Sensor devices, system and method for determining fluid flow through a pipeline

Abstract

A sensor device (1) for a pipeline carrying a fluid, in particular for leak detection, comprises a housing (7) for arrangement on the pipeline (6), a first temperature sensor (3), a second temperature sensor (2) and an evaluation device (4), wherein the temperature sensors (2, 3) are arranged in and / or on the housing (7) in such a way that, when the housing is fastened to the pipeline (6), the first temperature sensor (3) is at a greater distance from the pipeline than the second temperature sensor (2), wherein the evaluation device (4) is designed to a) process first sensor data from the first temperature sensor (3) and second sensor data from the second temperature sensor (2) as a local computing unit and / or b) transmit the first and second sensor data to a remote computing unit for processing in order to determine a fluid flow through the pipeline (6).

Description

The invention relates to a sensor device for a pipe line carrying a fluid, in particular a sensor device for detecting leaks.The invention further relates to a computing unit, in particular for processing sensor data of such a sensor device.Furthermore, the invention relates to a system having such a sensor device and a remote computing unit.The invention further relates to a method for determining a fluid flow through a pipeline, in particular for leak detection, preferably using such a sensor device.Sensor devices for detecting leaks in pipelines are known from practice. For example, there are water meters which can also be used for detecting leaks. However, water meters installed on the main water pipe can generally detect water leaks only when the flow of water through the leak is sufficiently large. The conventional models available on the market detect water streams only from a flow rate of 50 liters per hour. The same applies to water meters in homes. Other devices for detecting leaks are likewise integrated into the line system, as a result of which retrofitting is made more difficult.Devices for detecting leaks are also known, which are designed independently of a water meter or the water circuit. For example, EP 3 254 076 A1 discloses a concept for analyzing a water flow. An analysis unit comprises a housing with electronic components. Led out of the housing are one or more temperature sensors which can be attached, for example, to different sections of a domestic water line, for example before and after a faucet. Such an analyzer is bulky and requires a remote power source. In addition, leakage detection takes place directly in the device, which limits the processor power available. The use is possible only in single-family houses. The technique relies on equilibrium states at the temperatures measured.Furthermore, systems for detecting leaks are known in advance from EP 3 479 079 B1, WO 2016 / 146500 A1, EP 3 243 080 B1, DE 10 2021 211 940 A1 and WO 2019 / 08689 A2.The object of the present invention is to provide a sensor device for detecting leaks, which can also be used in homes or multi-family homes in order to detect leaks. Furthermore, the object is to make it possible to detect leakage even at very low water flows. Furthermore, a computing unit, a system and a method for detecting leaks should be specified.According to the invention, the above object is achieved by the features of claim 1. Claim 1 relates to a sensor device for a pipe conducting a fluid, in particular for detecting leaks. The sensor device comprises a housing for arrangement on the pipeline. The sensor device comprises a first temperature sensor, a second temperature sensor and an evaluation device. The temperature sensors are arranged in and / or on the housing in such a way that, in the state of the housing fastened to the pipeline, the first temperature sensor is at a greater distance from the pipeline than the second temperature sensor. The evaluation device is designed to process a) as a local computing unit first sensor data of the first temperature sensor and second sensor data of the second temperature sensor and / or to transmit b) the first and second sensor data to a remote computing unit for processing in order to determine a fluid flow through the pipeline.In a manner according to the invention, it has first been recognized that the flow of a fluid, and in particular of a liquid, through a pipeline can be recognized in a simple manner in that the temperature of the pipeline changes when the fluid flows through. Although the flow rate of the fluid also affects the ambient temperature, this temperature change is substantially less and is only noticeable at the time greater amounts of fluid flow through the tubing. When the flow of the fluid is terminated, the ambient temperature returns to the original value within a short time, while the temperature of the piping requires a longer time for it. This makes it possible to make a prediction, based on the temperature change in the ambient temperature, of how the temperature of the pipeline would have to change if the fluid flow is completely blocked. If the temperature of the pipeline does not correspond to this prediction, it can be assumed that there is still a low fluid flow which has an effect on the temperature of the pipeline. This low fluid flow can be detected even at low fluid flows of a few milliliters of liquid per minute and a small distance between the two temperature measurement points, as long as one of the measurements is carried out closer to the pipeline, preferably with a direct thermal coupling to the pipeline, and the other measurement is carried out further away from the pipeline. This allows the sensor device with temperature sensors to be arranged within a compact housing. Such a compact sensor device can be arranged at any point of a pipeline system, i.e. also within a home or on a pipeline which is divided by a plurality of consumers in a multifamily home. By detecting leaks based on the development of the temperature over time, it is possible to use them in particular also in complex topologies with many consumers in which an equilibrium between the temperatures occurs only rarely or never.The evaluation device can advantageously comprise a communication device, preferably configured for wireless communication, or be coupled to such a communication device. The sensor data can preferably be transmitted to the remote computing unit via such a communication device, wherein the communication device is configured to transmit the sensor data to the remote computing unit. By transmitting the sensor data for processing to the remote computing unit, the sensor data can be processed by a remote computing unit with a high computing power. This is more efficient because the high computing power can be shared by different sensor devices. Moreover, more complex calculations and, in particular, the use of complex machine learning models are possible, which improves the accuracy of determining the fluid flow. In addition, the sensor device is relieved of load, so that the sensor device can be operated with a battery over a longer time. As a result, the sensor device can also be used at locations where a power connection is not available.Alternatively, the results of the detection of the fluid flow can be