The present disclosure relates to monitoring electrical current consumption and temperature characteristics of cold storage devices to learn their operating behavior and to signal an alarm indication when abnormal behavior is detected.
Cold storage devices such as refrigerators and freezers play a mission-critical role in settings such as medical research labs and tissue banks, where an overnight malfunction can put many years of research and investment dollars to waste. In settings such as these, there is a clear need for automated remote health monitoring of these cold storage devices to provide stakeholders with an early warning indication if any abnormalities are detected.
In one form, the present disclosure describes a cold storage monitoring system that monitors the temperature inside and electrical current supplied to a cold storage unit such as a refrigerator or freezer, learns how the system behaves in a normal or baseline state, and signals an alarm indication when abnormal behavior is detected.
More specifically, a process is provided that includes monitoring one or more of the electrical current consumption of and temperature inside a cold storage device. Several embodiments of a monitoring device are described herein to obtain signals/data representing the electrical current consumption of and temperature inside a cold storage device. The process includes identifying operational state changes of the cold storage device using detected changes in the electrical current consumption, and calculating a feature vector of electrical and thermal properties of the cold storage device between two consecutive operational state changes. Furthermore, the process includes a learning process that includes: accumulating the feature vectors over a period of time; identifying clusters of accumulated feature vectors; associating one or more functional operating states of the cold storage device with one or more of the clusters; calculating learning statistics based on one or more of: a frequency that the cold storage device enters the one or more functional operating states or a variation of a feature vector parameter within one or more of the clusters; and generating an alarm threshold from the learning statistics; The process also includes a monitoring process that includes: determining a nearest cluster to the feature vector; determining one or more current functional operating states of the cold storage device from the functional operating states associated with the nearest cluster; calculating a monitoring statistic based on one or more of: the one or more current functional operating states or one or more feature vector components/elements; and sending an alarm notification if the monitoring statistic exceeds the alarm threshold.
The foregoing and other features of the present disclosure will become apparent to those skilled in the art to which the present disclosure relates upon reading the following description with reference to the accompanying drawings.
The present disclosure relates generally to a monitoring system for a cold storage device such as a vapor compression refrigerator or freezer. The monitoring system learns operating characteristics of the cold storage device and issues alarm notifications when abnormal behavior is detected. Such a system can be used as an “early warning system” to flag when a cold storage device is not operating properly. Such a system could be particularly valuable in applications that make mission-critical use of cold storage devices, e.g., biomedical or pharmaceutical research labs, blood or tissue banks, grocery stores, restaurants, and the like.
Referring to
The server 20 includes a communication interface (e.g., network interface card(s)) 20A, memory 20B and one or more processors 20C. The memory 20B may take the form of any non-transitory computer readable storage media, such as random access memory, read only memory, etc. The memory 20B may store or be encoded with instructions that, when executed by the one or more processors 20C, cause the one or more processors 20C to perform the server operations described herein.
Referring to
The monitor 12 may include a processor or central processing unit (CPU) 66 configured to execute instructions to, among other things, determine the operating state of the cold storage device using the measurement data obtained from the current sensor 54 and temperature sensor 72, characterize the behavior of these sensor signals over time, and send alarm notifications to the server 20 via the Wi-Fi transceiver 68 if abnormal behavior is detected.
The a BLE transceiver 70 may be used to pass measurement data and/or configuration data between the monitor's processor 66 and an external smartphone, tablet computing device or other portable/mobile device. Another potential use for the BLE transceiver 70 is to read measurement data from one or more external wireless sensors such as temperature, humidity, or air pressure sensors that support the BLE protocol.
