GB2576358A - Sensing fluid flow for estimating fluid flow state - Google Patents
Sensing fluid flow for estimating fluid flow state Download PDFInfo
- Publication number
- GB2576358A GB2576358A GB1813395.9A GB201813395A GB2576358A GB 2576358 A GB2576358 A GB 2576358A GB 201813395 A GB201813395 A GB 201813395A GB 2576358 A GB2576358 A GB 2576358A
- Authority
- GB
- United Kingdom
- Prior art keywords
- decision tree
- conduit
- data
- temperature
- fluid flow
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/02—Investigating fluid-tightness of structures by using fluid or vacuum
- G01M3/26—Investigating fluid-tightness of structures by using fluid or vacuum by measuring rate of loss or gain of fluid, e.g. by pressure-responsive devices, by flow detectors
- G01M3/28—Investigating fluid-tightness of structures by using fluid or vacuum by measuring rate of loss or gain of fluid, e.g. by pressure-responsive devices, by flow detectors for pipes, cables or tubes; for pipe joints or seals; for valves ; for welds
- G01M3/2807—Investigating fluid-tightness of structures by using fluid or vacuum by measuring rate of loss or gain of fluid, e.g. by pressure-responsive devices, by flow detectors for pipes, cables or tubes; for pipe joints or seals; for valves ; for welds for pipes
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F17—STORING OR DISTRIBUTING GASES OR LIQUIDS
- F17D—PIPE-LINE SYSTEMS; PIPE-LINES
- F17D5/00—Protection or supervision of installations
- F17D5/02—Preventing, monitoring, or locating loss
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01F—MEASURING VOLUME, VOLUME FLOW, MASS FLOW OR LIQUID LEVEL; METERING BY VOLUME
- G01F25/00—Testing or calibration of apparatus for measuring volume, volume flow or liquid level or for metering by volume
- G01F25/10—Testing or calibration of apparatus for measuring volume, volume flow or liquid level or for metering by volume of flowmeters
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/007—Leak detector calibration, standard leaks
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Fluid Mechanics (AREA)
- Examining Or Testing Airtightness (AREA)
Abstract
A method of producing a device configured to estimate at least one fluid flow state in a conduit based on input data sensed by the device, comprises: obtaining training data S1 comprising information corresponding to at least one known fluid flow state and associated input data; applying a machine learning algorithm S2 to the obtained training data, in order to generate a decision tree S3 configured to output a fluid flow state based on sensed input data; and storing the generated decision tree S4 in a memory of the device. The device may comprise a sensor and may be installed at a first property, wherein obtaining the training data may comprise sensing input data using one or more devices installed at one or more further properties or at a laboratory installation. The input data may comprise temperature data, which may be the conduit temperature or an ambient temperature. The fluid flow state may be a leak state or a no-leak state, or the probability of either state.
Description
Sensing fluid flow for estimating fluid flow state
Field of the invention
The invention relates but is not limited to sensing fluid flow or detecting leaks or a method for estimating a fluid flow state in a conduit of a property. In examples of the disclosure, the fluid flow state may be estimated as a leak state or a non-leak state.
Background
Leaks in a property can cause significant damage and may be difficult to detect until serious damage has occurred. A technical solution which provides early warning of potential leaks would therefore be advantageous but despite the long-standing problem, no simple and effective solution has been proposed. Placing a network of fluid detectors adjacent pipes, pressure and/or flow sensors on pipes and fluid consumers and throughout a property’s pipe network can give some warning but is expensive and technically difficult to install and configure as pipes may be inaccessible and in any event such a solution still does not reliably detect leaks. It is difficult to detect a leak in a conduit of a property based simply on fluid flow, because it is difficult to distinguish between an intended use of the fluid (such as a flush and/or a shower in a case where the fluid is water) and a problematic leak where the fluid is simply wasted.
Summary of the Invention
Aspects and embodiments of the invention are set out in the appended claims. These and other aspects and embodiments of the invention are also described herein.
In a first aspect of the invention, there is provided a method of producing or configuring a device configured to estimate at least one fluid flow state in a conduit based on input data sensed by the device, the method comprising: obtaining training data comprising information corresponding to at least one known fluid flow state and associated input data; applying a machine learning algorithm to the obtained training data, in order to generate a decision tree configured to output a fluid flow state based on sensed input data; and storing the generated decision tree in a memory of the device.
-2The device may comprise one or more sensors and obtaining the training data preferably comprises sensing input data using the device and associating at least one known fluid flow state with said sensed data.
The device may be installed, or intended for installation at, a first property, and obtaining the training data may comprise: sensing input data using one or more devices installed at one or more further properties or at a laboratory installation and associating at least one known fluid flow state with said sensed data.
The training data preferably comprises a plurality of training samples, each comprising a set of input feature data values and a flow state classification indicating a known fluid flow state corresponding to the input feature data values, optionally wherein feature data values comprise sensor data values recorded by one or more sensors and/or derived data values derived from sensor data values.
The input data (or feature data) preferably comprises temperature data. The temperature data may comprise one or more of: a conduit temperature of a monitored conduit; an ambient temperature in a vicinity of the monitored conduit, for example in a room of the property where the monitored conduit is located; a difference between the ambient temperature and the conduit temperature; an average temperature or temperature difference over a period of time; a temperature gradient over a period of time; a range of temperatures or temperature differences over a period of time; and a ratio of temperatures or temperature differences.
The at least one fluid flow state may include at least one of: one of a predetermined set of discrete states including at least one of a leak state, a non-leak state, a low severity leak state, a medium severity leak state, a high severity leak state, a device off-pipe state, a device poor connection state, and a filling of header tanks state; and/or one value of a continuous set of values associated with the fluid flow in the conduit of the property, such as a flow rate in the conduit. In an example, the output flow state of the decision tree is one of a leak state and a non-leak state.
-3The decision tree is preferably arranged to classify a fluid flow state based on input data comprising some or all of the feature data used to generate the decision tree by the machine learning algorithm.
