Methods and systems for maximizing the uptime of ophthalmic medical devices through predictive health monitoring and proactive preventative maintenance.

Abstract

Certain aspects of this disclosure provide techniques for predicting the likelihood of future failures of components in an ophthalmic medical device and for performing preventative maintenance on the device. Example methods typically involve receiving measurements of one or more operating parameters associated with the ophthalmic medical device. Predicting future failures of the ophthalmic medical device using one or more models. Generating predictions based at least in part on the received measurements of the one or more operating parameters. Taking one or more actions to perform preventative maintenance on the ophthalmic medical device based on the predicted future failures.

Description

[0001] introduction

[0002] Various aspects of this disclosure relate to ophthalmic medical devices, and more specifically to predicting the failure of such devices based at least in part on data obtained from them, and to performing preventative maintenance on these devices in response to the prediction of future failures, in order to maximize device uptime. Background Technology

[0003] Ophthalmic medical devices typically include devices, consoles, and / or systems for measuring or characterizing the anatomical properties of a patient's eye, performing surgery on a patient's eye to correct a diagnosed ophthalmic problem, and / or measuring patient outcomes after surgery. These devices include a variety of optical, electrical, and mechanical components that may need to be adjusted or replaced over time. For example, a light source may deteriorate over time (e.g., it may become dimmer or require more current to produce the same light intensity) and may adversely affect the device's effectiveness due to reduced detail captured by the device, increased time required to complete light-based treatments (e.g., laser-based retinal reattachment surgery), etc. In another example, the power provided by a battery may decrease over time, and noise in the output generated by other power supply components may increase over time; both of these can adversely affect the operation of the ophthalmic medical device due to the need for more frequent battery replacements or the introduction of power instability into mechanical components that may damage the device. In yet another example, mechanical components typically wear out over time. Blades typically become dull with use, vacuum pumps may become unable to produce an effective vacuum over time, gears may wear out, motor power may decrease, or the consistency of power provided by the motor may deteriorate, and so on. Furthermore, the performance of various components in ophthalmic medical devices may deteriorate in different ways due to usage patterns, environmental parameters, and other factors.

[0004] Because the performance of the optical, electrical, or mechanical components of ophthalmic medical devices typically deteriorates over time, these devices are often taken out of service periodically to replace these components and restore the device to its intended performance level. In many cases, the device is taken out of service for maintenance when a component fails or otherwise produces undesirable results, and the damaged component can be analyzed to determine the root cause of the failure. However, this can lead to unpredictable availability of the ophthalmic medical device to medical professionals, as professionals may not know or be able to predict when the device will need to be taken out of service, and multiple recurring failures may occur before the root cause of the component failure is discovered.

[0005] However, in some cases, preventative maintenance or other remedial actions can be performed to restore the performance level of an ophthalmic medical device to its expected level. For example, preventative maintenance can allow for the recalibration (automatic or manual) of components within the ophthalmic medical device to restore the performance of the recalibrated components to their expected level. Additionally, preventative maintenance can be used to identify components that need replacement before they fail or damage other related components within the ophthalmic medical device. However, preventative maintenance and other remedial actions can be time-consuming processes and are often inefficient when not needed. Furthermore, in the field of preventative maintenance involving ophthalmic medical devices, existing technologies for monitoring such devices may not be able to accurately predict when problems will occur before they do.

[0006] Therefore, there is a need for technologies to accurately predict the likelihood of failure and the timing of potential failures in ophthalmic medical devices or one or more of their components, as well as to perform preventative maintenance or remedial actions in response to such predictions. Summary of the Invention

[0007] Some embodiments provide a method for preventative maintenance of an ophthalmic medical device based on predictive modeling. The method typically includes receiving measurements of one or more operating parameters associated with the ophthalmic medical device. One or more models are used to predict future failures of the ophthalmic medical device. The prediction is generated, at least in part, based on the received measurements of the one or more operating parameters. One or more actions can be taken based on the predicted future failures of the ophthalmic medical device to perform preventative maintenance on the ophthalmic medical device or its components.

[0008] Some embodiments provide a method for training a predictive model to predict failure events on an ophthalmic medical device. The method typically includes generating a training dataset from a set of measurements of operating parameters associated with the ophthalmic medical device. The training dataset typically includes multiple records. Each of the multiple records identifies the measurement result of the operating parameter, the time the operating parameter was measured, and the difference between the time of measurement and the time when a failure event occurs on the ophthalmic medical device. One or more machine learning models are trained based on the training dataset to generate one or more failure predictions about the ophthalmic medical device. The trained one or more machine learning models are deployed to one or more computing systems to predict failures of the ophthalmic medical device and perform remedial or preventative maintenance actions based on the predictions.

[0009] Some embodiments provide a method for preventative maintenance of an ophthalmic medical device based on predictive modeling. The method typically includes generating a training dataset from a set of measurements of operating parameters associated with the ophthalmic medical device. The training dataset may include multiple records, each identifying: a measurement of the operating parameter, the time the operating parameter was measured, and the difference between the time of measurement and the time of a failure event of the ophthalmic medical device. One or more machine learning models are trained based on the training dataset to generate one or more failure predictions for the ophthalmic medical device. Measurements of one or more operating parameters associated with the ophthalmic medical device are received. One or more trained machine learning models are used, at least in part, based on the received measurements of the one or more operating parameters, to predict future failures of the ophthalmic medical device. One or more actions are taken to perform preventative maintenance on the component based on the predicted future failures of the ophthalmic medical device.

[0010] Various aspects of this disclosure provide apparatus, devices, processors, and computer-readable media for performing the methods described herein.

[0011] To accomplish the foregoing and related objectives, the one or more aspects include the features fully described below and specifically specified in the claims. The following description and drawings illustrate certain illustrative features of one or more aspects in detail. However, these features only indicate a few of the various ways in which the principles of each aspect can be employed. Attached Figure Description

[0012] The accompanying drawings depict certain aspects of one or more embodiments and should not be construed as limiting the scope of this disclosure.

[0013] Figures 1A to 1B An example environment is depicted based on certain aspects described herein, in which one or more machine learning models are trained and deployed to predict failures of ophthalmic medical devices and / or to perform preventative maintenance or other remedial actions based on the predictions.

[0014] Figure 2 Example operations based on certain aspects described herein are demonstrated, which can be performed by a computing system within a networked computing environment to perform preventative maintenance of ophthalmic medical devices based on predictive modeling.

[0015] Figure 3 Example operations based on certain aspects described herein are shown, which can be performed by one or more computing systems to train one or more machine learning models to predict the probability of failure of one or more components of an ophthalmic medical device and / or the time at which one or more components may fail.