transmitted by the local computing unit by means of the communication device to a remote device, such as a server or a mobile device. As a result, a user can be informed of a leak independently of location.The sensor device is provided to be fastened to the pipeline. For this purpose, the housing can preferably comprise a fastening device with at least one, in particular two, preferably slot-like, opening in order to fasten the housing to the pipeline by means of one or more fastening elements, preferably cable ties. The openings in the housing allow the sensor device to be fastened without clamps or similar components, which have to be adapted to the thickness of the pipeline. This simplifies the use of the sensor device with pipelines of different thickness.Preferably, the at least one opening, preferably the two openings, defines or define a fastening region of the housing. This fastening region is in contact with the pipeline in the state of the housing arranged on the pipeline. The second temperature sensor may be arranged inside the fastening region and the first temperature sensor may be arranged outside the fastening region of the housing. The fastening region is distinguished in particular by a stronger thermal coupling to the pipeline, which is not present at the first temperature sensor, so that changes in the pipe temperature have an increased effect on the measurements of the second temperature sensor and, if at all, only slightly or negligibly on the measurements of the first temperature sensor.The thermal coupling between the pipeline and the second temperature sensor can be improved by using heat-conducting elements. The heat-conducting element, preferably a heat-conducting pad, can advantageously be arranged on the second temperature sensor in order to improve the thermal coupling to the pipeline.For a compact and favorable production of the sensor device, the temperature sensors and the evaluation device can be arranged on a printed circuit board. This enables the sensor device to be mounted by means of automatic placement machines, as a result of which the production costs can be reduced.While the use of a common printed circuit board for the temperature sensors and the evaluation device offers advantages with regard to production, such a printed circuit board also acts as a heat conductor which thermally couples the two temperature sensors to one another and to the evaluation device. In order to reduce the effects of this thermal coupling, recesses can be provided in the printed circuit board in order to thermally decouple the respective component from one another. Thus, the printed circuit board can have a main surface on which the evaluation device is arranged, and two secondary surfaces on each of which a temperature sensor is arranged. The printed circuit board can have a recess in each case between the secondary surfaces and the main surface. In particular, the secondary surfaces can be connected to the main surface only by webs. As a result, the thermal coupling between the components can be reduced, whereby the accuracy of the determination of the fluid flow can be improved.In order to further improve the thermal coupling, a planar metallization can be provided on the respective rear side of the printed circuit board. In other words, at least one of the temperature sensors can be arranged on a first side of the printed circuit board. On a second side opposite the first side (i.e. the rear side, relative to the front side on which the respective temperature sensor is arranged), the printed circuit board can have at least one planar metallization which corresponds from its lateral position to a lateral position of the at least one temperature sensor. As a result, the thermal coupling to the pipeline or to the environment (in the case of the first temperature sensor) can be improved. In particular, the heat conducting device can be attached to the flat metallization.In a manner according to the invention, the evaluation device and the temperature sensors are arranged such that a lateral distance between the evaluation device and the temperature sensors corresponds in each case to at least 50% of a greatest lateral extent of the housing. As a result, a thermal influence of the evaluation device on the temperature measurements can be reduced, as a result of which the accuracy of the determination of the fluid flow can be improved.Advantageously, the sensor device can be designed as self-sufficient as possible. In particular, the sensor device can be supplied with energy by means of a battery, so that it can also be mounted at locations at which supply via a power grid is not possible. For example, the sensor device can comprise a circuit for connecting a battery, wherein the circuit is designed to provide electrical energy of the battery for the temperature sensors and the evaluation device. The battery, which may be a disposable battery or a rechargeable battery, may also be arranged in the housing. In particular, the sensor device can further comprise the battery. This makes the sensor device independent of a remote power supply, so that it can also be mounted in locations that are difficult to access.A further aspect of the present invention relates to the computing unit which can be used to process the sensor data of the temperature sensors. With respect to the computing unit, the object is achieved by the features of claim 9. The computing unit comprises an interface for obtaining first sensor data and second sensor data from the sensor device according to the invention. The first sensor data represent temperature measurements at a first position in and / or on a housing with a greater distance from a pipeline, in particular an ambient temperature. The second sensor data represent temperature measurements at a second position in and / or on the housing with a smaller distance from the pipeline, in particular a surface temperature of the pipeline. The computing unit comprises a processor configured to determine a fluid flow through the pipeline based on the first sensor data and the second sensor data. As explained above, this makes it possible to determine a fluid flow even with low flow quantities of a few milliliters per minute and also to detect leaks even in complex topologies with many consumers in which an equilibrium between measured temperatures occurs only rarely or never. For example, the processor may be configured to determine a leak based on a detected fluid flow through the pipeline.In particular, the processor can be configured to determine a fluid flow through the pipeline based on a temporal profile of the temperature measurements which are contained at least in the first sensor data and based on the second sensor data. Based on the time profile of the temperature measurements of the ambient temperature, the tube temperature can be predicted. If the pipe temperature does not match the prediction, an abnormality can be assumed, which may be caused by a leak, for example. By taking into account the profile of the temperature measurements, and not exclusively the setting of an equilibrium between the temperatures, the fluid flow, such as a leakage, can be established even in complex topologies with many consumers in which an equilibrium between measured temperatures is set only rarely