ​​that the criterion can achieve are defined. For the predictive model criterion, there is simply a set of different predictive models. For the deadline criterion, deadlines can be categorized as follows: deadlines within one week, deadlines between one and two weeks, etc. For the customer contract criterion, the number of contract types, etc., can be used.

[0050] For each criterion, a share must be allocated to each possible value such that the sum of shares over all values ​​for a given criterion totals 1 (1.00). The share for the given criterion value is obtained by simply multiplying the shares of all criterions 36 for a given alarm 30. Using this shared selection, the fair queuing algorithm can be used again. For each individual criterion, consecutive values ​​will still receive their fair share in the allocation.

[0051] As explained for the case of a single criterion, one or more values ​​of the criterion do not appear in the current list 42 of the alerts. In this case, the share of the criterion's occurrence values ​​increases proportionally, so that they still add up to 1.

[0052] For example, a situation may involve two criteria: one criterion has two values. and And a criterion has three values. , and The share of the first criterion is determined by... and Given, and the share of the second criterion is determined by , and Given. For each of the two criteria, the sum of the shares is 1. From the... The alarm number 30 was given as a share. And the total share of alarms with this pair of values ​​is made up of The matrix below (Table 1) shows the total share of alarms for each possible value pair.

[0053]

[0054] Table 1

[0055] By summing the columns in Table 1, the data is allocated to... and The total shares remain 0.6 and 0.4 respectively. Similarly, by summing the rows in Table 1, the shares allocated to... , and Their total shares remain at 0.5, 0.3 and 0.2 respectively.

[0056] For example, if each of the six possible value pairs contains several alarms 30, the successive positions of the sorted list 42 can be determined as given in Table 2:

[0057]

[0058] Table 2

[0059] Considering each criterion individually, each value of the criterion receives its fair share. Except for the last column, the subsequent columns relate to combinations. The successive entries in these columns show how the creditworthiness of this combination changes for consecutive time slots. The last column shows the combinations to which successive time slots are assigned.

[0060] Assigned to The total number of time slots 40 is 6 in the first 10 positions, and is allocated to The total number of time slots is 4 in the first 10 positions. Similarly, allocated to The total number of time slots is 5, allocated to The total number of time slots is 3, and they are allocated to... The total number of time slots is 2.

[0061] If, for one or more values ​​of one or more criteria 36, ​​there are currently no queued alarms 30, then the shares can be reweighted for each criterion so that the sum of the shares again totals 1. For example, for the second criterion, there may be no alarms with... The alarm. In this situation... and The share is changed through the following:

[0062]

[0063] therefore, and And Table 1 was updated to Table 3.

[0064]

[0065] Table 3

[0066] The multi-criteria weighted fair queuing process 38 can also be used to adapt the list 42 of alarm sorting to the specific RME currently processing that list. Therefore, modalities can be added as additional criteria. If an RME has expertise in modal CT and interventional X-ray (IXR) but not in MR, a share value can be assigned. , and For another RME with the most expertise in MR and less expertise in CT, the share value can be allocated as follows: , and .

Claims

1. A non-transient computer-readable medium (18) storing instructions, said instructions, when executed by at least one electronic processor (16), causing said at least one electronic processor (16) to perform a method (100) for generating an ordered list (42) of alarms (30) based on a file (32) of a plurality of electronic devices (12), said method (100) comprising: The predictive model (34) is applied to the data in the file to generate maintenance alerts for the multiple electronic devices; For each maintenance alarm, discrete values ​​are assigned to a set of criteria (36), and a share corresponding to the priority of the maintenance alarm is calculated for each maintenance alarm, wherein the share for each maintenance alarm is calculated as the product of the assigned discrete values; A credit-based queuing process (38) is performed to allocate the maintenance alarms to time slots in a sorted list, wherein, during the credit-based queuing process, the maintenance alarms accumulate credit based on the share calculated for each maintenance alarm; and The maintenance alarm is displayed as the sorted list on a display device (24) operatively connected to the at least one electronic processor, according to the allocation of the time slots from the maintenance alarm to the sorted list.

2. The non-transient computer-readable medium (18) according to claim 1, wherein, Each of the criteria (36) assumes a set of discrete values ​​in a set of discrete values, wherein the sum of the discrete values ​​in the set of discrete values ​​is equal to 1.