Preferably, the decision tree comprises a plurality of decision nodes, preferably wherein each decision node applies a condition to input data, optionally by comparing input data to a threshold, to determine a tree traversal path. The decision tree preferably comprises a plurality of leaf nodes defining output states of the decision tree, preferably wherein traversal of the tree by applying a given set of decision nodes along a traversal path terminates at a given leaf node indicating an output of the decision tree. The decision tree is preferably arranged to output, based on input data, one of: a flow state classification, optionally as one of a leak state and a no-leak state; and a probability of a given flow state, optionally a leak probability.
The decision tree may comprise at least 10 nodes, preferably at least 20 nodes, more preferably at least 50 nodes, and more preferably still at least 100 nodes.
Preferably, the device further comprises a processor configured to apply the generated decision tree to input data sensed by the device for estimating at least one fluid flow state in the conduit of the property.
Generating the decision tree may comprise sensing input data of a conduit at a property, and assigning at least one known fluid flow state to the sensed input data, based on knowledge of fluid flow in that conduit.
The method may be performed (partly or wholly) at a computer system separate (preferably remote) from the device. Preferably, the storing step comprises transmitting the generated decision tree to the device via a network, the device receiving and storing the decision tree. The process may be repeated e.g. periodically to update the stored decision tree at the device.
-4The decision tree may be generated, stored and/or transmitted in the form of one or more of: a data representation of the decision tree; and executable code for applying the decision tree to input data.
In a further aspect, the invention provides a method for estimating at least one fluid flow state in a conduit of a property, comprising: obtaining, at a device (preferably produced according to any method set out above), input data sensed by the device; applying the decision tree to the sensed input data; and estimating the at least one fluid flow state in the conduit of the property, based on the applying. The device may comprise one or more sensors and wherein obtaining the sensed input data comprises: the sensor(s) of the device sensing the input data. Preferably, the sensed input data comprises at least one of: temperature data associated with a temperature of the conduit and/or a temperature of the property (e.g. ambient temperature).
The invention further provides apparatus (e.g. a detection device) configured to estimate at least one fluid flow state in a conduit at a property based on input data sensed by the apparatus, the apparatus comprising a memory storing a decision tree generated by a method comprising: obtaining training data comprising information corresponding to at least one known fluid flow state and associated input data; and applying a machine learning algorithm to the obtained training data, in order to generate the decision tree. The apparatus may be produced or configured by any method as set out above. The apparatus may further comprise a processor, and wherein the memory of the apparatus further comprises instructions which, when executed by the processor, enable the processor to perform any method as set out herein.
The invention also provides a (tangible) computer program or a computer program product I computer readable medium, comprising instructions which, when executed by a processor, enable the processor to perform any method as set out herein and/or to provide, produce or configure a detection device or other apparatus as described herein.
Any feature in one aspect of the invention may be applied to other aspects of the invention, in any appropriate combination. In particular, method aspects may be applied to apparatus and computer program aspects, and vice versa.
-5Furthermore, features implemented in hardware may generally be implemented in software, and vice versa. Any reference to software and hardware features herein should be construed accordingly.
Brief Description of Drawings
Embodiments of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which:
Figure 1 shows a flow chart illustrating an example method according to the disclosure;
Figure 2 schematically illustrates an example system and an example device configured to implement the example method of Figure 1;
Figure 3 illustrates an example decision tree according to the disclosure;
Figure 4 schematically illustrates example data from a property over a three month period;
Figure 5 schematically illustrates the ambient and conduit temperatures as measured by a device, and a water temperature as estimated by a water supply estimation; and
Figure 6 shows a flow chart illustrating another example method according to the disclosure.
In the figures, similar elements bear identical numerical references.
Detailed Description of Example Embodiments
Embodiments of the invention provide a process for configuring a flow state detection device, to enable the device to estimate or detect a fluid flow state in a pipe or other conduit (or in a system of such conduits). In typical examples the device is configured to detect presence or absence of a leak, though other fluid flow states may be detected.
Figure 1 shows a flow chart illustrating an example method 100 according to the disclosure. The method 100 is illustrated in Figure 1 in connection with Figure 2, which
-6shows a flow state detection device 15 configurable by the method 100 to estimate at least one fluid flow state in a conduit 61 of a property 60 based on input data sensed by the device 15.
As described in more detail below, configuration of the device 15 involves storing a decision tree at the device. The decision tree is derived from training data using a machine learning algorithm, and is arranged to produce an output indicative of a flow state in a monitored conduit, in particular examples either a leak or a no-leak state.
Note that while reference is made herein to the flow state of the monitored conduit, a particular detected state may actually indicate a state or event elsewhere in a network of conduits including the monitored conduit. For example, a leak elsewhere in the network (typically but not necessarily downstream of the monitored conduit) may be reflected in, and hence detected from, the flow state of the monitored conduit. Thus, it will be understood that detecting a “leak” or “no-leak” or similar state based on monitoring of a given conduit does not necessarily indicate that the leak is at or near the monitored conduit, or the location where relevant sensors are positioned.
The decision tree is arranged to be simple to evaluate, after it is stored in the memory of the device 15, even though the process for deriving it from training data may be computationally intensive.
Once configured, the device 15 may provide an accurate estimation of the fluid flow state, e.g. for a leak detection, by applying the decision tree to input data sensed by the device 15. The detection process is illustrated (as process 200) in Figure 6 (described later).
As illustrated in Figure 1, the process 100 for configuring a detection device comprises, in overview:
obtaining, at S1, training data comprising information corresponding to at least one known fluid flow state and associated input data;
applying, at S2, a machine learning algorithm to the obtained training data; generating, at S3, a decision tree, based on the applying; and
-7storing, at S4, the generated decision tree in a memory of the device.
Computer system and detection device
Figure 2 schematically illustrates an example computer system 10 and device 15 configured to implement, at least partly, the example method 100 of Figure 1. In particular, in a preferred embodiment, computer system 10 executes the machine learning algorithm to generate the decision tree to be stored on the device 15. Although a single device 15 is shown for clarity, the computer system 10 may communicate and interact with multiple such devices. The training data may itself be acquired using the device 15, using other, similar devices and/or using other sensors and data sources.