or never.The processor can advantageously be designed to determine the fluid flow through the pipeline on the basis of temperature slopes or temperature slopes within the time profile of the temperature measurements. In particular, the steepness of the temperature slopes or temperature gradients can be taken into account. Thus, not only (only) the pure temperature differences have an influence on the determination of the fluid flow, but also temperature changes, their speed (for instance as a first derivative of the temperature changes) or their accelerations (for instance as a second derivative of the temperature changes). By taking into account the temperature slopes and temperature gradients, a fluid flow can be established already early, i.e. before an equilibrium is established.The processor can preferably be designed to determine a predicted time profile of the temperature measurements at the second position on the basis of a time profile of the temperature measurements at the first position. The processor may be configured to compare the predicted time profile of the temperature measurements at the second position with at least one temperature measurement at the second position to determine a fluid flow through the conduit. For example, a fluid flow, such as caused by a leak, may be assumed when the difference between the predicted time profile of the temperature measurements at the second position and the measured temperatures at the second position indicates that the measured temperature is increasing (or decreasing) less rapidly than the predicted temperature. In other words, the processor may be configured to compare a (positive or negative) slope of the predicted temporal profile of the temperature measurements at the second position with a (positive or negative) slope of a measured temporal profile of the temperature measurements at the second position to determine a fluid flow through the conduit (wherein there is a fluid flow / leakage if the slope of the prediction is greater than the slope of the measurement). The determination of the fluid flow can thus be based on checking whether the tube temperature behaves as expected. If the temperature profile does not behave as expected, there may be a leak.In a particularly advantageous manner, a prediction of the time profile of the temperature measurements at the second position can be achieved by means of a machine learning model. Machine learning is a sub-area of artificial intelligence that includes the development of algorithms and models that allow computers to learn and make predictions or decisions without being expressly programmed. The focus is on the development of systems that can improve their performance over time by learning from data.Training a machine learning model refers to the process of making the model accurate predictions or decisions. During training, the model is faced with a large amount of data used to adjust the internal parameters or weights of the model. The model learns from the training data patterns, relationships or rules that allow it to generalize and make predictions for new unseen data.Training data is a number of examples or instances used to train a machine learning model. Often, it is marked data, i.e., each example is associated with a known result or target value. The training data consists of input features as well as the corresponding output or target variable. The model learns from this data by analyzing the patterns and relationships between the input features and the target variable. For training the machine learning model, training algorithms such as supervised learning, semi-supervised learning, unsupervised learning, or reinforcement learning may be used.Machine learning models, such as the machine learning models that can be used within the scope of the invention, are frequently implemented as artificial neural networks (ANNs), in particular as deep neural networks, or as support vector machines, decision tree models or random forest models.The processor may be configured, for example, to determine the predicted time profile of the temperature measurements at the second position by means of a prediction machine learning model. In particular, the prediction machine learning model may be an RNN (a recurrent neural network, such as a simple recurrent neural network, an RNN with a so-called long short-term memory (long short-term memory), or an RNN with a gated recurrent unit (gated unit)), or a convolutional neural network (convolutional neural network), or a feedforward neural network (FNN).Machine learning models can be adapted to the installation situation and the resulting temperature changes of the respective sensor device from which the sensor data originate in order to improve the prediction accuracy. This can be done by iteratively adapting the prediction machine learning model by means of training. The processor may be configured to adapt the prediction machine learning model based on the first and second sensor data, preferably by supervised learning ("supervised learning"). In the present case, the prediction machine learning model may be used to predict the time profile of the temperature measurements at the second position based on the time profile of the temperature measurements at the first position. In this case, the time profile of the temperature measurements at the first position can be used as input data and the time profile of the temperature measurements at the second position can be output by the model. In order to adapt the prediction machine learning model to the respective installation situation, the parameters (also called weights) of the prediction machine learning model can now be adapted by means of a suitable mathematical method, such as by means of a gradient method (gradient distance) until the predicted time profile of the temperature measurements at the second position when the time profile of the temperature measurements at the first position is input corresponds as exactly as possible to the actually measured time profile of the temperature measurements at the second position (i.e. as long as a certain improvement of the prediction is to be determined). In this case, the actually measured temporal profile of the temperature measurements at the second position can be used as desired output, and the appropriately measured temporal profile of the temperature measurements at the first position can be used as input of the supervised learning algorithm in order to adapt the prediction machine learning model. However, the initial stage is an already pre-trained machine learning model that has been pre-trained based on temperature profiles in various installation situations to ensure the general functionality of the predictive machine learning model. The adaptation serves to adapt the prediction for the temperature profile of a sensor device. The machine learning model already works without later adaptation to the installation situation not with fixed (defined) threshold values but with dynamic (learned) threshold values. In each house, or installation, these may be different and may be learned by the adaptation of the machine learning model. As such, the adaptation of the machine learning model may also be referred to as calibration, although ideally not a one-time calibration. This