[0016] Figure 4An example system on which embodiments of this disclosure can be executed is shown.

[0017] For ease of understanding, the same reference numerals are used as much as possible to denote common elements in the figures. It is conceivable that elements and features of one embodiment can be advantageously combined with those of other embodiments without further description. Detailed Implementation

[0018] As discussed above, various ophthalmic medical devices (hereinafter referred to as "OMDs") can be used in ophthalmology to diagnose and treat conditions or diseases of a patient's eye. These OMDs include, for example, surgical systems and consoles, as well as diagnostic and measuring devices and systems. Surgical systems and consoles include systems and consoles for performing various ophthalmic surgical procedures, such as vitreoretinal surgery, cataract surgery, LASIK, and / or any other ophthalmic surgical procedures known to a person skilled in the art. Diagnostic and measuring devices and systems include devices and systems for diagnosing conditions or diseases related to a patient's eye or for measuring various anatomical characteristics of the eye. Examples of diagnostic and measuring devices are refractive diagnostic devices, keratometers, optical coherence tomography (OCT) devices, and / or any other ophthalmic diagnostic and measuring devices and systems known to a person skilled in the art. Each of these OMDs typically includes various components that wear or otherwise deteriorate over time. Therefore, over time, the performance of these OMDs may deteriorate to the point where it is necessary to discontinue use of the OMD to replace worn or damaged components.

[0019] Discontinuing the use of an OMD typically means that a professional cannot use the OMD until a damaged or otherwise deteriorated component is replaced. Furthermore, a considerable amount of time can pass between replacing a damaged or deteriorated component and initially determining the root cause of that damage or deterioration. Therefore, the underlying condition that caused the component's damage or deterioration may persist, leading to similar damage or deterioration of the replaced component, potentially resulting in additional downtime when the replacement component also needs to be replaced.

[0020] In some cases, components of an OMD may malfunction during use (e.g., during surgery). In such situations, it may be beneficial to take proactive action before component failure occurs to prevent OMD malfunction during use. For example, if a component in the OMD is still running but about to fail, the OMD may be stopped or disabled.

[0021] This disclosure provides techniques for using predictive modeling to predict the likelihood of impending failure of one or more components of an OMD, and the timing of such failure, given currently measured operating parameters of the OMD. Predictive modeling allows for the accurate prediction of component failure, the timing of such failure, and / or the cause of such failure, using a large amount of OMD operating data (e.g., and other information that can be used to predict the likelihood, timing, and / or cause of component failure). Furthermore, the predicted likelihood of failure, the timing of such failure, and / or the predicted cause of such failure can be used as triggers for various remedial or preventative maintenance actions on the OMD, which can extend the service life of these components. Preventative maintenance actions, as used herein, refer to various actions that may be performed to address situations related to failure or impending failure of OMD components. Such actions may include, for example, adjusting various configuration parameters used when operating the OMD; generating notifications to users or maintenance personnel identifying components that are failing or may fail; adjusting the operation of the OMD to disable features associated with components that are failing or may fail; disabling the OMD until maintenance is performed, etc. By performing these remedial or preventative maintenance operations, all aspects of this disclosure can extend the service life of OMD components and reduce downtime involved in taking the OMD out of service to replace faulty components.

[0022] Example computing environment for preventative maintenance of ophthalmic medical devices based on predictive modeling.

[0023] Various techniques can be used to train and deploy machine learning models that predict the likelihood, timing, and / or root cause of future failures of one or more components in an OMD (also referred to as failure prediction or future failure prediction in this paper). Figures 1A to 1B Various deployments are shown in the document. For example, Figure 1A The following deployment is demonstrated: a machine learning model is trained and executed on a remote server connected to the monitored OMD in order to predict failures and / or to perform preventative maintenance actions in response to failure predictions. Figure 1B The following deployment is demonstrated: a machine learning model is trained on a remote server and deployed to enable the OMD to perform partial failure prediction and / or preventative maintenance actions in response to failure predictions. However, it should be recognized that various other techniques can be considered for training and deploying machine learning models to predict future failures of components in the OMD, and... Figures 1A to 1B The deployment shown is a non-restrictive illustrative example.

[0024] Figure 1AAn example computing environment 100A is demonstrated, in which an OMD 110 and a server 120 are connected via a network to train one or more ML models for predicting future failures of one or more components in the OMD 110 and / or for preventative maintenance actions on the OMD 110 based on the predictions. As discussed in further detail herein, the ML models can generate predictions at least in part based on operating parameters provided by the OMD regarding various electrical, optical, and / or mechanical components of the OMD. As used herein, operating parameters typically include parameters indicating the operating state of one or more corresponding components of the OMD. For example, operating parameters may include input or output voltage or current for electrical components, light intensity generated by lighting components, mechanical operating information for mechanical devices (e.g., the rotational speed of a motor output shaft or a device powered by a motor, motor temperature), etc.

[0025] OMD 110 typically refers to various devices that ophthalmic surgeons can use to diagnose patients and / or perform ophthalmic surgical procedures on such patients. OMD 110 typically includes one or more components that may wear out or otherwise deteriorate over time. For example, as shown, OMD 110 includes mechanical components 112, electrical components 114, optical components 116, and / or other types of components. Each of these components may be equipped with various sensors or other metrological devices that allow the measurement of various operating parameters of such components. Typically, any number of OMDs may be included in computing environment 100A and generate different sets of operating parameter measurements that can be used as input to one or more ML models predicting future OMD failures. Each OMD 110 in computing environment 100A may generate measurements associated with operating parameters and provide these measurements to server 120 and / or operating history repository 140.

[0026] In some respects, the OMD 110 can additionally generate usage pattern data and send it to the server 120 for analysis. As used herein, usage pattern data typically refers to information defining historical usage patterns over various time windows. For example, usage pattern data may include information about the number of times the OMD is used at varying time granularities (e.g., during the day, during specific parts of the day, etc.). This usage pattern data can indicate operating conditions that cause additional wear on components of the OMD 110, potentially increasing the likelihood of component failure and a corresponding reduction in the lifespan or maintenance intervals associated with those components. For example, each time the OMD is used, components in the OMD may generate heat, and this heat may cause component performance degradation, and the accumulation of heat may lead to further performance degradation. Therefore, usage pattern data can also be used as input to a failure prediction system to determine the likelihood and timing of failures.