The computer system 10 of Figure 2 comprises a memory 11, a processor 12 and a communications interface 13.
The system 10 may be configured to communicate with one or more fluid meters 20 and/or one or more temperature sensors 21 and/or one or more devices 15, via the interface 13 and a first link 30 (e.g. Wi-Fi connectivity).
The system 10 of Figure 2 is also configured to be connected to one or more user interfaces 50, via the interface 13 and a second link 40 (e.g. Wi-Fi connectivity) between the interface 13 and the user interfaces 50.
The memory 11 is configured to store data, for example for use by the processor 12. The memory 11 may also comprise a first database server 111 configured to store data such as the training data.
The memory 11 may also comprise a second database server 112 configured to store data received from the user interfaces 50 over the link 40.
In some examples, the processor 12 of the system 10 may be configured to perform, at least partly, at least some of the steps of the method 100 of Figure 1 and/or the method 200 of Figure 6. Alternatively or additionally, some of the steps of the above methods
-8may be performed, at least partly, by another entity in the system 10, such as the server 111 or 112 as non-limiting examples.
The detection device 15 of Figure 2 comprises a memory 151, a processor 152 and a communications interface 153 (e.g. Wi-Fi connectivity) allowing connection to the interface 13. The device 15 may also comprise one or more temperature sensors 21 (e.g. thermistors) and/or one or more user interfaces 50. In some examples, the processor 152 of the device 15 may be configured to perform, at least partly, at least some of the steps of the method 100 and/or the method 200. Temperature sensors 21 may be integrated into the device or connected to the device by wired or wireless connection and are used to sense one or more temperature values relevant to flow/leak detection, such as pipe temperature, ambient temperature, fluid temperature and the like.
The device is installed at a property 60 (e.g. a house, apartment, office or other building), for example in a vicinity of (or affixed to) a conduit 61 (e.g. water or other fluid pipe) being monitored. The device may be attached such that a temperature sensor 21 directly contacts the conduit to measure the temperature of the conduit surface.
In some examples, the device 15 is a battery-powered device with the interface 153 having Wi-Fi connectivity. In some examples, the device 15 may only connect to the system 10, e.g. every 6 hours (e.g. to save battery life), unless a leak has started or ended, in which case the device 15 may connect to the system 10 more frequently, e.g. immediately. In some examples, the device 15 performs some or all of the steps of the method 200 (Figure 6), e.g. in order to determine whether a leak has recently started or ended.
Alternatively or additionally, in some examples, the device 15 may send all new temperature data (e.g. sampled at ten second intervals) to the system 10, each time the device 15 connects to the system 10.
The system 10 may output trigger data (e.g. the system 10 may send a notification to the person living in the property) if a new leak has been detected.
-9Decision-tree based flow classification
Flow state classification is based on a decision tree, an example of which is depicted in Figure 3. The decision tree comprises a set of internal (non-leaf) decision nodes (e.g. S40, S41, S42, S46), each defining some test or condition on a particular feature of the input data (e.g. comparison of a temperature to a threshold). The leaf nodes (e.g. S45, S44, S43, S48, S47) define outputs of the decision tree, in the form of flow state indications.
The decision tree is applied by following the tree from the root node, applying the tests specified at internal decision nodes, and following a given branch from each decision node depending on the test outcomes. Typically the test outcome is binary (e.g. does a temperature value exceed some threshold) with two outgoing branches from the decision node corresponding to yes/no outcomes, but nodes with more than two possible outcomes and outgoing branches are possible. The tree is traversed, applying the test conditions in order and following branches as dictated by the test outcomes, until a leaf node is reached. Thus, a given traversal of the tree corresponds to a path from the root to a leaf node. The leaf node then determines the output state of the decision tree. Each path through the decision tree from root to a leaf node essentially encodes a complex classification rule as a conjunction of test conditions.
In the example of Figure 3, a given path may traverse decision nodes S40 (yes branch), S41 (no branch), and S42 (no branch) arriving at leaf node S43, which is associated with a “leak” classification. Typical decision node test conditions include e.g. numerical comparison of values (e.g. greater than I less than I equals comparisons), such as comparison of two input feature values or comparison of an input feature value to a predetermined or dynamically determined threshold.
The leaf node output states may specify leak classifications (e.g. one of a set of classification labels such as “leak” and “no leak”), as depicted in Figure 3. In some embodiments, instead of leak I no leak classification, the leaf nodes may each specify a leak probability (or equivalently a no-leak probability). This is then converted into the final
-10classification, e.g. by applying a threshold (e.g. such that leak probabilities exceeding the threshold are classified as leaks; otherwise the output classification is no-leak). For example, in Figure 3 the leaf nodes could specify as output values leak probabilities as follows: S45: 20%, S44: 27%, S43: 100%, S48: 46%, S47: 93%. Applying a leak probability threshold of e.g. 70% then yields the final classifications shown in Figure 3.
While in this example the decision tree is used simply to detect presence or absence of a leak, in some examples, the at least one fluid flow state detectable may include additional states, for example one or more of a predetermined set of discrete states including: a leak state, a non-leak state, a low severity leak state, a medium severity leak state, a high severity leak state, a device off-pipe state, a device poor connection state, and/or a filling of header tanks state. Alternatively or additionally, the decision tree may output a value of a continuous set of values associated with the fluid flow in the conduit of the property, such as a flow rate in the conduit (or a leak probability as previously indicated).
The decision tree performs flow state classification on the basis of input data for a predetermined set of features (“feature data”). These features may directly correspond to input data measured by a sensor (e.g. pipe temperature or fluid temperature, measured using device sensors 21) or may be derived values computed from one or more input data values (e.g. an input feature may be the computed difference between pipe and ambient temperature, a pipe temperature gradient over time etc.)
Referring back to Figure 1, the decision tree is created by a machine learning algorithm and stored (S4) in the detection device 15. The decision tree may be created and stored using any suitable representation, for example as a data description comprising data elements specifying decision nodes and their test conditions, leaf nodes and their flow classification outputs (e.g. flow state labels or probabilities), and node connections. Such a data description could be encoded e.g. using XML or using a bespoke binary representation. The data description is then interpreted by a module running on device 15 when applying the decision tree.