3. The non-transient computer-readable medium (18) according to claim 2, wherein, The method (100) further includes: The set of discrete values ​​is recalculated for each individual criterion in the set of criteria (36), while ensuring that the sum of the discrete values ​​in the set of discrete values ​​remains equal to 1.

4. The non-transient computer-readable medium (18) according to any one of claims 1-3, wherein, The credit-based queuing process (38) is a credit-based fair queuing process, wherein each iteration of the credit-based fair queuing process includes assigning the maintenance alert to the time slot of the sorted list by: The credit for each maintenance alarm (30) will be increased by an amount equal to the share calculated for the maintenance alarm; The maintenance alert with the highest cumulative credit is assigned to the current time slot of the sorted list; and Following the allocation, the credit of the maintenance alarm allocated to the current time slot will be reduced by a predetermined amount.

5. The non-transient computer-readable medium (18) according to claim 4, wherein, The predetermined quantity includes 1.

00.

6. The non-transient computer-readable medium (18) according to any one of claims 1-3, wherein, The set of criteria (36) includes: a prediction model criterion, the allocation value of which for the maintenance alarm (30) corresponds to the prediction model that generates the maintenance alarm, and the time range that the prediction model assigns to the maintenance alarm.

7. The non-transient computer-readable medium (18) according to claim 6, wherein, The set of criteria (36) also includes one or more of the following: Based on guidelines for maintenance contracts for electronic devices (2) that generate maintenance alerts for the plurality of electronic devices (12); Based on criteria regarding the satisfaction level of the owners of the electronic devices for which the maintenance alarms are generated.

8. The non-transient computer-readable medium (18) according to claim 7, wherein, The criteria, based on the satisfaction level of the owners of the electronic devices (2) from among multiple electronic devices (12) for which maintenance alarms (30) are generated, include the potential downtime during the maintenance of the electronic devices, and the credit-based queuing process includes: Machine learning algorithms are used to analyze the generated maintenance alerts to predict the possible downtime of the electronic equipment.

9. The non-transient computer-readable medium (18) according to claim 6, wherein, The plurality of electronic devices (12) include medical imaging devices with various imaging modalities, and the set of criteria (36) further includes: Based on the criteria for the imaging modality of the medical imaging device for which the maintenance alarm is generated.

10. The non-transient computer-readable medium (18) according to claim 6, wherein, The plurality of electronic devices (12) includes a plurality of medical imaging devices (2), and the set of criteria (36) also includes the amount of examinations performed using the plurality of medical imaging devices relative to the number of medical imaging devices.

11. The non-transient computer-readable medium (18) according to any one of claims 1-3, wherein, The method (100) further includes: Discard maintenance alerts with a share of zero (30).

12. The non-transient computer-readable medium (18) according to any one of claims 1-3, wherein, The method (100) further includes: Remove maintenance alerts (30) that have shares with indeterminate values.

13. A method (100) for generating an ordered list (42) of alarms (30) based on files (32) of multiple electronic devices (12), the method comprising: The predictive model (34) is applied to the data in the file to generate maintenance alerts for the multiple electronic devices; For each maintenance alarm, discrete values ​​are assigned to a set of criteria (36), and a share corresponding to the priority of the maintenance alarm is calculated for each maintenance alarm, wherein the share for each maintenance alarm is calculated as a product of the assigned discrete values, wherein the set of criteria includes: a prediction model criterion, the assigned value for the maintenance alarm corresponding to the prediction model that generated the maintenance alarm, and a time range assigned to the maintenance alarm by the prediction model; A credit-based queuing process (38) is executed to allocate the maintenance alarms to time slots in a sorted list, wherein, in the credit-based queuing process, the maintenance alarms accumulate credit based on the share calculated for each maintenance alarm, wherein the credit-based queuing process (38) is a credit-based fair queuing process, wherein each iteration of the credit-based fair queuing process includes: The credit for each maintenance alarm (30) will be increased by an amount equal to the share calculated for the maintenance alarm; Assign the maintenance alert with the highest accumulated credit to the current time slot of the sorted list; and Following the allocation, the credit of the maintenance alarm allocated to the current time slot will be reduced by a predetermined amount; and The maintenance alarm is displayed as the sorted list on a display device (24) operatively connected to at least one electronic processor, according to the allocation of the time slots from the maintenance alarm to the sorted list.

Images