Reference is now made to
Turning now to
The monitor 12 uses an operating state detection algorithm to determine the current operating state or states of the cold storage device 14, to learn how the operating state varies over time its behavior over time, and to generate an alarm indication if abnormal behavior is detected. A block diagram showing the signal processing flow of the algorithm is shown in
Referring now to
Turning back to
The clustering module 582 feeds key characteristics about each cluster into a “functional state association” module 586 which associates one or more functional operating states of the cold storage device with one or more of the clusters. Examples of functional operating states include: whether a cold storage device's compressor is on, whether its defroster is on, whether a door is opened and a door light is on, whether a refrigerator-to-freezer cooling fan is on and damper is opened, and the like. A feature vector is made of feature vector components, also called feature vector elements. The key characteristics could include the centroid of each cluster; the min and max values for each feature vector component over all feature vectors contained within the cluster; the 90 percent min and max values for each feature vector component, which could be obtained by sorting each feature vector component's values over all feature vectors contained within the cluster, then taking the value that's within 10% of the sorted min or max; or the median of each cluster, which could be obtained by computing the median of each feature vector component value over all feature vectors contained within the cluster. In some cases the key characteristics could include all the feature vectors contained within each cluster.
The clustering and functional state association modules 582 and 586 also take statistics on the accumulated feature vectors and their associated functional operating states and generate one or more alarm thresholds based on this information which are passed on to the “Alert Gen” module 598 to send alarm notifications as part of the monitoring process. In certain cases, the statistics taken on the accumulated feature vectors are often only computed over a subset of one or more clusters. For example, mean compressor current overshoot would only be computed for feature vectors occupying clusters associated with a compressor off to compressor on functional state transition. Example statistics on the associated functional operating states include compressor duty cycle, compressor period, defroster duty cycle and defroster period, compressor steady-state current or on duration, compressor off duration, defroster on current, and defroster on duration.
The algorithm 560 uses a “Monitoring” process depicted at reference numeral 588 to determine the operating state of the cold storage device from the most recently received feature vector, to calculate statistics on the operating state over a period of time, and to signal an alarm indication if a malfunction is detected.
The first part of the Monitoring process 588 is to map a feature vector to its nearest cluster. This mapping process is done in the “Find Nearest Cluster” module 590. The mapping could be done by finding the cluster centroid having the minimum distance to the feature vector, wherein the distance is measured using a Euclidean norm or some other appropriate distance metric. Alternatively, the mapping could be done using decision regions generated from the clustering module 582 during learning. After the nearest cluster is determined, the current functional operating state 594 of the cold storage device is calculated in the “Functional State Mapping” module 592 using the nearest cluster and its associated functional operating state, the mapping for which was computed by the “Functional State Association” module 586.
The current functional operating state 594 is fed into a “Stats” module 596 that computes time-based statistics on the behavior of the functional operating state 594 or one or more of the feature vector 580 components. Examples statistics on the functional operating state include compressor duty cycle, mean compressor on duration, max compressor off duration, defroster duty cycle, and the like. Example statistics on the feature vector components include max/min/mean current overshoot level when compressor on, mean defroster on current, and mean defroster on duration.
The “Alert Gen” module 598 generates alert or alarm indications if it is determined that any of the statistics calculated in the Stats module 596 exceed a threshold. The alarm thresholds used in this calculation are supplied from the “Functional State Association” 586 and “Clustering” 582 modules. The alarm indications could be sent to the server 20 via the Wi-Fi transceiver 68 and then sent from the server to a user via his or her smartphone 30 or other mobile device (
The Learning and Monitoring processes 578 and 588, respectively, may or may not be used to process all of the feature vectors 580. In some cases, the Monitoring process could be disabled entirely until the Learning process 578 has had a chance to learn the cold storage device's characteristics over some minimum time period of, say, several days. After that minimum time period, Monitoring could begin. At that point, Learning could either stop altogether, or could continue for some other minimum duration or perhaps forever.
The “Alert Gen” module 598 could generate an “unrecognized operating state” alarm condition when the distance between a feature vector and its nearest cluster is excessively large or when the feature vector falls outside of some appropriate detection region. This alarm could be used as a “catchall” alarm in case an abnormality wasn't detected via some other means. The “Alert Gen” module 598 may also generate a “device disconnected from AC power” alarm when the RMS current is abnormally low for some period of time.
The algorithm 560 may detect an electrical current surge by looking for feature vectors indicating large spikes in the transient current level—levels that are significantly higher than the cold storage device exhibits during normal operation. When this condition is detected, the monitor could notify the user via the server 20 that a large surge in current was detected that could have damaged the cold storage device.