[0027] In some respects, the OMD 110 may include various diagnostic and measurement devices for generating anatomical measurements of a patient's eye. Typically, anatomical measurements are measurements of one or more anatomical characteristics of the patient's eye, such as white-to-white distance, anterior chamber depth, axial length, or corneal curvature. These anatomical measurements generally fall within the expected range, which can be defined as values ​​within two standard deviations of the average measurement. Because approximately 95% of measurements are expected to fall within the expected range, statistically, it is unlikely that measurements for more than a certain number of patients will fall outside the expected range. Generally, as discussed in further detail below, a measurement pattern provided by the OMD consistently falling outside the expected range can indicate an existing or future malfunction of one or more components of the corresponding OMD.

[0028] In some respects, the calibration data, usage pattern data, and / or anatomical measurement data discussed above, along with other information, can be used as input to an ML model to predict future failures of components of the OMD 110. When calibration data, usage pattern data, and / or anatomical measurement data are used as input to an ML model, the input vector processed by the ML model can include operating parameter measurements obtained from the OMD 110, as well as calibration data, usage pattern data, and / or anatomical measurement data, as discussed in further detail below. In some respects, calibration data, usage pattern data, and / or anatomical measurement data can be used to determine whether components of the OMD are of interest for further monitoring or analysis using multiple ML models. For example, as further described below, in some respects, calibration data, usage pattern data, and / or anatomical measurement data can trigger an ML model to predict failures of components of the OMD 110.

[0029] To maintain the OMD 110, calibration operations can be performed periodically. The calibration data generated during these operations may include information, for example, defining the adjustments made to the OMD 110 to bring it to a known baseline state. Typically, over time, calibration operations may require larger adjustments to the components of the OMD 110 to compensate for additional degradation in their performance characteristics. Calibration operations can be performed periodically or when the operating parameters of the OMD 110's components deteriorate to a defined threshold or setpoint. When calibration operations are performed periodically, the amount of adjustment required to return the OMD 110 to a known baseline state can be used to determine if the OMD may be failing, as discussed in further detail below. Similarly, when calibration operations are performed in response to the detection that measurements of operating parameters have deteriorated (e.g., to a defined threshold or setpoint), the frequency with which such calibration operations are performed can be used to determine if the OMD may be failing. For example, if calibrations are performed more frequently due to more frequent deterioration of operating parameters, or if calibrations require larger adjustments to compensate for the degradation of the OMD's components, this may indicate that the OMD may be failing.

[0030] Server 120 typically represents a single computing device or cluster of computing devices on which training datasets can be generated and used to train one or more ML models to predict future failures of one or more components in the OMD and / or to perform preventative maintenance actions on the OMD based on these predictions. Server 120 is communicatively coupled to OMD 110 and an operation history repository 130 (hereinafter referred to as "repository 130"), which stores records of historical operating parameters and failure event information (e.g., causes of failures, time information associated with the failures, etc.). In some respects, repository 130 may be or include a database server for receiving information from OMD 110 and / or server 120 and storing the information in corresponding records in a structured and organized manner.

[0031] In some respects, each record in repository 130 may include the following information: the identifier of the component of OMD110 associated with the operating parameters, the operating parameters measured by the sensor or other metering device associated with the component, the time of measurement of the operating parameters, and the time when the component eventually failed.

[0032] Server 120 uses records of these historical operating parameters and time information associated with measurements of these operating parameters and failures of corresponding components in OMD 110 to train an ML model to predict future failures of components in the OMD. More specifically, such as Figure 1AAs shown, server 120 includes training data generator 122 (hereinafter referred to as "TDG 122"), model trainer 124, fault predictor 126, and fault corrector 128. TDG 122 retrieves data from repository 130 to generate data for model trainer 124 to train ML models, which are used by fault predictor 126 to predict future faults of one or more components in OMD.

[0033] Model trainer 124 includes or refers to one or more machine learning algorithms (hereinafter referred to as "ML algorithms") configured to train an ML model using a training dataset. In some embodiments, a trained ML model refers, for example, a function with weights and parameters, used to generate or predict future failures of parts of an OMD for a given set of inputs. Various ML algorithms can be used to generate different types of outputs for a given set of inputs.

[0034] ML algorithms typically include supervised learning algorithms, unsupervised learning algorithms, and / or semi-supervised learning algorithms. Unsupervised learning is a machine learning algorithm used to make inferences from a dataset consisting of input data with unlabeled responses. Supervised learning is a machine learning task that learns a function, such as mapping inputs to outputs based on example input-output pairs. Supervised learning algorithms typically include regression algorithms, classification algorithms, decision trees, neural networks, etc. A description of the dataset is provided below.

[0035] Once trained and deployed, the ML model is able to make predictions related to component failures in the OMD 110 based on a specific set of inputs (including operating parameters of components in the OMD 110 and / or temporal information associated with component failures). Typically, the predictions made by the ML model can include the probability of a component failing within a given time period and / or the time at which a component may fail. In some aspects, the model trainer 124 trains multiple ML models configured to predict failure events for different categories of components in the OMD 110. For example, a first ML model might be configured to predict future failures of a mechanical component based on operating parameters associated with that component in the OMD (e.g., input power noise level, rotational speed, temperature, etc.); a second ML model might be configured to predict failures of electrical components in the OMD based on operating parameters associated with electrical components (e.g., input power noise level, output voltage, output power noise level, temperature, etc.); a third ML model might be configured to predict failures of light-emitting devices based on operating parameters specific to light-emitting components (e.g., input power noise, input power level, output brightness, etc.); and so on. In the example, the probability of failure can be the general probability of failure, such as a 20% probability, a 30% probability, or that a specific component or OMD will fail. In another example, the probability of failure is the probability of failure within a specific time period, such as a 20% probability of failure occurring within the next two days for a specific component or the entire OMD.

[0036] Each sample in the dataset used to train the ML models includes operating parameters specific to the device type targeted by the ML models and / or temporal information relating to the amount of time elapsed between the acquisition of the operating parameters at OMD 110 and the failure of the associated components of the OMD. In some cases, each sample in the dataset may further include calibration data and usage pattern information associated with the measured operating parameters. Usage pattern information may include, for example, the number of times the device was used during the time period between generating a previous set of operating parameters and generating the current set of measured operating parameters. Usage pattern information can be used to further predict future failures based on the assumption that heavier or more intense usage patterns may accelerate or otherwise affect the rate of component degradation in the OMD.

