Patent Application
20220333823
This invention relates, but is not limited to, a method, a device, a computer program product and/or apparatus for diagnostics or fault determination of the operation of a boiler.
Boilers are used to provide heated water in homes and other buildings. The heated water is commonly provided for domestic usage, such as shower and kitchen water as well as being used to provide heating, for example through radiators. Boilers commonly use gas or electricity as a heating source and their usage can form a large portion of a domestic energy usage. There are different types of boilers. A combination or ‘combi’ boiler is both a water heater and a central heating boiler in a single compact unit. Combi boilers heat water directly from the mains when activated. This removes the need for a hot water storage cylinder or a cold water storage tank.
Although the basic operation of boilers is widely known, boilers are becoming increasingly complex and may have multiple ways of operating, or operation modes, which can operate dependent on the demand for heated water, instructions from a controller or the settings of the boiler. In modern boilers, the mode of operation of the boiler and any operation faults may be available through a control panel. However, users are often not aware of the mode or fault because they do not closely monitor their boiler and a faulty boiler may still produce hot water for some time period. The reporting or raising of a fault is commonly different depending on the boiler, because there is a large range of different models and manufacturers. Furthermore, proprietary systems or codes may be used which limit the ability of a user to identify or interpret the mode or fault of the boiler.
US20140222366 shows an apparatus and method for monitoring and analysing the performance of a heating and cooling system. It shows the use of temperature sensors on the inflow and return temperatures on the pipes of a heat plant or boiler, which are described as indicating the relative performance of a boiler, or if the radiators are balanced. This relies on a comparison between the two temperatures.
EP3117566 shows a method for determining a state of operation of a domestic appliance in which time series data is obtained over a cycle of operation of the domestic appliance. The state of operation is then determined by comparing this time series data to an expected model of time series data of a plurality of domestic appliances. This system determines a current operation of the boiler, such as whether the boiler is in a burn state, based on pattern matching to particular boiler action. A current operation describes what the boiler is doing at each time (e.g. ignition flame), rather than a mode of operation which the boiler may be in for a period of time.
U.S. Pat. No. 9,638,436 shows a method of monitoring HVAC performance remotely by measuring an aggregate current supplied to the HVAC system and determining a failure based on the current data. This uses careful analysis of aggregate current readings to determine and assess operation of individual components of the HVAC systems. Disaggregation of energy signals is often complex and may result in situations where it is not possible to identify particular components.
Embodiments of the invention seek to provide approaches for allowing detection of faults or boiler failures.
Accordingly, in a first aspect of the invention, this is provided a method of building a model for detecting the fault status of boilers, the method comprising the steps of:
Corresponding time series means that the time series cover the same, or a subset of the same time periods. This provides an association between the electrical energy data and the outlet temperature data, so that the outlet temperature data can be used to identify key periods in the electrical energy data where clear fault status decisions can be made. The method then selects a time window in the temperature data and identifying cold flow events in this time period, wherein the same or preferably an overlapping (in whole or part) time window is selected in the electrical energy data. The model is trained on the basis of fault status in each of the electrical energy data time windows. This provides a model which links the fault status of the boiler (e.g. faulty, pre-fault or normal) to the electrical energy data in a time window around a cold flow event. This provides a more accurate measure of the fault status that the electrical energy data alone because it limits the potentially complex electrical energy signals that need to be processed.
Preferably the steps of identifying the fault status and identifying the one or more cold flow events comprise at least one of:
Alternatively the steps of identifying the fault status and identifying the one or more cold flow events comprises at least one of:
The fault status of the training data may be applied by a skilled human operator to allow the automated learning of the model, for instance by reviewing the time series information in the training data and identifying times faulty behaviour associated with cold flow events. It is also possible that this information would be associated with the time series data, for instance by being collected from the boiler controller when collecting the data, available through a smart controller, or otherwise obtained.
Preferably the time window of the electrical energy data includes the time window of the cold flow event. More preferably the time window of the electrical energy data is larger than the time window of the cold flow event. More preferably the time window of the electrical energy data and the cold flow event have the same end time.
Preferably the model comprises a decision tree. Preferably the model comprises a measure of how constant or stable the electrical energy data is during the time period.
Preferably the model is configured to determine the presence of a fault based on the time series of electrical energy data preceding a cold flow event and the time series of outlet temperature data, wherein the two time series are combined in a mathematical function. More preferably the mathematical function is a product.