In some cases when an alarm threshold is breached and an alarm notification is sent, the algorithm 560 cannot determine with certainty whether the cold storage device has indeed malfunctioned or if the detected behavior is in fact normal and acceptable. In such cases, the alarm notification sent to the user may give the user a way to provide feedback on which is the case. For example, if an abnormal compressor duty cycle is detected, the alarm message sent to the user via a text message could read “An abnormally high compressor duty cycle on freezer XYZ was detected. Please respond with an “OK” if the freezer seems to be behaving normally.” When the user feedback indicates that the cold storage device is behaving normally, the monitor could pass any feature vectors that have accumulated since the alarm was triggered through the Learning process 578 to ensure that the new behavior is included in the learning statistics and that the alarm doesn't retrigger again in the future.
The algorithm 560 could also give the user the ability to clear its learning state and re-start the Learning process 578 via a user interface. This could be useful if the system was just repaired or serviced and is now known to be in a healthy state.
Reference is now made to
As described herein, the functional operating states may include one or more of: compressor or defroster of the cold storage device is on; one or more dampers of the cold storage device are open; one or more fans of the cold storage device are on; or one or more door lights of the cold storage device are on.
The learning statistics may include one or more of: compressor duty cycle, compressor on duration, compressor off duration, compressor period, defroster duty cycle, defroster on duration, defroster off duration, defroster period, compressor and defroster off current, rate-of-cooling when compressor on, rate-of-heating when compressor off, max temp when defroster on, rate-of-heating when defroster on.
The alarm notifications may include one or more of: compressor powered on for an unusually large time period, compressor powered off for unusually large time period, short-term average compressor duty cycle uncharacteristically high or low, long-term average compressor duty cycle uncharacteristically high or low, uncharacteristically low rate of cooling when compressor powered on, abnormal rate of heating when defroster powered on, abnormal rate of heating when compressor powered off, unexpected defroster “on” duration, missing defrost cycle, unexpected defroster “off” duration, irregular compressor power-up transient behavior, irregular compressor current consumption while powered on, or unexpected defroster current consumption.
The feature vector may include a component for the temperature inside the cold storage device, wherein the alarm thresholds include thresholds to indicate a temperature out-of-range condition inside the cold storage device, wherein the functional operating states include a defroster of the cold storage device is on, and wherein different values for the temperature alarm thresholds are used when the defroster has recently been determined to be on versus otherwise.
The methods presented herein may determine whether there is a defroster in the cold storage device. For example, referring back to
In one form, the feature vector may include a component for the temperature inside the cold storage device, wherein the alarm thresholds include thresholds to indicate a temperature-too-high condition inside the cold storage device and the length of time that the temperature has been too high, wherein the functional operating states include whether the compressor is on, wherein sending includes sending an alarm notification a period of time after a temperature-too-high condition has been detected and the cold storage device's compressor is determined to be powered on, and sending an alarm notification immediately and without delay if the compressor is determined to not be powered on when the temperature-out-of-range condition is first detected.
In still another form, the feature vector may include a component for the temperature inside the cold storage device, wherein the alarm thresholds include thresholds to indicate a temperature out-of-range condition inside the cold storage device, wherein the functional operating states include whether the defroster is on, and wherein different values for the temperature alarm thresholds are used when the defroster has recently been determined to be on versus otherwise.
In yet another form, the feature vector includes components for one or more of: the ambient temperature and humidity outside of the cold storage device, wherein the functional operating states include an indication of whether the compressor is on, wherein the learning statistics include the compressor duty cycle, and further comprising adjusting the functional operating state alarm thresholds as a function of one or more of the ambient temperature and humidity.
The methods presented herein may further include reading, with a radio frequency identifier (RFID) interrogator, RFID tags associated with items stored in the cold storage unit in order to determine a type of material being stored inside the cold storage device, and adjusting one or more of the alarm thresholds based on the type of material determined to be stored inside the cold storage device.
The monitoring, identifying and calculating operations may be performed on a plurality of cold storage devices. In this case, the accumulating in the learning process further includes accumulating the feature vectors over time from the plurality of cold storage devices, and wherein the calculating in the monitoring process is performed on a single cold storage device that may or may not be one of the plurality of cold storage devices.