[0037] To train the ML models(s), model trainer 124 runs the ML models on the input data for each sample to generate predictions associated with the input data. In some aspects, the predictions can be probability distributions over a specified time period (e.g., n days), where each day is associated with the probability that a part of the OMD 110 will fail on that day. Typically, since the part of the OMD 110 will eventually fail, the total area of ​​the probability distribution will be close to 1 (i.e., a 100% probability of failure); however, since the device can fail in a predictable manner within a predictable time period, the probability distribution can include multiple low values ​​and one or more spikes, indicating that the part is more likely to fail at some point in the time window. Typically, the difference between the predicted time information (i.e., Y^ predicted for each sample in the dataset) and the actual time information included in each sample in the dataset (i.e., Y included in each sample in the dataset) can be calculated, and these differences can be used to train the ML models(s). For example, the predicted time information could correspond to the predicted number of days until the part fails, and the actual time information included in each sample could be the actual number of days elapsed between recording the input included in each sample and the part failing. It should be noted that the time information discussed above is merely an example, and multiple ML models can be trained to predict Y^ by learning from the actual values ​​of Y that involve other parameters or output types (e.g., parameters or output types that do not involve time).

[0038] In some respects, the model trainer 124 can train (or refine) the ML model based on the error (i.e., YY^) between the actual and predicted time information. In other words, the model trainer 124 adjusts the weights in the ML model to minimize the error (or bias) between the predicted and actual time information. As the model trainer 124 runs more samples through the ML model and continues to adjust the weights, the accuracy of the ML model can improve, allowing it to begin making very accurate predictions with a very low error rate. At this point, the ML model is ready to be deployed to make fault predictions about various components of the OMD 110 (e.g., mechanical components 112, electrical components 114, and / or optical components 116). Figure 1A In the example, the trained ML model can be deployed to the fault predictor 126 to predict the failure of components of the OMD110 based on the captured operating parameters, as described in further detail below.

[0039] In some respects, time-series-based algorithms and datasets can be used to train ML models. For example, Long Short-Term Memory (LSTM) ML algorithms can be used to train ML models, which are capable of learning temporal relationships (or other order dependencies) in the data. Typically, these LSTM models learn a function that maps a sequence of past observations (i.e., past operating parameters of the OMD 110 parts) to output observations (i.e., the probability that the part will fail and the time when the part will fail). Other ML algorithms (such as recurrent neural networks or other algorithms capable of learning temporal relationships between different inputs) can also be used, or alternatively, to predict future failures of OMD 110 parts based on captured operating parameters.

[0040] Model trainer 124 can train a single ML model to roughly generate failure predictions for OMD 110 (e.g., predicting future failures of any component in OMD 110). In this case, the dataset used to train the single ML model may include aggregated operating parameter measurements and / or temporal information related to failures of OMD 110 across multiple monitored components in OMD 110. In another example, model trainer 124 can train multiple ML models to generate failure predictions for different components of OMD 110. The multiple ML models may include a first model trained to generate failure predictions for mechanical component 112, a second model trained to generate failure predictions for electrical component 114, a third model trained to generate failure predictions for optical component 116, and so on. The dataset used to train each of these ML models may include input data that can be used to generate failure predictions for each specific type of component.

[0041] Typically, the model trainer 124 can retrain multiple ML models using the dynamic datasets described herein. The datasets can be described as dynamic because they continuously receive and reflect new data points collected from the OMD 110 and other OMDs. Retraining can be performed periodically (according to a schedule), such as after a threshold number of new entries have been added to the historical datasets used by the model trainer 124 to train and retrain the ML models, or manually. By retraining the ML models using these dynamic datasets, the model trainer 124 can generate ML models that, over time, can make more accurate fault predictions about the OMDs or their components. This improved accuracy in OMD fault prediction, in turn, allows for more timely preventative maintenance actions on one or more components of the OMD 110, and consequently increases system uptime.

[0042] Fault predictor 126 typically uses captured operating parameters, additional information from OMD 110, and a trained ML model to determine whether a component of the OMD is likely to fail. Fault predictor 126 may receive operating parameter measurements and other information from OMD 110 in real time or in periodic batch reports for use in predicting component failures of OMD 110.

[0043] Typically, the fault predictor 126 can use at least the operating parameter measurements of components generated by the OMD 110 and various models (such as prior-defined models applicable to the component or trained ML models(s)) to predict future faults of components in the OMD (e.g., components associated with operating parameter measurements or upstream or downstream components whose faults can be indicated by operating parameter measurements associated with that component). Predictions of future faults of components in the OMD can include, for example, the probability of component failure and / or the likely time of component failure, as well as other predictions. Typically, the prior-defined models do not need to be trained by the model trainer 124 and can be used to determine that a component is failing (e.g., and its timing / probability) based on known characteristics of these components. Trained ML models can be used in conjunction with or in place of prior-defined models to generate fault predictions for components of the OMD 110 (or related components, since measurements of operating parameters of a component can actually indicate that an upstream or downstream component is actually failing).

[0044] As discussed, (multiple) ML models can take a set of operating parameter measurements as input and generate a probability distribution as output. The probability distribution can show the probability of a component failing at each of several time periods following the generation of the operating parameter measurements. For example, within an n-day time window, the probability distribution can show the failure probabilities after 1 day, 2 days, and so on up to n days. Peaks in the probability distribution typically indicate the time when the component is most likely to fail.

[0045] In some respects, the (multiple) ML models can incorporate other information, along with the measurements of this set of operating parameters, as input to predict the likelihood of component failure and / or the timing of component failure. For example, the (multiple) ML models can use calibration data, usage pattern information, and / or measurements of anatomical parameters generated by the OMD 110 (or information derived therefrom), in conjunction with the operating parameter measurements, to predict future failures of the OMD 110 and / or the timing of component failure. Usage pattern information may include, for example, the number of times the OMD 110 is used within a given time period, or more granular information about utilization within a given time period (e.g., utilization during different segments of that given time period). Measurements of anatomical parameters generated within a given time period can be used, for example, to generate information about the proportion of measurements outside the expected range for a given anatomical parameter.

[0046] In some aspects, a priori-defined models can be used to determine whether a component is malfunctioning based on prior known values ​​of normal and abnormal operating parameter measurements of the OMD 110. For example, the priori-defined model can define normal values ​​for operating parameter measurements based on minimum and maximum values. The maximum value can correspond to the value of an operating parameter measurement where a known component (or upstream or downstream component) has malfunctioned. If the received operating parameter measurement of a component in the OMD 110 exceeds this maximum value, the fault predictor 126 can determine that the component has malfunctioned and can instruct the fault corrector 128 to take one or more actions against the component. In some aspects, a priori-defined models can be used initially, and the ML model can be trained over time based on captured historical data, so that the trained ML model can eventually be used in place of or in combination with the priori-defined model.