-11Alternatively, the machine learning algorithm may generate the decision tree directly as executable code (e.g. machine code, virtual machine byte code or interpretable script). This may be in the form of a code routine that the device 15 can invoke to apply the decision tree. An example is shown below in the form of an excerpt from a function definition taking the input feature data as parameters (line 1), applying test conditions (lines 2-11) and generating a return value (line 12) specifying the output flow state probability or flow state label (in numerical representation). Only a single tree path is shown:
i def tr^e(abs^eenbiff} difffronpe, pipings, pipMfttedranch ratio.„st.a):
if ratio 6,800568518567:
if rati0.,.Std <·· 0,0598599091173:
if ratio <- 0,15074078493:
:> U dtKtoncte 0 045000m7581:
if abs^eaaDiff <-· 0 8.65^0011444:
0183878155759:
If o B,18S0000033M:
if absM^anDiff 0,0590000003576:;
:? if atffRanae 'J if ratio 4 ^85804797 return O
Regardless of the representation of the decision tree, the decision tree effectively defines a decision algorithm (comprising a set of rules) for classifying flow status based on input data.
Obtaining the training data
Referring back to Figure 1, the decision tree is generated based on training data (S1).
The training data consists of training samples, each comprising a set of input data in the form of feature values and labelled with a known fluid flow state classification associated with those feature values. The known fluid flow state classifications are also referred to as the “ground truth”. Thus, a training sample corresponds to an instance when the flow state classification (e.g. “leak” or “no leak”) is known. The associated measured characteristics of the conduit and environment during that state (e.g. temperature
-12measurements) form the basis for the input data (feature values) associated with that known state. Thus a training sample may take the form <F1, F2, ... Fn, label> where F1Fn are the input feature values and “label” is the associated classification label.
Training samples may correspond to past data acquired in situ by the same device 15 that is being configured, e.g. using sensors 21 of the device 15 as depicted in Figure 2. Alternatively, training samples may have been obtained in a different environment, e.g. using a similar device (or equivalent set of sensors) installed in a different (but preferably similar) building environment, or in a controlled test configuration in a laboratory environment.
The classification labels for the training samples (specifying known fluid flow states) may be known in advance (e.g. when training samples are derived experimentally in a laboratory setting) or may be obtained using a secondary leak detection system. In some examples, a domain specialist may manually label the training data samples with the ground truth classification (e.g. ‘leak’ or ‘no leak’ of an event).
The training samples preferably comprise (or are derived from) data samples recorded at intervals of less than 24 hours or at least once per day. Most preferably the samples may include both historical measures associated with the property and/or the flow itself (such as a history of known fluid flow states). More preferably the samples may include several readings per day, typically at least morning, day, evening and night or at least about four readings per day. In some examples the training samples may comprise readings separated by a much lower interval, such as a second, five seconds or ten seconds as non-limiting examples. The sampling intervals need not be equal and in some embodiments sampling times may be adjusted based on observed fluid usage, for example so that usage at morning, evening, night, etc., corresponds to times of typical usage.
The input feature data may be based on temperature data (e.g. measured using temperature sensors 21), flow measurements (e.g. measured using flow meters 20), or other sensor data. Feature data may also include derived or estimated data. For example, input features may include one or more of:
-13a conduit temperature T1 (e.g. measured on a surface of the conduit);
a fluid temperature T2 (this may be measure or estimated as described below); an ambient temperature T3 in a vicinity of the conduit (inside the property); an external temperature T4 (outside the property);
for any of the above temperatures:
a difference between two temperatures e.g. between the ambient temperature and the conduit temperature;
an average temperature or temperature difference over a period of time;
a temperature gradient over a period of time;
a range of temperatures or temperature differences over a period of time;
a ratio of temperatures or temperature differences.
In some examples, the temperature data may be provided and/or sensed e.g. by temperature sensors 21 and/or the device 15 as described above.
Alternatively or additionally, in some examples, the feature data may comprise features which are calculated from a window of temperature data (e.g. the past 100 temperature readings ~ e.g. 15 mins of data, as a non-limiting example).
In general, the input features used by the machine learning algorithm (whether base or derived values) are the same as the features applied by the generated decision tree (i.e. in the test conditions specified by decision nodes of the tree). However, it may be the case that one or more features of the training samples are not found by the machine learning algorithm to be sufficiently indicative of flow status and thus may not appear in decision node conditions in the final generated tree.
In some examples, a flow ratio may be another calculated feature used as an input into the machine learning algorithm and/or the decision tree.
In some examples, the flow ratio data may comprise data associated with at least one of:
• a ratio calculated by dividing, on the one hand, a difference between the ambient temperature T3 and the conduit temperature T1 by, on the other hand, a
-14difference between the ambient temperature T3 and a flow temperature T2, such that:
ratio=(T3-T1)/(T3-T2); and/or • a standard deviation of the ratio.
As described in more detail below, in some examples, the fluid temperature T2 may be estimated from multiple days of temperature data in relation to the temperature T1 of the conduit 61 and/or the temperature T3 of the air surrounding the conduit 61.
In some examples, the sampling frequency of the raw temperature data (such as the measurements of the temperatures T1 and T3) and/or of the calculated features (e.g. used as input into the machine learning algorithm and/or the decision tree) may be e.g. between 1 second and 15 minutes, e.g. every 10 seconds as non-limiting examples.
Generating the decision tree
Referring back to Figure 1, the decision tree is built by applying a machine learning algorithm to the training data. Any suitable machine learning algorithm may be used for building the decision tree. For example, approaches based on recursive partitioning or evolutionary algorithms may be used. Specific examples of decision tree learning algorithms include ID3 and C4.5.
The learning process is typically computationally intensive and may involve large volumes of training data. In some examples, the processor 12 of system 10 may comprise greater computational power and memory resources than the processor 152 of the device 15. The decision tree generation is therefore performed at least partly, remotely from the device 15, at computer system 10. In some examples, at least steps S2 and/or S3 of the method 100 are performed by the processor 12 of the computer system 10. However, if sufficient processing power is available locally then decision tree learning could be performed (at least partly) by processor 152 of device 15.