The monitoring may further include monitoring one or more of the humidity and temperature both inside and outside the cold storage device, wherein the functional operating states include an indication of whether the compressor is on, wherein the learning and monitoring statistics include a compressor duty cycle, wherein the learning and monitoring statistics also include statistics on how the compressor duty cycle varies as a function of the one or more of the humidity and temperature both inside and outside the cold storage device, and wherein the alarm notification is used to indicate that the monitored compressor duty cycle is outside of a normal range at the current settings for the one or more of the humidity and temperature both inside and outside the cold storage device.
Further still, the feature vector may include components for one or more of: transient current overshoot level; transient current overshoot duration; post-overshoot minimum, maximum or average current level; minimum, maximum or average temperature; minimum, maximum or average temperature rate-of-change. In this case, the methods may further include determining whether an electrical surge has occurred using the transient current overshoot level and the duration and wherein sending an alert notification in the monitoring process is used indicate that an electrical surge has occurred.
In one form, the monitoring process further includes receiving from one or more recipients of the alarm notification, feedback as to whether the alarm notification is indicative of a malfunction of the cold storage device; and if the feedback indicates that the alarm notification is not indicative of a malfunction, updating the learning process using the feature vector or feature vectors that triggered the alarm notification such that the parameter that triggered the alarm notification is not deemed to be associated with a malfunction of the cold storage device.
In one form, the learning process and monitoring process may both be executed for each calculated feature vector. In another form, only one but not both of the learning process and monitoring process are executed for a subset of the calculated feature vectors.
The monitoring process may further include sending an unrecognized operating state alarm indication if the distance to the nearest cluster or cluster centroid exceeds an alarm threshold.
One well-known tradeoff with temperature monitoring systems for cold storage devices lies between the time it takes to detect a problem and false alarm probability. For example, if a monitoring system is configured to generate a temperature-out-of-range alarm if the temperature is out of range for, say, one minute, the chances for generating a false alarm could be relatively high. A false alarm could be easily generated if someone leaves the refrigerator door open for a minute or two to re-stock the refrigerator. The false alarm probability could be lowered significantly by increasing the temperature-out-of-range detection period from, say, 1 to 60 minutes. But increasing the minimum detection time in this way would also have the undesirable effect of increasing the time required to detect any legitimate issue with the refrigerator—for example, if someone leaves the refrigerator door permanently open or if the AC power becomes unplugged.
A cold storage monitor can be configured to run the operating state detection algorithms described herein to achieve improved performance relative to the detection latency vs. false alarm probability tradeoff because the cold storage monitor can monitor both temperature and electrical current.
The cold storage monitor can offer improved performance for cold storage devices equipped with an automatic defroster. For these types of cold storages devices, the temperature monitoring algorithm running in the cold storage monitor could be configured to use a different set of detection thresholds during and just after a defrost cycle than it does at all other times. For example, the temperature monitoring algorithm could be configured to generate a temperature-out-of-range alarm if the temperature in its cold storage device exceeds 8 degrees Celsius for more than 5 minutes unless it is within 30 minutes of the defroster being turned on, in which case the temperature and time thresholds could be increased to 9 degrees Celsius and 30 minutes, respectively. Using this approach, the cold storage monitor would almost always have a detection latency of at most 5 minutes for refrigerator malfunctions that cause the temperature to go out-of-range, unless the malfunction occurs within 30 minutes of the defroster going on. Since the defroster in many refrigerators goes on at most once per 1 to 3 days, this is a significant improvement.
Another way the monitor can offer improved detection latency performance involves alarming immediately if the cold storage device does not appear to be changing its functional operating state in an appropriate way. For example, as mentioned earlier, to mitigate false alarms, a relatively large detection time could be used before a temperature out-of-range alarm is generated. However, the cold storage monitor could alarm right away if the temperature is too high and the compressor has not been turned on, or if the temperature is too low and the compressor has not been turned off since in both of these cases, it is not 100% certain there is a malfunction and there is no need to wait extra time to prevent a false alarm.