[0047] In some respects, various thresholding techniques can be used to determine whether the fault predictor 126 should use (multiple) trained ML models to predict future failures of components in the OMD 110. By using these thresholding techniques, the ML models can be used to generate predictions for components of interest in the OMD 110, rather than predictions for all components of the OMD 110 regardless of whether the components exhibit signs of impending failure. Therefore, computational resources can be saved by using ML models to generate predictions only for components of interest used for additional monitoring.

[0048] For example, comparing the received operating parameter measurement results with the expected operating parameter measurement results can be used to determine whether the ML model should be used to generate fault predictions for the components of the OMD 110. Typically, each component of the OMD 110 can be associated with a predefined normal range of operating parameters. For example, mechanical devices can be associated with normal ranges of input current / voltage, rotational speed, vibration measurements, etc. Electrical devices can be associated with normal ranges of output current / voltage, output noise, heat, etc. Light-emitting devices can be associated with normal ranges of requested input power, output brightness levels, etc. If the OMD 110 reports operating parameters for a component within such a normal range, the fault predictor 126 can determine that the component is operating normally, and it is not necessary to use the ML model to predict future faults of the OMD 110 components, because the OMD components are unlikely to fail immediately or require closer attention in the near future.

[0049] In some respects, the fault predictor 126 may use trend analysis or other statistical analysis to determine whether to use an ML model to generate fault predictions for components of the OMD 110. As discussed, the performance of mechanical, electrical, optical, and light-emitting components in the OMD 110 deteriorates over time. In some cases, these components may produce consistent operating parameters for a period of time and then deteriorate abruptly. In other cases, these components may exhibit monotonic patterns of change in operating parameters, which change more drastically as the component reaches the end of its service life or otherwise approaches a failure state. Based on the prior known degradation characteristics of the device's operating parameters, the fault predictor 126 may examine the operating parameters captured for the components within a certain time window to detect trends in the reported operating parameters. If the trend indicates an impending failure (e.g., the trend shows a significant change in operating parameters within the time window), the fault predictor 126 may determine that the components of the OMD 110 should be monitored more closely, and thus a fault prediction for that component may be generated using an ML model trained by the model trainer 124.

[0050] In some respects, the fault predictor 126 may also, or alternatively, use information about the calibration procedures performed on the OMD 110 to determine which ML models(s) should be used to inspect the components of the OMD. Typically, the OMD 110 may be calibrated periodically (e.g., daily) to bring the components to a known basic state. Since the performance of components in the OMD may deteriorate over time, the calibration procedures may gradually require more and more changes to bring the components of the OMD to a known basic state. Furthermore, because the performance characteristics of various components may have known patterns, the calibration procedures used to bring the components of the OMD to a known basic state may have patterns that reflect the performance characteristic patterns of the components. Therefore, the fault predictor 126 may use trend analysis or a known range of calibration parameters to determine whether to use an ML model to generate a fault prediction for the component.

[0051] In another example, fault predictor 126 can use the anatomical measurements generated by OMD 110 to determine which parts of OMD 110 should be examined using multiple ML models. As discussed, anatomical measurements typically fall within a known value distribution. However, if OMD 110 consistently generates anatomical measurements outside the known value distribution, it is more likely that OMD 110 is malfunctioning or may malfunction in the near future. For example, if the number of patients whose generated anatomical measurements fall outside the known distribution exceeds a threshold percentage of patients assessed by OMD, fault predictor 126 can determine that multiple ML models should be used to examine OMD 110. For example, anatomical measurements may include eye-tracking information showing patient eye movements recorded at the time the anatomical measurements were obtained, axial length (i.e., the distance between the anterior cornea and the retina), corneal thickness, anterior chamber depth (i.e., the distance between the anterior cornea and the anterior lens surface), white-to-white diameter (i.e., the distance between the corneal and scleral boundaries on both sides of the eye), lens thickness, lens curvature, and other anatomical measurements that may be recorded by the OMD during use.

[0052] After the fault predictor 126 determines that an OMD component may fail and / or identifies the likely time when the component will fail, the fault predictor 126 may provide the information predicting and identifying the faulty component to the fault corrector 128 for further processing. Typically, the fault corrector 128 is configured to perform various preventative maintenance actions on one or more components of the OMD 110. These preventative maintenance actions may vary based on the component's identity, current operating parameter measurements recorded for the component, and other information that can be used to determine whether remote maintenance of the component is possible.

[0053] In some aspects, the fault predictor 126 may generate one or more notifications to be displayed to users of the OMD and / or support personnel maintaining the OMD, including information about predicted future faults of the OMD. These notifications may include, for example, information about actions taken by the fault corrector 128 in response to a prediction of a component failure, the reasons for taking the action (e.g., in response to a prediction based on operating parameter measurements), and / or other information that may help understand the state of the component in the OMD, problems with the component in the OMD, and future actions to be taken in response to a prediction of a component failure.

[0054] Fault predictor 126 can be used to perform fault prediction for OMDs located in various operating environments and geographical regions. Because fault predictor 126 can generate fault predictions for devices in changing operating environments and geographical regions, model trainer 124 and fault predictor 126 can utilize extensive data to train and predict a wide range of faults across deployed OMDs. Furthermore, data points obtained from widely deployed OMDs can allow for various analyses based on large datasets. For example, fault predictor 126 can divide widely deployed OMDs into multiple groups (e.g., region groups, environment groups, etc.) to identify faults that are more or less prevalent in these groups. Information about faults more prevalent in a group can be provided by fault predictor 126 to the engineering team responsible for the specific component of the OMD to identify the root causes(s) of these faults and new designs or procedures for these components that may reduce the prevalence of faults in the specific user group.

[0055] In some aspects, the fault corrector 128 may include a library of programmed solutions that can be executed to remedy faults or anticipated faults in components of the OMD 110. These programmed solutions may be associated with, for example, measurements of operating parameters for specific types of components and specific groups. Typically, the programmed solutions may include executable code that remedies abnormal performance at the OMD 110 when pushed to the OMD 110 for execution. These programmed solutions may include, for example, software patches controlling the timing of various operations, changes to calibration targets for adjusting the calibration of the OMD 110, and so on. When a match is found between a programmed solution and the component and operating parameter measurements reported by the OMD 110, the fault corrector 128 may push the programmed solution to the OMD 110 for execution.