The machine learning step involves inferring behaviours and patterns based on the training data (e.g. history of the past known fluid state and/or input feature data) and
-15encoding the detected patterns in the form of the decision tree. The generated decision tree may thus represent patterns in the input data.
The machine learning algorithm may be controlled based on hyper-parameters. The hyper-parameters may constrain the search process and/or define a complexity and/or a size of the output decision tree (e.g. maximum nodes/tree levels). In practice, given suitable training data, the resulting decision tree may be highly complex, with tens or hundreds of nodes or more, encoding hundreds or even thousands of classification rules. After the initial learning stage (S2), decision tree pruning may be used during generation of the final decision tree (S3) to reduce the size of the tree, e.g. to ensure that the complexity of the tree remains within the computational and storage resources of the detection device 15.
As discussed above, the final decision tree provides a predictive model which predicts a flow state based on unseen samples of feature data (corresponding to the feature data on which the tree was learnt). In some examples, the device 15 may be configured to use the decision tree to predict a probability of an event being a leak. Alternatively or additionally, an example decision made by the device 15 may be determining a presence or absence of a leak.
After the decision tree is generated at S3, the decision tree is stored in the memory 151 of the device 15. The device 15 may be connected temporarily to the system 10 to transfer the generated decision tree (e.g. as a data file or executable code) or transfer may occur using a storage medium (e.g. memory card). In a preferred approach, the decision tree is transferred to the device 15 from system 10 over the network connection (this could include transmission over the Internet from a central location of system 10 to a local network in property 60). The decision tree is then installed at the device 15. The decision tree could be installed as part of a firmware update of device software, or independently.
Installation of the decision tree may be performed once (e.g. at time of manufacture or installation) or repeatedly (e.g. as a regular update). The latter approach can allow the
-16classification performance of the decision tree to be improved over time, as new training data becomes available.
Applying the decision tree to perform flow state estimation
After the device 15 has been configured with the decision tree, the device can use the decision tree based on locally recorded sensor data to detect flow (e.g. leak) states.
Figure 6 shows a flow chart illustrating an example method 200 for flow state detection. The method 200 is performed by device 15 (as shown in Figure 2).
The method 200 comprises:
obtaining, at S10, input data sensed by the device 15;
applying, at S20, the decision tree to the sensed input data; and estimating, at S30, the at least one fluid flow state in the conduit of the property, based on the applying.
The decision tree is applied to feature data corresponding to the feature data on which the learning algorithm operated when generating the decision tree. As indicated above, this may include direct sensor measurements (e.g. conduit temperature T1, ambient temperature T3 etc.) as well as derived feature values (e.g. averages, differences etc.) Thus, step S10 includes deriving or computing the relevant input feature values from raw acquired sensor data.
Application of the decision tree in step S20 then involves applying the test conditions specified by the decision nodes of the tree, traversing the tree along a path from the root node to a leaf node. In step S30, the tree output defined for the relevant leaf node is then used to determine the fluid flow state (e.g. based directly on a flow state classification specified by the leaf node or by applying a threshold to a leak probability value specified by the leaf node to determine whether a leak is present).
Optionally the method 200 may further comprise outputting, e.g. at S50, trigger data to trigger an intervention in response to estimating that the flow state is a leak state. The
-17intervention may include outputting an alarm signal and/or trigger a further estimation of the state (e.g. for verification). In some examples, several types of alarm signals may be considered. The different types of alarm signals may enable taking into account the seriousness of the estimated state. A first type of alarm signal may cause a message being sent to a person living in the property (such as a text message, a phone call, etc.). A second type of alarm signal may cause a visit of a professional maintenance staff (such as a plumber) to the property. Other types of alarm signal are envisaged.
Further examples and details
The following describes, purely by way of example, application of the described techniques in a real-world setting.
In this example, five features are extracted as input data for the machine learning algorithm I decision tree from the ambient temperature T3 and the conduit temperature T1:
1. MeanDiff - e.g. an average of the difference between the ambient temperature T3 and the conduit temperature T1, e.g. over a specific past time period;
2. DiffRange - e.g. a range of MeanDiff over a past time period;
3. PipeRange - e.g. a range of the conduit temperatures over a past time period;
4. PipeGradient - e.g. a gradient of the conduit temperature over a given time window, which may be approximated by subtracting a value at a start of the window from a value at an end of the window;
5. PipeLinearity - e.g. an average absolute difference between the conduit temperature and a straight line connecting the most recent pipe temperature and the pipe temperature a given time earlier.
The fluid supply temperature T2 (e.g. water temperature) may be estimated as follows.
Figure 4 schematically illustrates example data from a property over a three month period. Figure 4 schematically illustrates that the conduit temperature drops by a few degrees every time a high flow event occurs.
-18In some examples, the temperature of the water supply (e.g. water temperature T2) may vary from one property to the next, and even across the year for the same property. As a result, the estimate of the supply temperature T2 may be updated for each property every few days. In some examples, a minimum conduit temperature over a rolling three day window may provide an estimate of the water supply temperature (see Figure 4), since the conduit typically reaches a temperature close to the water supply temperature under a sustained full-bore flow (e.g. a shower).
In some examples, during periods when water is not used (e.g. the property is unoccupied) the conduit temperature tracks the ambient temperature T3, and the minimum conduit temperature may not be a good estimate of water temperature (see Figure 4). In such examples, the three day minimum conduit temperature may be updated when there may be a sustained full-bore flow during that three day period, which may be indicated by a difference between the ambient temperature and the conduit temperature greater than 1.5 degrees. In these cases, the last good estimate of the water supply temperature may be used.
The estimation of the flow ratio may be performed as follows.
Figure 5 schematically illustrates the ambient and conduit temperatures as measured by the device 15, and the water temperature as estimated by the water supply estimation described above. Figure 5 also indicates the numerator and denominator of the flow ratio calculation.