Preferably the model comprises a measure of the relationship between the electrical energy data and the outlet temperature data.
Preferably the model comprises an adjustable confidence threshold. This allows a constructor or user of the model to adjust the accuracy of the system, for instance to reduce the occurrence of false alarms.
Preferably the training data sets are obtained from a plurality of further boilers in a plurality of different locations. Preferably the further boiler information comprises multiple days of use of the plurality of further boilers. The use a wide variety of further boilers over multiple provides a broad training data set for the system and may improve performance. Preferably the training set comprises at least 50, 100 or 200 further boilers and/or time series.
Accordingly, in a second aspect of the invention, there is provided a method for determining a fault status of a boiler, the method comprising the steps of:
The determination of a cold flow event is used to identify periods where clear identification of fault status can be achieved in the electrical energy data. This is because the electrical energy data is often complex and difficult to effectively disaggregate due to overlap between faulty and normal operation of boilers at different time or operation modes. Cold flow events occur when cold water flows from the boiler output pipe. Cold flow events can, for instance, be caused by a failure in the heating element of the boiler, or a pressurization problem, or may occur as a boiler is beginning operation. A time window of electrical energy data, ending at the point where a cold flow event has been determined is then processed to determine if the boiler is faulty. This restricts the amount of electrical energy data which must be processed to identify the fault. Advantageously the use of cold flow events to indicate time windows for fault status identification in the electrical energy data provides a much narrower range of expected electrical energy data, allowing for the construction and use of a more accurate model. This is because cold flow events are both reasonably frequent and are often associated with a clear boiler activation signals, or attempted signals, in the electrical energy data.
Preferably the model is created according to the first aspect.
A cold flow event occurs where there is a decrease in the outlet pipe temperature. This occurs when water (which may be ambient water temperature, or temperature beneath a base or expected temperature in the outlet pipe) leaves the boiler before it has been heated. It commonly occurs when the boiler is initially activated to deliver hot water (for instance a hot water faucet being opened) because there is a delay in the activation of the boiler element. This means that the temperature signal indicates that the outlet pipe is unusually cold, generally because water has passed through the boiler without heating, or without heating as much as usual.
Preferably the cold flow event is determined through a comparison of the shape of the temperature data with an expected shape function. Preferably the comparison comprises a classification. More preferably the classification uses the KNN (K-nearest neighbours) algorithm or a Random Forests algorithm. The expected shape function may comprise a cold stream range, wherein a cold flow event is determined when the temperature data remains within the cold stream range.
Preferably the method comprises the step of normalising the temperature data. More preferably it is normalised to the average temperature of the data. Normalising the temperature data reduces the impact of different starting temperatures across the temperature data sets, this enables an improved comparison between temperature data, for instance to determine cold flow events.
Preferably the method comprises the step of selecting or obtaining a time window of temperature data. Preferably the time window is a rolling time window, such that a constant length of temperature data is provided. Preferably the time window is less than 10 minutes, more preferably less than 8 minutes and most preferably 5 minutes.
Alternatively the cold flow event is determined by a measure of the gradient of the temperature data. Preferably the cold flow event is determined by a comparison of the temperature data to a reference temperature, such as a base temperature. The base temperature may be a pre-set value, or may be estimated from historical temperature data. More preferably the cold flow event is determined by a combination of the gradient of the temperature data and the comparison of the temperature data to a reference temperature. Preferably the cold flow event is determined by repeated application of the comparison(s).
This alternative method (although it may be used in conjunction with the earlier method) of determining a cold flow event determines a gradient (or rate of change) of the temperature data. This is effective because a negative gradient indicates that the water in the pipe is getting colder and a large negative gradient suggests this is not simply due to cooling of the pipe. A comparison to a reference temperature, such as the ambient or a defined base level temperature, improves performance because it ensure that the decrease in temperature is from a relatively low temperature, removing false alarms due to high rates of cooling (large gradients) due to high temperature cooling.
Preferably the method comprises a first stage to determine the cold flow event and a second stage to determine the fault status, wherein the second stage is only activated or performed when a cold flow event is detected.
Through the use of a first stage in which a cold flow event is identified and a second stage in which the electrical energy is examined for a time window prior to the cold flow event occurring the processing requirements can be reduced, because the electrical energy is only processed when a fault can be detected.
Preferably the electrical energy time window is less than 20 minutes, more preferably less than 10 minutes, more preferably 3-8 minutes, most preferably 8 minutes.