In a large-scale implementation containing a large number of cold storage monitors and cold storage devices, characteristics from multiple instances of the same cold storage device could be used to detect performance problems. For example, the server 20 (
The cold storage monitor could be configured to measure the ambient temperature outside the cold storage device and possibly also the humidity inside and outside of the cold storage device. The cold storage device will generally have to consume more energy, on average, to maintain a fixed internal temperature when there is high ambient temperature or humidity than it would otherwise. With this in mind, instead of using fixed alarm thresholds for an unexpected compressor duty cycle alarm, a performance improvement could be realized if the cold storage monitor made the alarm thresholds vary as a function of one or more of the internal operating temperature, external ambient temperature and external humidity. A cold storage monitor could build data in a table over time characterizing how the mean compressor duty cycle varies as a function of these three variables and generate an alarm if the duty cycle exceeds the mean at a particular temp and humidity by an appropriate threshold (e.g., 2 standard deviations). To further improve performance, the data could be stored in a database on the server 20 and could be built from multiple instances of the same cold storage device (i.e., multiple instances from the same manufacturer and model number).
The operating state detection procedure 600 is further described using an example based on current and temperature data taken from a pharmaceutical refrigerator over a period of 7 days. Reference is made to
Using the above feature vector definition,
The feature vectors in cluster 905 have a relatively small amount of current overshoot, and a small steady-state current level, and are therefore associated with the compressor, defroster and door light all being turned off. Cluster 910 has a low overshoot but slightly larger steady-state current, and is therefore associated with the compressor and defroster being off and door light being on. Clusters 915 and 920 are associated with the defroster being on while the compressor is off. The door light is off for vectors in cluster 915 and on for those in cluster 920. Clusters 925 and 930 are associated with the compressor being on while the defroster is off; the door light is off for vectors in cluster 925 and on for those in cluster 930.
The learning process step 710 of identifying clusters of accumulated feature vectors using any one of a number of well-known clustering algorithms would identify the 6 clusters 905, 910, 915, 920, 925 and 930 shown in
The step 712 of associating functional operating states with clusters would assign all clusters having a relatively high current overshoot level and a medium-to-high steady-state current level (since this is characteristic of the compressor being on) to the “defroster off, compressor on” functional operating state. Since there are two such clusters (925 and 930) in this example, the cluster having the higher steady-state level (930) would be assigned to the “defroster off, compressor on, door light on” operating state; the cluster having the lower steady-state level (925) would be assigned to the “defroster off, compressor on, door light off” state. The associating step 712 would further assign any clusters having a low overshoot current and low steady-state current (characteristics of the compressor being off and defroster being on—clusters 905 and 910 in this example) to the “defroster off, compressor off, door light on” and “defroster off, compressor off, door light off” states, and assign any clusters having a high steady-state current and low overshoot current to the “defroster on, compressor off, door light on” and “defroster on, compressor off, door light off” states. The functional operating states and decision regions associated with each of the six clusters for this example are summarized in table 1050 of
In steps 715 and 720 of the learning process, learning statistics are calculated from which alarm thresholds are generated. There are two learning statistics used in this example: compressor duty cycle, and min/max overshoot.
Referring back to
Turning again to
Step 820 of the monitoring process compares the current overshoot level extracted from feature vector 1035 to the min and max alarm thresholds for current overshoot generated from the learning process, and generates an alarm notification if either threshold is breached. Also in step 820, the 24 hour moving average compressor duty cycle is compared to the min and max alarm thresholds from the learning process, and generates an alarm notification if either threshold is breached.
The foregoing description of the example depicted with reference to
In one form, a method is provided. The method determines the operating state of a cold storage device, and comprises: monitoring electrical current consumption and temperature inside a cold storage device; identifying operational state changes of the cold storage device using detected changes in the electrical current consumption; calculating a multi-dimensional feature vector comprising a plurality of electrical and thermal parameters derived from the monitored electrical current consumption and temperature of the cold storage device between consecutive operational state changes; performing a learning process that includes: accumulating feature vectors over a period of time; identifying clusters of accumulated feature vectors; associating one or more functional operating states of the cold storage device with one or more of the clusters; calculating learning statistics based on one or more of: a frequency that the cold storage device enters the one or more functional operating states; a variation of a feature vector parameter within one or more of the clusters; and generating an alarm threshold from the learning statistics; performing a monitoring process that includes: determining a nearest cluster to the feature vector; determining one or more current functional operating states of the cold storage device from the functional operating states associated with the nearest cluster; calculating a monitoring statistic based on one or more of: the one or more current functional operating states; one or more feature vector components; and sending an alarm notification if the monitoring statistic exceeds the alarm threshold.