In some embodiments, each leak may be classified as either a large or small leak. The flow rate may be assumed to be proportional to the decrease in the conduit temperature T1 during the leak, relative to the water temperature T2. As such, higher flow rates may result in a conduit temperature T1 closer to the water supply temperature T2, while lower flow rates may result in a conduit temperature T1 closer to the ambient temperature T3.
In some examples, the flow ratio r may be calculated by dividing the difference between the ambient temperature T3 and the conduit temperature T1 by the difference between the ambient temperature T3 and the water temperature T2, such that:
-19Use of the flow ratio may allow the decision tree to classify a leak as a large leak if the ratio r is greater than some value (e.g. 0.6), while lower ratios r may be classified as small leaks. In this way, the flow state classification could thus be extended to three discrete states (no leak, small leak, large leak).
Some further examples of discrete output flow states that could be detectable using the described techniques are described in more detail below.
In some examples the device 15 may be mounted on at least one conduit 61, as illustrated in Figure 2. In most cases, the installation of the device 15 is straightforward. However, conduits in the properties vary widely from a property to another property. This results in a small fraction (approx. 10-20%) of poorly installed devices 15. A poor mounting of the device 15 on the conduit 61 may create a poor customer experience, due to e.g. missed leaks and false positive alerts.
In some examples, the data collected during a mounting phase and/or within an initial period may be used to detect cases of bad installation, potentially enhanced by repeating the collection of the data at regular intervals for the whole life of the device 15.
The main reasons for poor installations may comprise at least one of the following (in descending order of frequency): the device 15 not being attached to a conduit (about 50% of the cases), the device 15 being attached to the wrong conduit (i.e., hot water, gas or secondary branch pipe), the device 15 having a poor connection to the conduit (e.g. the device 15 being attached at a joint/bend in the conduit), and the device 15 being attached to a pipe with a warm conduction (e.g. due to proximity with boiler or other hot pipes).
Each of the above mentioned scenarios manifests itself differently in the conduit temperature and in the ambient temperature. Therefore a different detection strategy may be adopted for each of the bad installations mentioned above. Alternatively or additionally, the ability to identify the specific reason for the poor installation may allow
-20notifying the customer and/or the person living in the property with a more tailored suggestion on how to fix the problem. Each type of bad installation is described in more detail in the following.
In cases where the device 15 is not on the conduit, as the device 15 is not in contact with any conduit (e.g. via at least one temperature sensor 21) and is in thermal equilibrium with the surrounding air, the conduit temperature and the ambient temperature fluctuate in a very similar way.
Moreover, no steep change in temperature is observed, which would be normally caused by attaching the device 15 to a conduit and by usage events. Therefore in some examples detection of devices 15 that are not attached to conduits may be performed by simply checking for absence of usage events in a given period of time.
In cases where the device 15 is installed on hot conduits (e.g. conduits that carry hot water), fluid flows are typically rare events so that the conduit is normally cold at the time of installation. As a result, the absence of usage may not be used to detect devices 15 installed on hot conduits. Therefore in some examples detection of devices 15 installed on hot conduits may be performed by using a threshold on the conduit temperature.
In cases where the device 15 is installed on gas pipes (or non-water carrying conduits), the devices 15 are characterised by a drop in temperature when the device 15 is attached to the conduit and by a different response of the conduit temperature to changes in ambient temperature, as the device 15 is in contact with a solid body that has a different thermal capacity than air. Therefore in some examples detection of devices 15 installed on non-water carrying conduits may be performed by the continued absence of usage after the initial temperature drop, once the device 15 has reached thermal equilibrium with the conduit.
By design the device 15 may only be able to detect leaks that are downstream of its location on the conduit. Therefore the device 15 is usually installed upstream of any branch. It should be noted that cases where the device 15 is installed on a branch pipe might be an intentional installation, as the person living in the property might be
-21interested only on leaks on a specific pipe. However the more common occurrence of this scenario is due to the presence of multiple water conduits at the location where the water enters the property, which causes the customer to mistakenly install the device 15 on a branch conduit. In cases where the device 15 is installed on a branch pipe, detection may be performed by the rare detection of usages, approximately one or two per day, which characterise the flow in secondary branch conduits.
In cases where the device 15 is poorly connected to the conduit, a low difference exists between ambient temperature and conduit temperature when water is flowing. This problem usually causes smoother curves at otherwise sharp points like start and end of flow. Detection may be performed by using a classifier, developed to identify lowdifference situations, in order to detect bad installations caused by poor connection.
In cases where the device 15 is connected to a conduit with a warm conduction (e.g. from nearby heat sources, i.e. boilers or hot pipes), during periods where water flow is absent, the conduit temperature may converge to a value higher than the ambient temperature, preventing convergence of temperature from being detected by the device 15 and possibly resulting in false leak alerts. Unlike in all previous cases, in this situation the device 15 can still potentially detect real leaks. The detection may be performed by comparing the sign of the (ambient temperature-conduit temperature) difference, averaged within a long sliding window or during stable periods, and compare the sign with the same value measured during usage events. When the two signs are different then the conduit is classified as being affected by warm conduction.
In some examples, detection of each of the above categories of bad installation is preferably performed as soon as possible. The first opportunity to detect a bad installation is during the mounting, although there is very little data available at this point. Subsequent opportunities exist 24 hours after the mounting process while the person living in the property is still highly engaged, but also this check can be applied at any time during the device’s lifetime (e.g. to check the device 15 has not been knocked off the conduit).
-22ln some examples detection may be performed at the mounting of the device 15. It may be possible to detect a usage (e.g. water flowing) if the user follows these steps: attach the device 15 to the conduit, flush the toilet (causing water to flow through the conduit, and therefore changing rapidly the conduit temperature) and causing the device 15 to upload the sensed data. A similar drop in conduit temperature may be observed when a device 15 is attached to a conduit, as the stop-tap is typically in a location at a different temperature and the conduit is a very good heat conductor. For these reasons, at the mounting it may be possible to accurately detect the case when a device 15 is not attached to a conduit.
Alternatively or additionally, bad installations may be detected after having waited for a period of time after the installation, so that it is more likely to observe the different signatures ofthe various bad installation scenarios. In some examples, a 24-hour period may capture most of the bad installations, given that it allows the device 15 to observe a full daily schedule ofthe person living in the property.