Preferably the electrical energy data comprises current data and/or voltage data.
Preferably the model comprises a decision tree. Preferably the model calculates statistical features of the electrical energy time series. Preferably the statistical features comprise any one or more of the sum, mean, variance, skewness, or other statistical moments. More preferably the statistical features comprise the sum and the first five statistical moments. Preferably the decision tree uses the statistical features and the electrical energy time series.
Preferably the model calculates a measure combining the temperature time series and electrical energy time series data. Preferably the combination is a mathematical function of the temperature and electrical energy time series data, where the product or sum is calculated for each matching time point in the temperature and electrical energy time series data. Preferably the mathematical function is a product or sum.
Preferably the boiler is a combination boiler (or combi-boiler). A combi boiler provides heat on demand by a combination of a water heater and central heating boiler in one unit. This reduces or removes the need for a hot water storage tank.
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 system, apparatus and computer program aspects, and vice versa.
Furthermore, 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.
Preferred features of the present invention will now be described, purely by way of example, with reference to the accompanying drawings, in which:
The monitoring system of
The sensors are preferably able to communicate with a controller 102. The controller 102 may be a microcontroller or other control device. It is preferably connected to the sensors by a connection means 113 which may be wired (e.g. a wire) or wireless (e.g. Bluetooth™ or Zigbee™) for each sensor. In many cases it is preferably wired to allow the controller to power the sensors. The controller 102 is preferably linked to a server 101 (either directly or through one or more intermediate devices), with which it may communicate wired or wirelessly. The server may also be in communication with a user device 103 this is typically using the internet, although it is also possible a user device would connect on a local area network (including to the controller or an intermediary device such as a hub). The user device 103, (for instance a personal electronic device such as a mobile telephone or smart, or other user interactive device) may allow a user to interact with the controller and/or to send instructions to the controller and/or to receive notifications from the controller, for instance regarding a diagnostic problem or a determined mode of operation. The notification may be by text message or notification in an application on the user device 103. The notification may prompt the user to take an action to control the boiler, or may inform the user of an action already taken by a boiler controller. Alternatively the server may control, or send instructions to, the boiler directly based on a fault determination.
The server will typically include other conventional hardware and software components as known to those skilled in the art. While a specific architecture is shown by way of example and specific software technologies and vendors have been mentioned, any appropriate hardware/software architecture may be employed. Functional components indicated as separate may be combined and vice versa. For example, the functions of server 101 may in practice be implemented by multiple separate server devices, e.g. by a cluster of servers.
The method first processes the temperature data 302. this processing of the temperature data may include a normalisation to make comparison with historical data easier and more accurate, as well as a windowing. Windowing the data (selecting a specified time length of data within the data received from the boiler) provides a fixed length of data which allows easier comparison with historical or comparative data. Preferably the windowing is achieved by building a vector each time a new temperature sample arrives, wherein the new temperature data is added to the vector and the oldest temperature data is removed, maintaining the fixed length of the temperature data.
The method then determines if a cold stream event 304 has occurred in the temperature data. This requires processing the temperature data to identify a cold stream event, methods may include shape matching as described below in relation to
The processing of the electrical data preferably includes a windowing similar to the temperature data. However, the selected window is preferably longer than the temperature window, such that the electrical data extends further back in time, since relevant electrical activity often precedes any flow of water from the boiler. The reference of the length of time of the window is preferably from the end of the window (e.g. the last temperature reading) as this represents the most up-to-date reading. In a preferred system a five minute temperature window is used and an eight minute current window, where the current window therefore covers a three minute period before the temperature window begins. In other embodiments the overlap between the windows is not complete, for instance the current window may cover a five minute period beginning two minutes before the start of the five minute temperature window and finishing three minutes into the temperature window.
The method then determines if a fault has occurred in the boiler 306, i.e. that the boiler operation is abnormal. This step preferably provides a classifier which classifies the boiler into two or more states or categories including faulty and non-faulty or normal. It may also be able to classify potentially faulty states, such as when performance degradation is found. In a preferred example a decision tree algorithm is used. A decision tree algorithm allows the program to make a fault status decision through a series of decisions steps (e.g. working from the trunk to the branches of the decision tree). In a preferred example the output from the decision tree is expressed as a probability or on a data range. For example, the output may be a number between 0 and 1, where 1 is a definite failure and 0 is normal operation, allowing for an accuracy of specificity of the detection to be controlled. If no fault is determined the process returns to the beginning and waits for further data from the boiler, at which point the process can repeat.