[0056] In some respects, fault corrector 128 can determine that components identified as faulty or likely to fail immediately should be replaced to remedy the fault or anticipated failure. Fault corrector 128 can determine the component to be replaced based on information indicating that the component has failed due to an electrical or mechanical fault that cannot be compensated for by software. For example, if a gradually increasing amount of power is required to generate the expected output of a component, fault corrector 128 can determine that an electrical component should be replaced because the increased power demand (e.g., a corresponding increase in resistance in the electrical component) may indicate an impending failure that cannot be corrected by software. In another example, if reported operating parameter measurements for a mechanical component indicate performance degradation (e.g., inconsistent rotational speed of the motor) or indicate that the device has a heartbeat (e.g., electrical connection to the controller at OMD 110) but is mechanically inoperable, fault corrector 128 can determine that the mechanical component should be replaced. In this case, fault corrector 128 can push one or more instructions to OMD 110 to disable the OMD until the identified component that has failed or is about to fail has been replaced.

[0057] In some respects, the fault corrector 128 may perform diagnostic tests or other actions on components on the OMD 110 to further determine whether the component is currently failing or is about to fail. Diagnostic tests may include instructions to cause the OMD to perform specific operations and provide the fault corrector 128 with additional operating parameter measurements for analysis. If the additional measurements indicate that the component is failing or may fail, the fault corrector 128 may determine that preventative maintenance actions, as discussed above, should be performed on the component.

[0058] Figure 1B Another example computing environment 100B is shown, in which training and using machine learning models can be performed to predict future failures of components in the OMD for preventative maintenance. As shown, computing environment 100B includes a measurement device 110, a server 120, and an operation history repository 130. Figure 1BIn the example shown, TDG 122 and model trainer 124 are executed on server 120, while fault predictor 126 and fault corrector 128 are executed on OMD 110 using data generated by device component instruments associated with mechanical component 112, electrical component 114, and / or optical component 116. Device component instruments typically include various sensors and other metrological devices capable of monitoring and measuring various operating parameters of components on the OMD. As discussed, these operating parameter measurements can be used alone or in combination with other data, such as calibration data, usage history, or measurements of various patient anatomical parameters, to predict potential component failures on the OMD and, based on this prediction, to identify preventative maintenance actions for the OMD to prevent component failures.

[0059] Example methods for preventative maintenance of ophthalmic medical devices based on predictive modeling

[0060] Figure 2 Example operations 200 are demonstrated. These example operations can be executed by a computing system to predict future failures of components in the OMD, thereby enabling preventative maintenance of the OMD. Operation 200 can be performed by... Figures 1A to 1B It can be executed by one or more of the OMD 110 or server 120 shown in the figure.

[0061] As shown, operation 200 can begin at box 210, where the system receives measurement results of one or more operating parameters associated with components of the OMD. The measurement results may include operating parameter measurements at different times over a period of time. In some cases, measurement results may be received periodically (e.g., according to a heartbeat monitor performed on the OMD or a server receiving the measurement results). For example, measurement results may be received daily after each use of the OMD, or according to some other periodicity defined for receiving measurement results from the OMD.

[0062] At box 220, the system uses one or more models, at least in part, based on measurements of one or more received operating parameters, to predict the probability of future component failure and / or the timing of potential component failure. This model may include a priori-defined models that identify whether a component is failing based on known values ​​of operating parameter measurements corresponding to normal and abnormal operation, or based on a trained ML model capable of predicting future failures of the OMD. Note that using a priori models to identify whether a component is failing may include predicting the probability of component failure and / or the timing of potential component failure, at least in part, based on measurements of one or more received operating parameters. In some cases, to reduce the computational resource utilization involved in predicting future component failures of the OMD, the system may predict future component failures of the OMD in response to determining that additional data from the OMD indicates that the component has deteriorated to the point requiring predictive modeling and preventative maintenance.

[0063] For example, to determine if predictive modeling is necessary, the system can check the measured operating parameters of the component against known specifications. Known specifications may include, for example, information about the expected range of the measured operating parameter values ​​for the component. If the measurements exceed the expected range (e.g., exceed a threshold amount), the system can determine that predictive modeling and preventative maintenance based on predictive modeling are necessary. In another example, the system can use trend analysis or statistical analysis of the measured optical parameters and historical measurement results to determine if the component exhibits signs of performance degradation indicating an impending failure.

[0064] In some cases, the system can examine data beyond the measured optical parameters or historical measurements of those parameters to determine if predictive modeling and preventative maintenance are necessary. For example, the results of calibration tests can be examined to determine if increasingly larger calibration adjustments are being used to calibrate the OMD. Since using larger calibration adjustments can indicate component degradation in the OMD, increasing calibration adjustments beyond a threshold amount can indicate that a component is failing or is likely to fail, thus necessitating predictive modeling and preventative maintenance. In another example, the system can examine patient anatomical measurements to determine if the OMD consistently generates abnormally distributed measurements relative to the values ​​of anatomical parameters. If the OMD continues to generate abnormal measurements, the system can determine that the OMD is performing abnormally and that predictive modeling and preventative maintenance are necessary.

[0065] At box 230, the system takes one or more actions to perform preventative maintenance on the component based on the predicted probability of failure and / or the time when the component may fail. One or more actions may include executing code on the OMD to address potential failure modes of the OMD, performing additional diagnostic tests on the OMD to determine further action procedures, or disabling the OMD until the faulty component is replaced.

[0066] Figure 3 Example operations 300 are demonstrated. These example operations can be executed by the system to train a machine learning model to predict future failures of components in the OMD, for preventative maintenance of the OMD. Operation 300 can be performed by, for example... Figures 1A to 1B The server 120 shown in the image is used to execute the command.

[0067] As shown, operation 300 begins at box 310, where the system generates a training dataset from measurements of a set of operating parameters associated with the OMD. In some aspects, the dataset may include measured operating parameters for various parts of the OMD and / or time information associated with the time the measurements were obtained at the OMD and the time the OMD failed. Generating the training dataset may require generating different training datasets for different parts or part categories of the OMD, as different parts or part categories may have different parameters indicating the likelihood of future failures.

[0068] At box 320, the system trains one or more machine learning models based on a training dataset. One or more machine learning models can be trained to generate fault predictions for the OMD. For example, a first ML model can be trained to generate fault predictions for the mechanical parts of the OMD. A second ML model can be trained to generate fault predictions for the electrical parts of the OMD. A third ML model can be trained to generate fault predictions for the light-emitting parts of the OMD. Further ML models can also be trained to generate fault predictions for other categories or types of parts in the OMD. In another example, ML models can be trained to generate fault predictions for each specific part in the OMD.