form of calibration may be repeated at multiple stages (intervals) as the model continues to learn by including new data in the learning.In some cases, particularly when the computing unit is a remote computing unit for processing sensor data from different sensor devices, such adaptation of the prediction machine learning model means that a separate prediction machine learning model is used for each sensor device. As a result, assuming that the sensor data is received at regular intervals, a plurality of prediction machine learning models must be simultaneously held in the memory of the computing unit or loaded into the memory, which makes a great demand on the amount of available memory and the loading architecture. To reduce this effort, the predictive machine learning model may include two components - a first basic model that is used for all predictions and is not changed during the adjustment, and a downstream refinement model that is adjusted to the sensor data of a particular sensor device. Here, the basic model, assuming the prediction machine learning model is used as a deep neural network (DNN, a neural network having one or more hidden layers in addition to an input layer and an output layer) may include a larger number of layers (such as the input layer and one or more hidden layers) than the refinement model (such as the output layer or one or more hidden layers and the output layer). These two models can then be interconnected to one another in order to output the prediction on the basis of the temporal profile of the temperature measurements at the first position.The output of the predictive machine learning model is the predicted time course of the temperature measurements at the second position. This can be achieved by the prediction machine learning model being a time-series prediction machine learning model and / or a regression machine learning model. In both cases, based on a series of previous temperature measurements (at the first position), the history of the temperatures at the second position is predicted, such as for a single time step (which is then combined with the predictions for other time steps to obtain the temperature history) or equally for a sequence of time steps.Alternatively or additionally to predicting the course, the determination as to whether a fluid flow (and thus possibly also a leakage) is present can be considered as a classification problem. Thus, the processor may be configured to determine a fluid flow through the pipeline, and in particular the presence of a leak, using a classification machine learning model based on the first and second sensor data. In particular, the classification machine learning model can be trained to output a classification about a fluid flow through the pipeline based on a temporal profile of the temperature measurements that are contained at least in the first sensor data and based on the second sensor data (for example a current value of the temperature measurement or a profile of the temperature measurements at the second position). For this purpose, the classification machine learning model can be trained by means of supervised learning in order to output a classification as to whether a fluid flow, such as a leakage, is present, based on first and second sensor data of sensor devices and corresponding labels, whether a fluid flow / a leakage is based on the respective temperature profiles.In some embodiments of the invention, the determination of the fluid flow is based on a course of the respective temperatures. These curves are provided as input to the model used in each case. In this case, the respective model can act stateless in a first variant, i.e. inputs made in the past into the model have no influence on a current output of the model, the output merely depending on the current input and the model parameters. In this case, the processor may be configured to provide the respective machine learning model with the time profile of the temperature measurements that are contained at least in the first sensor data (for example a section, a so-called window, thereof) as current input data. This procedure is particularly appropriate if the computing unit is a remote computing unit which processes the sensor data of many sensor devices, since the sensor data can be distributed without problems to different remote computing units in this case by means of a load distribution.Alternatively, the respective model can memorize the profile, so that only the temperature measurement value or the temperature measurement values of a time step are used as input in each case. In this case, the respective machine learning model can comprise a memory component, preferably a recurrent neural network, for example a long short-term memory, in order to keep the time profile of the temperature measurements which are contained at least in the first sensor data in memory. Recurrent neural networks, such as long short term memories or gated recurrent units, have mechanisms for weighting previous inputs and, in abstract form, using them as an additional internal state that influences the current output.One way to determine whether there is a flow of fluid through a conduit is to determine a state of equilibrium between different temperature measurements. While the proposed concept is based in particular on determining a fluid flow, such as a leak, on the basis of the temperature changes at both positions in or on the housing, the determination of the fluid flow can (additionally) also be based on the determination of an equilibrium state (equilibrium). In this case, in particular a virtual equilibrium can be defined or learned. This is relevant for a water pipe system which does not have absolute equilibrium states. For example, a permanent (known / accepted) water flow can lead to a "virtual equilibrium" already being reached if the temperature difference has a constant (for example) 1.2° C. This is also part of the learning of the model. For example, in an ideal house, the equilibrium may be defined with a 0° C. temperature difference, in a house that has a certain accepted water flow and in which there is always a minimum temperature difference of 1.2° C., but a "virtual equilibrium" equivalent to a real equilibrium may be defined. The processor can be configured, for example, to determine a constant temperature difference between the measured temperatures at the first position and at the second position based on a time profile of the temperature measurements (which temperature difference can also be 0° C. in an ideal house, or can be different from 0° C. by at least 0.1° C. (or at least 0.2° C., or at least 0.5° C.). Based on this constant temperature difference and based on current temperature measurements, a "virtual" equilibrium state can now be determined. A fluid flow through the tubing can now be determined based on whether such a "virtual" equilibrium state exists.In particular, if such an equilibrium state does not occur, a fluid flow, and thus a leakage, can be present.Up to now, the determination of the fluid flow has focused on finding out whether a fluid flow is currently present. Such a fluid flow can indicate, for example, a leakage if it continues for a longer time. In addition, to determine when there is fluid flow, the amount of fluid flow