Alternatively or additionally, testing for bad installations may be performed at periodic intervals of time. These may differ for each particular bad-installation type, depending on how easy it is to fix the detected problem. For example, the person living in the property might want to be notified within a day if the device 15 has fallen off the conduit, but might not want to receive notifications every day in the case of warm conduction if the person living in the property cannot fix the problem.
The described decision tree learning techniques may be adapted to detect the various installation problems described above based on the available sensor data, in addition to or instead of leak states.
It will be understood that the present invention has been described above purely by way of example, and modification of detail can be made within the scope of the invention.
For example, the above examples mainly focus on learning decision trees for leak detection based on temperature data. This is based on the fact that pipe, fluid and ambient temperatures can provide useful information on flow conditions in a pipe which
-23 can be obtained using small, cheap and non-invasive sensors (e.g. thermistors). For example the convergence (or lack thereof) between conduit and ambient temperatures can provide an indication of flow rate and of the presence of steady trickle flows which can indicate a downstream leak.
However, the invention is not limited to temperature data and depending on available sensors, other types of data may be used (alternatively or additionally) to produce the feature data input to the machine learning algorithm and decision tree. Examples include flow rate (e.g. measured using a flow meter), pressure or humidity. Data sourced from 10 external systems or services (e.g. weather data from an online weather service) could additionally be used.
Claims (24)
1. A method of producing a device configured to estimate at least one fluid flow state in a conduit based on input data sensed by the device, the method comprising:
obtaining training data comprising information corresponding to at least one known fluid flow state and associated input data;
applying a machine learning algorithm to the obtained training data, in order to generate a decision tree configured to output a fluid flow state based on sensed input data; and storing the generated decision tree in a memory of the device.
2. The method of claim 1, wherein the device comprises a sensor and wherein obtaining the training data comprises:
sensing input data using the device and associating at least one known fluid flow state with said sensed data.
3. The method of claim 1, wherein the device is installed or for installation at a first property, and wherein obtaining the training data comprises:
sensing input data using one or more devices installed at one or more further properties or at a laboratory installation and associating at least one known fluid flow state with said sensed data.
4. The method according to any of the preceding claims, wherein the training data comprises a plurality of training samples, each comprising a set of input feature data values and a flow state classification indicating a known fluid flow state corresponding to the input feature data values, optionally wherein feature data values comprise sensor data values recorded by one or more sensors and/or derived data values derived from sensor data values.
5. The method of any of the preceding claims, wherein the input data comprises temperature data.
6. The method of claim 5, wherein the temperature data comprises one or more of:
-25 a conduit temperature of a monitored conduit;
an ambient temperature in a vicinity of the monitored conduit, for example in a room of the property where the monitored conduit is located;
a difference between the ambient temperature and the conduit temperature;
an average temperature or temperature difference over a period of time;
a temperature gradient over a period of time;
a range of temperatures or temperature differences over a period of time;
a ratio of temperatures or temperature differences.
7. The method of any one of the preceding claims, wherein the at least one fluid flow state includes at least one of:
one of a predetermined set of discrete states including at least one of a leak state, a non-leak state, a low severity leak state, a medium severity leak state, a high severity leak state, a device off-pipe state, a device poor connection state, a filling of header tanks state; and/or one value of a continuous set of values associated with the fluid flow in the conduit of the property, such as a flow rate in the conduit.
8. The method of any of the preceding claims, wherein the decision tree is arranged to classify a fluid flow state based on input data comprising some or all of the feature data used to generate the decision tree by the machine learning algorithm.
9. The method of any one of claims 1 to 8, wherein the decision tree comprises a plurality of decision nodes, preferably wherein each decision node applies a condition to input data, optionally by comparing input data to a threshold, to determine a tree traversal path.
10. The method according to claim 9, wherein the decision tree comprises a plurality of leaf nodes defining output states of the decision tree, preferably wherein traversal of the tree by applying a given set of decision nodes along a traversal path terminates at a given leaf node indicating an output of the decision tree.
11. The method according to any of the preceding claims, wherein the decision tree is arranged to output, based on input data, one of:
a flow state classification, optionally as one of a leak state and a no-leak state; and a probability of a given flow state, optionally a leak probability.
12. The method according to any of the preceding claims, wherein the decision tree comprises at least 10 nodes, preferably at least 20 nodes, more preferably at least 50 nodes, and more preferably still at least 100 nodes.
13. The method of any one of the preceding claims, wherein the device further comprises a processor configured to apply the generated decision tree to input data sensed by the device for estimating at least one fluid flow state in the conduit of the property.
14. The method of any of the preceding claims, wherein generating the decision tree comprises sensing input data of a conduit at a property, and assigning at least one known fluid flow state to the sensed input data, based on knowledge of fluid flow in that conduit.
15. A method according to any of the preceding claims, wherein the method is performed at a computer system separate (preferably remote) from the device.
16. A method according to claim 15, wherein the storing step comprises transmitting the generated decision tree to the device via a network, the device receiving and storing the decision tree.
17. A method according to any of the preceding claims, wherein the decision tree is generated, stored and/or transmitted in the form of one or more of:
a data representation of the decision tree;
executable code for applying the decision tree to input data.
18. A method for estimating at least one fluid flow state in a conduit of a property, comprising:
obtaining, at a device produced according to any one of claims 1 to 17, input data sensed by the device;
applying the decision tree to the sensed input data; and estimating the at least one fluid flow state in the conduit of the property, based on the applying.
19. The method of claim 18, wherein the device comprises one or more sensors and wherein obtaining the sensed input data comprises:
the sensor(s) of the device sensing the input data.
20. The method of claim 18 or 19, wherein the sensed input data comprises at least one of:
temperature data associated with a temperature of the conduit and/or a temperature of the property.
21. Apparatus configured to estimate at least one fluid flow state in a conduit at a property based on input data sensed by the apparatus, the apparatus comprising a memory storing a decision tree generated by a method comprising:
obtaining training data comprising information corresponding to at least one known fluid flow state and associated input data;
applying a machine learning algorithm to the obtained training data, in order to generate the decision tree.