[0069] At box 330, the system deploys one or more trained ML models to one or more computing systems. In some aspects, the trained ML models can be deployed to a fault predictor that runs on the same system (or cluster of systems) used to train the ML models. Alternatively, the trained ML models can be deployed to a computing system local to the OMD, which can allow for fault prediction and preventative maintenance of OMDs that may not be connected to the central system or may have intermittent connections to the central system.

[0070] Example system for preventative maintenance of ophthalmic medical devices based on predictive modeling

[0071] Figure 4 An example system 400 is shown that uses a machine learning model to assist in performing ophthalmic surgical procedures (such as cataract surgery). For example, system 400 may correspond to one or more of the measuring device 110, server 120, and / or user console 130 shown in Figure 1.

[0072] As shown, system 400 includes a central processing unit (CPU) 402, one or more I / O device interfaces 404 that allow various I / O devices 414 (e.g., keyboard, display, mouse device, pen input, etc.) to be connected to system 400, a network interface 406 through which system 400 is connected to a network 490 (which may be a local network, intranet, Internet or any other group of computing devices communicatively connected to each other), memory 408, storage device 410, and interconnect 412.

[0073] CPU 402 can retrieve and execute programming instructions stored in memory 408. Similarly, CPU 402 can retrieve and store application data residing in memory 408. Interconnect 412 transfers programming instructions and application data between CPU 402, I / O device interface 404, network interface 406, memory 408, and storage device 410.

[0074] CPU 402 is included to indicate a single CPU, multiple CPUs, a single CPU with multiple processing cores, etc.

[0075] Memory 408 represents volatile memory such as random access memory and / or non-volatile memory such as non-volatile random access memory, phase change random access memory, etc. As shown, memory 408 includes device component instrumentation software 420, TDG 430, model trainer 440, fault predictor 450, fault corrector 460, and operation history repository 470. Device component instrumentation software 420 typically interfaces with sensors and other metrology devices in the OMD to obtain measurements of various operating parameters of the OMD's components (including but not limited to mechanical, electrical, and / or optical components). The operating parameter measurements obtained by device component instrumentation software 420 typically include various parameters that can predict whether a component of the OMD is malfunctioning or may malfunction, and if so, when such a malfunction may occur.

[0076] TDG 430 typically uses information obtained from the instrumentation software 420 of the device components to generate a training dataset for model trainer 440 to train one or more ML models to predict future failures of components in the OMD for preventative maintenance of the OMD. To generate the training dataset, TDG 430 can generate records that correlate various operating parameter measurements of the OMD components with time information related to component failures. These records in the training dataset can be persisted to the operation history repository 470 in storage device 410 for storage and future use in generating updated datasets to train ML models to predict future failures of components in the OMD for preventative maintenance of the OMD.

[0077] Model trainer 440 typically trains one or more ML models used by fault predictor 450 to predict future failures of components in OMD for preventative maintenance of OMD. As discussed, model trainer 440 can use a training dataset generated by TDG 430 to train the ML model, and the trained ML model can be deployed to fault predictor 450 (or a remote system) for use.

[0078] Fault predictor 450 typically uses operating parameter measurements obtained via device component instrumentation software 420 to determine whether a component of the OMD is currently failing or is likely to fail in the near future. If a component is failing or is likely to fail (e.g., based on operating parameter measurements, calibration data, or anatomical measurements being outside the expected range), fault predictor 450 can use a trained ML model to determine when the OMD is likely to fail. Based on the prediction, fault corrector 460 can be used to perform further diagnostic tests on the component, push preventative maintenance codes or remedial instructions to the OMD for execution, and / or disable the OMD until the failing or potentially failing component is replaced.

[0079] Additional considerations

[0080] The foregoing description is provided to enable any person skilled in the art to practice the various embodiments described herein. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments. For example, changes can be made to the function and arrangement of the elements discussed without departing from the scope of this disclosure. Various examples may be appropriately omitted, substituted, or added to various procedures or components. Furthermore, features described with respect to some examples may be combined in some other examples. For example, any number of aspects set forth herein can be used to implement an apparatus or practice. Additionally, the scope of this disclosure is intended to cover apparatuses or methods practiced using structures, functions, or structures and functions other than or different from the aspects set forth herein. It should be understood that any aspect of this disclosure may be embodied by one or more elements of the claims.

[0081] As used herein, the phrase “at least one” in a list of items refers to any combination of those items, including a single member. For example, “at least one of a, b, or c” is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination of multiples of the same element (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other order of a, b, and c).

[0082] As used herein, the term "determine" encompasses a wide variety of actions. For example, "determine" can include calculation, operation, processing, derivation, investigation, searching (e.g., searching in a table, database, or other data structure), ascertainment, etc. Furthermore, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Moreover, "determine" can include parsing, selecting, picking, building, etc.

[0083] The methods disclosed herein include one or more steps or actions for implementing the methods. The method steps and / or actions may be interchanged without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims. Furthermore, the various operations of the above methods can be performed by any suitable means capable of performing the corresponding functions. Such means may include various hardware and / or software components and / or modules, including but not limited to circuits, application-specific integrated circuits (ASICs), or processors. Typically, where operations are illustrated in the figures, those operations may have corresponding means and functional components with similar numbering.

[0084] The various illustrative logic blocks, modules, and circuits described in connection with this disclosure may be implemented or executed using a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but alternatively, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors incorporating a DSP core, or any other such configuration.

[0085] The processing system can be implemented using a bus architecture. Depending on the specific application and overall design constraints of the processing system, the bus can include any number of interconnect buses and bridges. The bus can link together various circuits, including processors, machine-readable media, and input / output devices. User interfaces (e.g., keypads, displays, mice, joysticks, etc.) can also be connected to the bus. The bus can also link various other circuits, such as timing sources, peripherals, voltage regulators, power management circuits, etc., which are well known in the art and therefore will not be described further. The processor can be implemented using one or more general-purpose and / or special-purpose processors. Examples include microprocessors, microcontrollers, DSP processors, and other circuit systems capable of executing software. Those skilled in the art will recognize how best to implement the functions of the processing system according to the specific application and the overall design constraints imposed on the system as a whole.