may also be estimated. Thus, the processor can be configured to estimate the volume flow of the fluid flowing and / or flowed through the pipeline based on a temporal profile of the temperature measurements, which are contained at least in the second sensor data, preferably by means of a regression machine learning model. For this purpose, the regression machine learning model can be trained by means of supervised learning in order to output an estimate for the quantity of fluid flow on the basis of the temporal profile of the temperature measurements. This can again be done by using as training input data the measured temporal profile of the temperature measurements at the second position, and optionally the measured temporal profile of the temperature measurements at the first position, and as desired output data the quantity of fluid flow corresponding thereto for training.In some embodiments of the invention, the sensor device can comprise the computing unit as a local computing unit. This enables the local determination of the fluid flow without the need to transmit data. A warning about a leak can be output, for example, via a loudspeaker of the sensor device.In one embodiment according to the invention, a system comprises the sensor device according to the invention and the arithmetic unit according to the invention, wherein the arithmetic unit is a remote arithmetic unit which is arranged separately from the sensor device. This variant enables improved processing of the sensor data, since the remote computing unit can be equipped with more computing power, since it does not rely on battery operation and is shared by many users.A further aspect of the present invention relates to a corresponding method for determining a fluid flow through a pipeline, in particular for leak detection. With respect to the method, the object is achieved by the features of claim 22. The method for determining a fluid flow through a pipeline, in particular for leak detection, using a sensor device according to the invention and a computing unit according to the invention comprises obtaining first sensor data of a first temperature sensor. The method includes obtaining second sensor data of a second temperature sensor, wherein the first temperature sensor is at a greater distance from the piping than the second temperature sensor. The method includes determining a flow of fluid through the conduit based on the first and second sensor data.The method can be implemented, for example, by means of the sensor device and the computing unit. Features which have been described in connection with the sensor device and the computing unit (and a system comprising the sensor device and the computing unit) can also be part of the corresponding method. In particular, the method can comprise that functionality which is provided by the interface and / or the processor of the computing unit.There are now various possibilities for advantageously embodying and developing the teaching of the present invention. For this purpose, reference is made on the one hand to the claims subordinate to claim 1 and on the other hand to the following explanation of preferred exemplary embodiments of the invention on the basis of the drawing. In conjunction with the explanation of the preferred exemplary embodiments of the invention on the basis of the drawing, preferred embodiments and refinements of the teaching are also generally explained. The drawing shows FIG. 1 shows a schematic drawing of a sensor device for determining a fluid flow through a pipeline, in particular for leak detection, which is optionally coupled to a remote computing unit; FIG. 2 shows a schematic drawing of a remote or local computing unit; FIG. 3 is a schematic drawing of a sensor device arranged on a pipeline; FIGS. 4 ato 4 d show diagrams of a relationship between temperature measurements and temperature predictions of a sensor device; FIG. 5 shows a printed circuit board of a sensor device; FIG. 6 shows a sensor device with a printed circuit board and housing; and FIG. 7 shows a flow diagram of a method for determining a fluid flow through a pipeline, in particular for leak detection.The present invention relates to a technology for detecting fluid flow in pipelines, and in particular to detecting tap water leaks.The invention provides a sensor for detecting water leaks and monitoring water consumption in real time. The sensor device used as an Internet of Things (Engl. Internet of things, IoT), can be configured using modern sensor technology and telecommunication hardware to measure and either process water flow data locally or transmit it to a remote server (i.e. a remote computing unit) for processing. The sensor data can be analyzed, in particular on the server, by means of artificial intelligence. The sensor-AI based analysis system by a local or remote computing unit may itself identify leaks with low water flow. Once an abnormality is detected, the system may send a notification to quickly respond.A first component of the invention is a sensor device 1. FIG. 1 shows a schematic drawing of such a sensor device 1 for determining a fluid flow through a pipeline 6 (shown in FIG. 3 ), in particular for leak detection, which is optionally coupled to a remote computing unit 4'. This sensor device 1 comprises two temperature sensors (temperature gauges) 2, 3. one of the two temperature sensors can, as is also shown in FIGS. 3, 5 and 6 in addition to FIG. 1, be positioned horizontally centrally and measure the tube temperature (temperature sensor 2, also referred to below as second temperature sensor) and provide it in the form of second sensor data. The other temperature sensor (temperature sensor 3, also referred to below as first temperature sensor) can, on the other hand, not be positioned centrally and measure the ambient temperature and provide it in the form of first sensor data. In particular, the temperature sensors 2, 3 are arranged in and / or on a housing 7 for arrangement on the pipeline 6 (see FIG. 3 ) in such a way that, in the state of the housing 7 fastened to the pipeline 6, the first temperature sensor 3 is at a greater distance from the pipeline 6 than the second temperature sensor 2.The temperature sensors 2, 3 measure the temperature of the room and an object such as a pipe that changes its temperature when a medium (e.g., water) flows therethrough. The sensor device 1 further comprises a processor and a transmitter unit (both shown as an evaluation device 4 in FIGS. 1, 3, 5 and 6 ). These components 4 are coupled to the temperature sensors 2, 3, process the measurement data (sensor data) and transmit information, for example to the remote computing unit 4'. The sensor device 1 further comprises a power supply. For example, a battery 5 can supply the sensor device 1 with the required energy, which makes it independent of power grids.The analysis of the sensor data is carried out by the evaluation device 4. This is designed to process a) as local computing unit 4 first sensor data of first temperature sensor 3 and second sensor data of second temperature sensor 2, and / or to transmit b) the first and second sensor data to remote computing unit 4' for processing in order to determine a fluid flow through pipeline 6. It is evident that two variants are possible-in a first