The method is then able to report the failure 307 or raise an alert of the problem. The alert may be an audio/visual alert to the homeowner such as a bell or flashing light. Alternatively, or in combination, the alert may be sent to a device, either directly or over a network. The device may be a personal electronic device such as a mobile phone. In a further embodiment the alert may be sent to a central controller or monitoring system who can contact the boiler owner or dispatch a suitable repair person. The method may also include a direct instruction to the boiler, for instance to pause operation until authorised by a user, or the problem is fixed. This may help reduce boiler failure by stopping operation before further damage occurs.
The decision tree preferably acts on a vector of a window of the current measurements. The decision tree preferably also looks at combinations or mathematical combinations of the current measurements, these may include the statistical moments of the vector, such as the first five statistical moments and/or the sum of the Current measurements, as well as the current measurements themselves. In further embodiments the temperature vector is also used, including the statistical moments of the temperature vector (and preferably also the temperature vector itself). In a preferred example a combined vector is formed, for instance by combining the temperature and electrical data. In a preferred embodiment this is formed by the product of the temperature and time (preferably normalised), where the product is of the temperature and time values at each common time step.
In a particular embodiment the window of current measurements is represented as the vector C=(Ck, Ck-1, . . . , Ck-N), where ck is the current measured at the timestamp k and the window covers a fixed timespan LC. A second vector which represents the response of the temperature to the current consumption is represented as R=(Cktk, Ck_1tk-1, . . . , Ck-Ntk-H), where H=min(N,M). The goal of the detector is to check if C corresponds to a failure or not. The training data for the decision tree is formed from the set {(Ci,Ri,yi)|yi∈{0, 1}} where yi is set to 1 where the window corresponds to a failure and to 0 otherwise is collected. This allows a statistical classifier (e.g. the decision tree) to be trained to predict yi. The features of C and R include the first 5 statistical moments and the sum of C and all the elements of the vector R.
In a specific example a temperature window is represented as the vector T=(tk, tk-1, . . . , tk-N), where tk is the temperature at the timestamp k and the window covers a fixed timespan LT. The training or historic data is formed by a set of previous or known examples of cold water events and non cold water events (or normal operation), such that the data provides a set D={(Ti, yi)|yi∈{0, 1}} where y is set to 1 where the window corresponds to a certain cold water event and 0 for a certain not cold water event. A statistical classifier W(Ti) is then trained to predict yi. The temperature values are used as predictive features of a process of resampling and normalization. Resampling allows sensor data to be made comparable, for instance by insuring they have the same number of samples, or similarly spaced samples. This addresses problems due to sensors which report on change, rather than sending results at fixed times. Results can, for instance, be achieved with Random Forests trained using Gradient Boosting and/or K-Nearest Neighbours (KNN) and other classification systems may be known. In use the vector T can be reconstructed each time step to add the new temperature value and discard the oldest temperature value. Using the KNN approach T is then compared with each of the historical/annotated vectors in D using a metric distance (for instance the Euclidean distance). If the most similar vector (or closest vector) is a failure then the KNN approach will classify the time window as a failure, otherwise it will move on to the next time point.
representing a fraction or multiple (depending on alpha) of the minimum temperature in an expected or previous window. The window for calculating the base temperature may be longer than the comparison window, for instance a 24-hour period. Alternatively, the base temperature may be a predefined value, or a value calculated based on further temperature readings. A cooling rate c can be calculated as a function of the difference between consecutive temperature values: Tid-1−Ti-1d-1. Other methods of calculating the cooling rate or gradient are also available. A cooling event is determined where Tid<b and Tid−Ti-1d<c for parameters b and c, preferably for w consecutive timestamps.
Any of the described methods, processes and techniques may be performed by software code executing on processor(s) of one or more processing devices (such as the controller 102 and/or server 101 of
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.
The above embodiments and examples are to be understood as illustrative examples. Further embodiments, aspects or examples are envisaged. It is to be understood that any feature described in relation to any one embodiment, aspect or example may be used alone, or in combination with other features described, and may also be used in combination with one or more features of any other of the embodiments, aspects or examples, or any combination of any other of the embodiments, aspects or examples.
Furthermore, equivalents and modifications not described above may also be employed without departing from the scope of the invention, which is defined in the accompanying claims.
Further aspects and features of the invention are set out in the following numbered clauses.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2105430.9 | Apr 2021 | GB | national |