[0086] If implemented in software, functionality can be stored or transmitted as one or more instructions or code on or through a computer-readable medium. Software should be interpreted broadly as instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or others. Computer-readable media includes both computer storage media and communication media (such as any medium that facilitates the transfer of computer programs from one place to another). The processor may be responsible for managing the bus and general processing, including the execution of software modules stored on the computer-readable storage medium. The computer-readable storage medium may be coupled to the processor, allowing the processor to read information from and write information to the storage medium. Alternatively, the storage medium may be integrated with the processor. For example, a computer-readable medium may include a transmission line, a carrier wave modulated by data, and / or a computer-readable storage medium on which instructions separate from the wireless node are stored, all accessible to the processor via a bus interface. Alternatively or additionally, the computer-readable medium or any portion thereof may be integrated into the processor, as in cases where it may have a cache and / or a general-purpose register file. Examples of machine-readable storage media may include RAM (random access memory), flash memory, ROM (read-only memory), PROM (programmable read-only memory), EPROM (erasable programmable read-only memory), EEPROM (electrically erasable programmable read-only memory), registers, disks, optical disks, hard disks, or any other suitable storage media, or any combination thereof. Machine-readable media may be embodied in computer program products.

[0087] Software modules may include single or multiple instructions and may be distributed across several different code segments, across different programs, and across multiple storage media. Computer-readable media may include multiple software modules. Software modules include instructions that, when executed by a device such as a processor, cause the processing system to perform various functions. Software modules may include transmission modules and reception modules. Each software module may reside in a single storage device or be distributed across multiple storage devices. For example, when a triggering event occurs, a software module may be loaded from a hard disk drive into RAM. During the execution of a software module, the processor may load some instructions into a cache to improve access speed. Then, one or more cache lines may be loaded into a general-purpose register file for processor execution. When referring to the functionality of a software module, it should be understood that this functionality is implemented by the processor when executing instructions from that software module.

[0088] The following claims are not intended to be limited to the embodiments shown herein, but are given the full scope consistent with the language of the claims. In the claims, unless specifically stated otherwise, references to singular elements are not intended to mean “one and only one”, but rather “one or more”. Unless otherwise specifically stated otherwise, the term “some” means one or more. Pursuant to 35 U.S.SC §112(f), no element of any claim will be interpreted unless the element is expressly described using the phrase “means for…” or, in the case of a method claim, using the phrase “steps for…”. All structural and functional equivalents of elements in all aspects described throughout this disclosure that are known to or will be known hereafter by one of ordinary skill in the art are expressly incorporated herein by reference and are intended to be covered by the claims. Furthermore, regardless of whether such disclosure is expressly described in the claims, the contents of this disclosure are not intended for public use.

Claims

1. A method for preventative maintenance of ophthalmic medical devices based on predictive modeling, the method comprising: Receive measurement results of one or more operating parameters associated with the ophthalmic medical device; Receive patient anatomical measurement results generated by the ophthalmic medical device for multiple patients; as well as Based on received patient anatomical measurements, it was determined that the performance of the ophthalmic medical device or its components had deteriorated. Prediction is made using one or more machine learning models, which are trained to predict future failures of the ophthalmic medical device based at least in part on measurements of the one or more operating parameters received and further on the determination that the performance of the ophthalmic medical device or the component has deteriorated, wherein determining that the performance of the ophthalmic medical device or the component has deteriorated includes: For each received patient anatomical measurement, determine whether the measurement is outside the threshold range of the data point represented by the measurement; Determining that at least a threshold number of the received patient anatomical measurements are outside the typical range; and Based on predicted future failures of the ophthalmic medical device, one or more actions are taken to preventively maintain the ophthalmic medical device, wherein the predicted future failures include at least one of the probability of future failure of the ophthalmic medical device or the time when the ophthalmic medical device may fail.

2. The method as described in claim 1, wherein, The one or more models include models that define values ​​for the one or more operating parameters corresponding to normal operation of the ophthalmic medical device and values ​​for the one or more operating parameters corresponding to malfunctions of the ophthalmic medical device.

3. The method as described in claim 1, wherein, The one or more operating parameters are associated with components of the ophthalmic medical device, and wherein predicting future failures of the ophthalmic medical device includes predicting future failures of the components or one or more other components in the ophthalmic medical device.

4. The method of claim 1, further comprising: The calibration data of the ophthalmic medical device or its components are received, wherein the prediction of future failures of the ophthalmic medical device is further based on the trend exhibited by the calibration data over time.

5. The method of claim 1, further comprising: The ophthalmic medical device is received using usage pattern data, wherein predicting future malfunctions of the ophthalmic medical device is further based on the usage pattern data.

6. The method of claim 5, wherein, The usage pattern data includes information related to system utilization over multiple time windows.

7. The method of claim 1, wherein, The patient anatomical measurements include eye-tracking information, which includes patient eye movements recorded during the collection of the patient anatomical measurements, and wherein determining that the performance of the ophthalmic medical device or the component has deteriorated includes determining that the recorded patient movements of the plurality of patients have exceeded a threshold amount.

8. The method of claim 1, further comprising: Based on a comparison between measured operating parameters of a component of the ophthalmic medical device and a range of values ​​defined as indicative of normal operation of the component, it is determined that the performance of the component has deteriorated, wherein predicting future failures of the ophthalmic medical device is further based on the determination that the performance of the component has deteriorated.

9. The method of claim 1, wherein, The one or more actions include: Identify programmed solutions for remedies to malfunctions or anticipated malfunctions of the ophthalmic medical device when performed on it; and The identified programmatic solution is pushed to the ophthalmic medical device for execution.

10. The method of claim 1, wherein, The one or more actions include: Identify one or more components of the ophthalmic medical device that require replacement to remedy a malfunction or anticipated malfunction of the ophthalmic medical device; and The ophthalmic medical device will be disabled until the identified component to be replaced is replaced.

11. The method of claim 1, wherein, The one or more operating parameters are associated with components of the ophthalmic medical device, and wherein the one or more actions include: Identify one or more actions to be performed using the components of the ophthalmic medical device; Transmit one or more instructions to the ophthalmic medical device to perform one or more identified actions; Receive additional measurement results of operating parameters from the ophthalmic medical device in response to the transmission of the one or more instructions; and Based on the received additional measurement results, one or more additional actions are determined to be performed to preventatively maintain the component based on the predicted future failures.

12. The method of claim 1, wherein, The one or more operating parameters are associated with components of the ophthalmic medical device, and the method further includes: An aggregated dataset is generated by aggregating measurements of one or more operating parameters associated with components of the ophthalmic medical device and the predicted future failures of the ophthalmic medical device with measurements and predictions from multiple other ophthalmic medical devices; and One or more analyses are generated based on the aggregated dataset, which show trends across multiple groups of ophthalmic medical devices.

13. The method of claim 1, further comprising: Output a notification for display to the user of the ophthalmic medical device. The notification includes information identifying one or more components of the ophthalmic medical device that may fail, the predicted future failure, and information related to the one or more actions to be taken for preventative maintenance.

Images