variant the processing of the sensor data takes place locally. In this variant, the evaluation device 4 comprises a computing unit as a local computing unit. Alternatively (or additionally), the sensor data can be processed in a second variant by the remote computing unit 4', for example by a server in the cloud. This remote computing unit 4' forms a system with the sensor device 1. In order to transmit the sensor data to the remote computing unit 4', the evaluation device 4 comprises a communication device, preferably configured for wireless communication, or is coupled to such a communication device. In the first variant, such a communication device can also be used to transmit information about the determination of the fluid flow to the user, for example via a server or directly to a terminal of the user. The communication device can be designed, for example, for communication via at least one of a local wireless network (WLAN), Bluetooth (a trademark of Bluetooth SIG), ZigBee (a trademark of Connectivity Standards Alliance), DECT (a trademark of European Telecommunications Standards Institute) or another wireless communication protocol, or else via Ethernet.The actual determination of the fluid flow is carried out by a computing unit, whether it be locally or remotely by a server. FIG. 2 shows a schematic drawing of a remote or local computing unit 4, 4'. The computing unit 4, 4' comprises an interface 4a for obtaining first sensor data and second sensor data, preferably from the sensor device 1. Thus, the interface 4 amay be configured to obtain the sensor data via a local data bus (such as I 2 C). In the embodiment as a remote computing unit, the interface 4 acan be an interface for communication via a computer network, for example for communication via the Internet, for example an Ethernet interface. The computing unit 4, 4' further comprises a processor 4b, for instance at least one microprocessor, at least one digital signal processor (DSP), or at least one FPGA (field-programmable gate array). The arithmetic unit 4, 4' further comprises a memory 4c. The memory 4 cmay be volatile or non-volatile memory and may include machine readable instructions executed by the processor 4 bto provide the functionality of the processor 4 b. In addition, the memory 4 cmay include at least one machine learning model that may be used by the processor 4 bto determine the fluid flow. For example, the computing unit 4, 4' may comprise an acceleration circuit, such as a graphics processor (not shown), which is used by the processor 4 bto determine the fluid flow by means of the machine learning model. The processor 4 bis coupled to the interface 4 aand to the memory 4 c. The processor 4b receives the sensor data and processes it to determine the fluid flow. In particular, the processor 4 bis configured to obtain the first and second sensor data via the interface 4 a. The processor 4 bis also configured to determine the fluid flow through the pipeline 6 based on the first sensor data and the second sensor data.The present invention is based on the fact that the sensor data are obtained at different positions in the housing 7, and in particular at different distances from the pipeline 6. Thus, the first sensor data represent temperature measurements (by the first temperature sensor 3) at a first position in and / or on the housing 7 with a greater distance from the pipeline 6, in particular temperature measurements of an ambient temperature. The second sensor data represent temperature measurements (by the second temperature sensor 2) at a second position in and / or on the housing 7 with a smaller distance from the pipeline 6, in particular temperature measurements of a surface temperature of the pipeline 6.FIG. 3 shows a schematic drawing of a sensor device 1 arranged on a pipeline 6. Although a pipeline 6 is always mentioned in the present application, the concept according to the invention is applicable to all objects which change their temperatures when a medium (such as water) flows or is moved through the object. FIG. 3 shows the sensor device 1 with a housing 7 and the arrangement of the components in the housing 7. FIG. 3 in particular shows the horizontally centrally positioned second temperature sensor 2 for measuring the tube temperature and the non-centrally positioned first temperature sensor 3 for measuring the room temperature. The housing 7 includes two slit-like openings (mounting slits) 7a for automatically positioning the sensor. By means of cable ties which are passed through the fastening slots 7 aand around the pipeline 6, the housing 7 can be fastened to the pipeline 6. The centrally located slots 7a for the cable ties allow simple and flexible installation of the sensor device 1 at different objects and locations.The fixing slots 7a are opposed to the pipe 6 and define a fixing portion 7b of the casing 7 which, when the casing 7 is fixed to the pipe 6, is in contact with the pipe 6. The second temperature sensor 2 that measures the piping temperature is disposed inside the mounting portion 7 a, but the first temperature sensor 3 is disposed outside. Thus, the second temperature sensor 2 is disposed in a portion 7 bof the housing 7 that is in contact with the piping 6, but the first temperature sensor 3 is disposed in a portion of the housing 7 that is not in contact with the piping 6. However, since the housing 7 is formed in one piece, thermal coupling between the pipeline 6 and the first temperature sensor 3 cannot be completely avoided. In this respect, it can be brought about only by the shorter distance between the second temperature sensor 2 and the pipeline 6 and by further measures, such as a heat conducting pad, that the thermal coupling between the pipeline 6 and the second temperature sensor 2 is greater than the thermal coupling between the pipeline 6 and the first temperature sensor 3.The two temperature sensors 2, 3 are also as far away as possible from the evaluation device 4 in order to prevent the heat radiation of the evaluation device 4 from distorting the measurement over a fee. In particular, the evaluation device 4 and the temperature sensors 2, 3 are arranged at different ends of the housing 7, such that a lateral distance between the evaluation device 4 and the temperature sensors 2, 3 corresponds in each case to at least 50% of a greatest lateral extent of the housing 7 (i.e. the length / height of the housing 7).The determination of the fluid flow and the detection of leaks are carried out by the respective arithmetic unit 4, 4'. In this case, the following relationships between the measured and optionally predicted temperature profiles can be taken into account, which are shown in FIGS. 4 ato 4 d. FIGS. 4a to 4d show diagrams of a relationship between temperature measurements and temperature predictions of a sensor device 1. in FIGS. 4a to 4d, curve 8 shows the measured ambient temperature, curve 9 shows the measured pipe temperature, and curve 10 shows a predicted pipe temperature.First, the arithmetic unit 4, 4' can determine the fluid flow based on an equilibrium state (equilibrium). In Fig. 4a it is shown how curves 8, 9 and 10 develop in this case. When the temperature 9 of the tube reaches room temperature 8, an equilibrium state (equilibrium) is assumed, which indicates that no