22. The apparatus of claim 21, produced by the method of any one of claims 2 to 17.
23. The apparatus of claim 21 or 22, further comprising a processor, and wherein the memory of the apparatus further comprises instructions which, when executed by the processor, enable the processor to perform the method of any one of claims 18 to 20.
24. A computer program or a computer program product comprising instructions which, when executed by a processor, enable the processor to perform the method according to
-28any one of claims 1 to 20 or to provide the apparatus according to any one of claims 21 to 23.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB1813395.9A GB2576358B (en) | 2018-08-16 | 2018-08-16 | Sensing fluid flow for estimating fluid flow state |
PCT/GB2019/052304 WO2020035696A1 (en) | 2018-08-16 | 2019-08-15 | Sensing fluid flow for estimating fluid flow state |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB1813395.9A GB2576358B (en) | 2018-08-16 | 2018-08-16 | Sensing fluid flow for estimating fluid flow state |
Publications (3)
Publication Number | Publication Date |
---|---|
GB201813395D0 GB201813395D0 (en) | 2018-10-03 |
GB2576358A true GB2576358A (en) | 2020-02-19 |
GB2576358B GB2576358B (en) | 2022-12-28 |
Family
ID=63668085
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
GB1813395.9A Active GB2576358B (en) | 2018-08-16 | 2018-08-16 | Sensing fluid flow for estimating fluid flow state |
Country Status (2)
Country | Link |
---|---|
GB (1) | GB2576358B (en) |
WO (1) | WO2020035696A1 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2595660A (en) * | 2020-06-01 | 2021-12-08 | Homeserve Plc | Flow event classification |
US11209333B2 (en) | 2015-01-07 | 2021-12-28 | Homeserve Plc | Flow detection device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013190396A2 (en) * | 2012-06-12 | 2013-12-27 | TaKaDu Ltd. | Method for locating a leak in a fluid network |
CN106228190A (en) * | 2016-07-26 | 2016-12-14 | 广州华工弈高科技有限公司 | Decision tree method of discrimination for resident's exception water |
US20170076563A1 (en) * | 2015-09-15 | 2017-03-16 | General Electric Company | Systems and methods to provide pipeline damage alerts |
US20180197048A1 (en) * | 2017-01-11 | 2018-07-12 | Ford Global Technologies, Llc | Generating Training Data for Automatic Vehicle Leak Detection |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2017228462B2 (en) * | 2016-03-04 | 2022-02-03 | Transocean Innovation Labs Ltd | Methods, apparatuses, and systems for human machine interface (HMI) operations |
-
2018
- 2018-08-16 GB GB1813395.9A patent/GB2576358B/en active Active
-
2019
- 2019-08-15 WO PCT/GB2019/052304 patent/WO2020035696A1/en active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013190396A2 (en) * | 2012-06-12 | 2013-12-27 | TaKaDu Ltd. | Method for locating a leak in a fluid network |
US20170076563A1 (en) * | 2015-09-15 | 2017-03-16 | General Electric Company | Systems and methods to provide pipeline damage alerts |
CN106228190A (en) * | 2016-07-26 | 2016-12-14 | 广州华工弈高科技有限公司 | Decision tree method of discrimination for resident's exception water |
US20180197048A1 (en) * | 2017-01-11 | 2018-07-12 | Ford Global Technologies, Llc | Generating Training Data for Automatic Vehicle Leak Detection |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11209333B2 (en) | 2015-01-07 | 2021-12-28 | Homeserve Plc | Flow detection device |
GB2595660A (en) * | 2020-06-01 | 2021-12-08 | Homeserve Plc | Flow event classification |
WO2021244799A1 (en) * | 2020-06-01 | 2021-12-09 | Homeserve Plc | Flow event classification |
Also Published As
Publication number | Publication date |
---|---|
GB201813395D0 (en) | 2018-10-03 |
GB2576358B (en) | 2022-12-28 |
WO2020035696A1 (en) | 2020-02-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108027594B (en) | Method for detecting anomalies in a water distribution system | |
US11449013B2 (en) | Linepack delay measurement in fluid delivery pipeline | |
US10387590B2 (en) | Techniques for iterative reduction of uncertainty in water distribution networks | |
US10401879B2 (en) | Topological connectivity and relative distances from temporal sensor measurements of physical delivery system | |
TWI391803B (en) | System and method for monitoring resources in a water utility network | |
Jung et al. | Water distribution system burst detection using a nonlinear Kalman filter | |
US10354195B2 (en) | Forecasting leaks in pipeline network | |
US9395262B1 (en) | Detecting small leaks in pipeline network | |
US20210116076A1 (en) | Anomaly detection in pipelines and flowlines | |
Eliades et al. | Leakage fault detection in district metered areas of water distribution systems | |
KR20190131046A (en) | Device motion signal processing system and method | |
US20140149054A1 (en) | Leak Detection Via a Stochastic Mass Balance | |
KR102435919B1 (en) | A in-house leakage detection and type classification device using multidimensional data, a method and a program thereof | |
JP6164311B1 (en) | Information processing apparatus, information processing method, and program | |
WO2020035696A1 (en) | Sensing fluid flow for estimating fluid flow state | |
BR112021003369A2 (en) | method for identifying the source of a pressure transient in a water network | |
WO2020153903A1 (en) | Method and system for monitoring a gas distribution network operating at low pressure | |
US20210172824A1 (en) | Sensing fluid flow | |
US20160275407A1 (en) | Diagnostic device, estimation method, non-transitory computer readable medium, and diagnostic system | |
US11391615B2 (en) | Fluid flow probe and analysis | |
Boudhaouia et al. | Water consumption analysis for real-time leakage detection in the context of a smart tertiary building | |
US11635342B2 (en) | Building type classification | |
KR20160026492A (en) | Anomaly detection method using confidence interval estimation based on time series analysis | |
US9699525B2 (en) | Method and system for automatic residual consumption | |
Kim et al. | Robust leakage detection and interval estimation of location in water distribution network |