The embodiments may take the form of a system. The system comprises: a monitoring device configured to monitor one or more of electrical current consumption and temperature inside a cold storage device; and a server coupled to the monitoring device, wherein the server is configured to perform operations including: identifying operational state changes of the cold storage device using detected changes in the electrical current consumption; calculating a multi-dimensional feature vector comprising a plurality of electrical and thermal parameters derived from the monitored electrical current consumption and temperature of the cold storage device between consecutive operational state changes; performing a learning process that includes: associating one or more functional operating states of the cold storage device with one or more of the clusters; calculating learning statistics based on one or more of: a frequency that the cold storage device enters the one or more functional operating states; a variation of a feature vector parameter within one or more of the clusters; and generating an alarm threshold from the learning statistics; performing a monitoring process that includes: determining a nearest cluster to the feature vector; determining one or more current functional operating states of the cold storage device from the functional operating states associated with the nearest cluster; calculating a monitoring statistic based on one or more of: the one or more current functional operating states; one or more feature vector components; and sending an alarm notification if the monitoring statistic exceeds the alarm threshold.
In addition, the embodiments presented herein may take the form of one or more non-transitory computer readable storage media encoded with instructions, that when executed by a processor, cause the processor to perform operations including: monitoring electrical current consumption and temperature inside a cold storage device; identifying operational state changes of the cold storage device using detected changes in the electrical current consumption; calculating a multi-dimensional feature vector comprising a plurality of electrical and thermal parameters derived from the monitored electrical current consumption and temperature of the cold storage device between consecutive operational state changes; performing a learning process that includes: associating one or more functional operating states of the cold storage device with one or more of the clusters; calculating learning statistics based on one or more of: a frequency that the cold storage device enters the one or more functional operating states; a variation of a feature vector parameter within one or more of the clusters; and generating an alarm threshold from the learning statistics; performing a monitoring process that includes: determining a nearest cluster to the feature vector; determining one or more current functional operating states of the cold storage device from the functional operating states associated with the nearest cluster; calculating a monitoring statistic based on one or more of: the one or more current functional operating states; one or more feature vector components; and sending an alarm notification if the monitoring statistic exceeds the alarm threshold.
From the above description, those skilled in the art will perceive improvements, changes and modifications. Such improvements, changes and modifications are within the skill of one in the art and are intended to be covered by the appended claims.
This application is a continuation-in-part of U.S. application Ser. No. 15/727,771, filed Oct. 9, 2017, which in turns claims the benefit of U.S. Provisional Patent Application No. 62/409,947, now abandoned, “Machine Learning Algorithms for Health Monitoring of Refrigerators, Freezers and Other Temperature Control Systems,” filed Oct. 19, 2016, and of U.S. Provisional Patent Application No. 62/456,897, “Misc Enhancements To Refrigerator Operating State Detection Algorithms,” filed Feb. 9, 2017, both of which are incorporated by reference herein in their entirety.
Number | Name | Date | Kind |
---|---|---|---|
3976985 | Schalow | Aug 1976 | A |
4325223 | Cantley | Apr 1982 | A |
4502287 | Hare et al. | Mar 1985 | A |
5402112 | Thompson | Mar 1995 | A |
8725455 | Kriss | May 2014 | B2 |
9020769 | Rada et al. | Apr 2015 | B2 |
20070156373 | Yamashita | Jul 2007 | A1 |
20080209925 | Pham | Sep 2008 | A1 |
20110144944 | Pham | Jun 2011 | A1 |
20140165614 | Manning et al. | Jun 2014 | A1 |
20140324388 | Kriss | Oct 2014 | A1 |
Number | Date | Country | |
---|---|---|---|
20180340729 A1 | Nov 2018 | US |
Number | Date | Country | |
---|---|---|---|
62456897 | Feb 2017 | US | |
62409947 | Oct 2016 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 15727771 | Oct 2017 | US |
Child | 16054208 | US |