water flows and no leakage is present.In addition, however, it is checked how the time curves of the ambient temperature and of the pipe temperature behave with respect to one another. Thus, the processor 4 bcan determine the fluid flow through the pipeline 6 on the basis of temperature slopes or temperature gradients within the time profile of the temperature measurements. Therefore, the processor 4 bis configured to determine a fluid flow through the pipeline 6 based on a temporal profile of the temperature measurements, which are contained at least in the first sensor data and optionally also in the second sensor data, and based on the second sensor data. In particular, it is possible to predict, on the basis of the temperature profile of the measured ambient temperature, how the temperature measurements of the tube temperature will develop. The processor 4 bmay then compare the predicted time profile of the temperature measurements at the second position with at least one temperature measurement at the second position to determine a fluid flow through the conduit 6. If the actually measured temperature curve 9 (for example a warming-up or a cooling curve) corresponds to the temperature curve 10 predicted by the algorithm, as in FIG. 4 b, no leak is likewise assumed. However, a temperature curve that is less steep than the predicted temperature curve (warm-up curve or cool-down curve) indicates a leak, wherein the computing unit can use differentials, derivatives and other relevant metrics to make an accurate diagnosis.To predict the time-course of the temperature measurements at the second position, the processor 4 bmay use a prediction machine learning model. In particular, a time-series prediction machine learning model or a regression machine learning model may be used to predict the time profile of the temperature measurements at the second position. Here, the time waveform of the temperature measurements at the first position is used as input values for the respective prediction machine learning model, and the time waveform of the temperature measurements at the second position is output as an output value. From the predicted and measured curves of the temperature measurements at the second position, the relevant metrics, such as the differentials and derivatives, can now be calculated to determine how the two curves negotiate with each other. If the curve of the measured course is less steep than the curve of the predicted course, this can indicate a leak.It is now shown in FIG. 4 d, how the measured ambient temperature 8, the measured pipe temperature 9 and the predicted pipe temperature 10 are related to one another, and what conclusions are drawn from this. The real water consumption is shown in the lower part of the diagram, the effects of the water consumption on the temperature curves 8, 9, 10 in the upper part. In this case, a distinction is made between the section types water consumption 11, non-noticeable region 12 and possible leakage 13. In particular, around 15:00 and between 21:00 and 22:30, cases occur repeatedly in which the predicted temperature development 10 produces a steeper curve than the actually measured temperature development 9, which indicates a leakage. Only between 20:00 and 21:00 and between 03:30 and 07:00 an equilibrium is established between the temperature curves. In the remaining periods, it cannot be recognized solely on the basis of the equilibrium whether or not there is a leak.In some implementations, the components of the sensor device 1 can be arranged on a printed circuit board 14 and coupled to one another via the latter. FIG. 5 shows such a printed circuit board 14 of a sensor device 1 with the temperature sensor 2, 3, the evaluation device 4 and a circuit 5 afor connecting a battery. However, the components are not only logically coupled to one another but also thermally coupled to one another by the printed circuit board 14. Therefore, the printed circuit board 14 can be divided into different partial regions, for example into a main surface 14 e, on which the evaluation device 4 is arranged, and two secondary surfaces 14 f, on each of which a temperature sensor 2, 3 is arranged. The thermal coupling is reduced in that the printed circuit board 14 has recesses (recesses) 14 bbetween the main surface 14 eand the secondary surfaces 14 fthat are intended to reduce the thermal coupling. In particular, the secondary surfaces 14 f, as shown in FIG. 5, can be connected to the main surface 14 eby thin webs (for example webs having a width of at most 3 mm). In order to improve the thermal coupling between the second temperature sensor 2 and the pipeline 6, a heat conducting pad may be provided on the rear side of the printed circuit board 14 (i.e. on the side of the printed circuit board which faces the pipeline 6). In addition, planar metallizations can be provided on the rear side of the printed circuit board 14, at the position of at least the second temperature sensor 2, in order to improve a thermal coupling to the pipeline 6 or to the ambient temperature.It can also be seen in FIG. 5 that the printed circuit board 14 can have further recesses in addition to the recesses 14 bfor the thermal insulation between the evaluation device 4 and the temperature sensors 2, 3. Specifically, the circuit board 14 has recesses 14a which, as shown in FIG. 6, interact with ridges 7c of the housing 7 to prevent the circuit board 14 from slipping relative to the housing 7. The printed circuit board 14 also has recesses 14 c, which correspond to the fastening slots 7 aof the housing 7 and make it possible to fasten the housing 7 to the pipeline 6 by means of cable ties. Finally, the circuit board 14 also has a recess 14 dthat enables a connecting cable 5 bto be inserted between the circuit 5 aand the battery. FIG. 6 shows the sensor device 1 with printed circuit board 14 and housing 7. It should be noted that the slits 7 a, 14 care continuous to allow the housing 7 to be fixed to the piping 6.The present invention relates not only to a sensor device 1 and a computing unit 4, 4', but also to a corresponding method which can be carried out, for example, with the aid of the sensor device 1 and / or computing unit 4, 4'. FIG. 7 shows a flow diagram of the method for determining a fluid flow through a pipeline 6, in particular for leak detection. The method comprises obtaining S 1 first sensor data of a first temperature sensor 3. the method comprises obtaining S 2 second sensor data of a second temperature sensor 2, wherein the first temperature sensor 2 has a greater distance from the pipeline 6 than the first temperature sensor 3. the method further comprises determining S 3 a fluid flow through the pipeline 6 based on the first and second sensor data.In one implementation, the present invention can already detect very small water flows starting from 3 milliliters per minute or 180 milliliters per hour. An advantage is that it is not designed to measure the exact water consumption, but mainly to detect the presence of water flows. It can have its own power supply and an independent radio connection, so that it also functions in the event of power, WLAN and network failures. The data may be processed not directly on the device but in an associated cloud system in some implementations, which enables advanced artificial intelligence to be deployed and avoids false alarms. In addition, not only equilibrium states will be detected in order to exclude leaks, but the dynamic changes can be analyzed by artificial intelligence. Accordingly, more complex buildings such as multi-family houses can also be monitored.In addition, the sensor device according to the invention enables non-invasive installation and can be fitted by home owners without expert and without tools. With a suitable battery and radio technology, it functions independently of remote energy or internet sources. In some implementations, the computing unit is based on an AI (artificial intelligence)-based monitoring and continuously recognizes patterns and anomalies in the water flow. The sensor device and evaluation by the computing unit enable damage prevention, because early detection enables rapid intervention in order to minimize potential damage and the loss of drinking water. By using this intelligent sensor, water resources can be effectively protected as well as costly water damage can be avoided.The invention relates to a water leakage sensor and a corresponding evaluation method which is carried out by the arithmetic unit. The sensor for detecting leaks is based on a measurement of room and pipe temperature, the room serving as a heat source and fresh water as a cooling source (or the room serving as a cooling source and hot water as a heat source). The water leakage sensor can have an autonomous radio connection and its own energy source. A calibration can be carried out automatically by means of an algorithm. The leakage detection can be carried out automatically by means of an artificial intelligence, i.e. by means of a machine learning model. Data compression can be used for sparse data transmission.With regard to further advantageous embodiments of the device according to the invention, reference is made to the general part of the description and to the appended claims in order to avoid repetitions.Finally, it should be expressly pointed out that the above-described exemplary embodiments of the device according to the invention serve merely for the discussion of the claimed teaching, but do not restrict it to the exemplary embodiments.List of reference characters1 Sensor device 2 Second temperature sensor, pipe-side temperature meter 3 First temperature sensor, room temperature meter 4 Evaluation device, local arithmetic unit 4' External arithmetic unit 4a Interface 4b Processor 4c Memory 5 Battery 5a Circuit for connecting a battery 5b Connecting cable 6 Pipe 7 Housing 7a Fastening device, Slot-shaped openings 7 b Befestigungs region 7 c Web of the housing 8 Ambient temperature 9 Pipe temperature 10 Prediction of the pipe temperature 11 Water consumption 12 No leakage 13 Possible leakage 14 Printed circuit board 14 a Ausnehmung for groove of the housing 14 b Ausnehmung for thermal insulation 14 c Slot-shaped opening for fasteners 14 d Ausnehmung for connecting cables for connection of a battery 14 e Hauptfläche surface of the printed circuit board 14 f Neben surface of the printed circuit board S 1 Obtaining first sensor data S 2 Obtaining second sensor data S 3 Determining a fluid flow

Claims

Sensor device (1) for a pipe (6) carrying a fluid, in particular for detecting leaks, comprising a housing (7) for arrangement on the pipe (6), a first temperature sensor (3), a second temperature sensor (2) and an evaluation device (4), wherein the temperature sensors (2, 3) are arranged in and / or on the housing (7) in such a way that, in the state of the housing fastened to the pipe (6), the first temperature sensor (3) has a greater distance from the pipe (6) than the second temperature sensor (2), wherein the evaluation device (4) is designed to process a) as a local arithmetic unit first sensor data of the first temperature sensor (3) and second sensor data of the second temperature sensor (2) and / or b) to transmit the first and second sensor data to a remote arithmetic unit (4') for processing, In order to determine a fluid flow through the pipeline (6), characterized in that the evaluation device (4) and the temperature sensors (2, 3) are arranged such that a lateral distance between the evaluation device (4) and the temperature sensors (2, 3) corresponds in each case to at least 50% of a greatest lateral extent of the housing (7).Sensor device (1) according to Claim 1, characterized in that the evaluation device (4) comprises a communication device, preferably configured for wireless communication, or is coupled to such a communication device, preferably wherein the communication device is configured to transmit the sensor data to the remote computing unit.Sensor device (1) according to claim 1 or 2, characterised in that the housing (7) comprises a fastening device (7a) with at least one, in particular two, preferably slot-like, opening (7a) in order to fasten the housing to the pipeline (6) by means of one or more fastening elements, preferably cable ties, preferably wherein the at least one opening (7a), preferably the two openings (7a), defines / define a fastening region (7b) of the housing (7), the fastening region (7b) is in contact with the pipeline (6) in the state of the housing (7) arranged on the pipeline (6), and the second temperature sensor (2) is arranged within the fastening region (7b) and the first temperature sensor (3) is arranged outside the fastening region (7b) of the housing (7).Sensor device (1) according to one of Claims 1 to 3, characterized in that a heat-conducting element, preferably a heat-conducting pad, is arranged on the second temperature sensor (2) in order to improve the thermal coupling to the pipeline (6).Sensor device (1) according to one of Claims 1 to 4, characterized in that the temperature sensors (2, 3) and the evaluation device (4) are arranged on a printed circuit board (14).Sensor device (1) according to Claim 5, characterized in that the printed circuit board (14) has a main surface (14e) on which the evaluation device (4) is arranged, and two secondary surfaces (14f) on which a temperature sensor (2, 3) is arranged in each case, wherein the printed circuit board (14) has a cutout (14b) in each case between the secondary surfaces (14f) and the main surface (14e), preferably wherein the secondary surfaces (14f) are connected to the main surface (14e) by webs.Sensor device (1) according to Claim 5 or 6, characterized in that at least one of the temperature sensors (2, 3) is arranged on a first side of the printed circuit board (14), wherein the printed circuit board (14) has, on a second side opposite the first side, at least one planar metallization which corresponds from its lateral position to a lateral position of the at least one temperature sensor (2, 3).Sensor device (1) according to one of Claims 1 to 7, comprising a circuit (5a) for connecting a battery (5), wherein the circuit (5a) is designed to provide electrical energy of the battery (5) to the temperature sensors (2, 3) and the evaluation device (4).Computing unit (4, 4') having an interface (4a) for obtaining first sensor data and second sensor data from a sensor device (1) according to one of Claims 1 to 8, wherein the first sensor data represent temperature measurements at a first position in and / or on the housing (7) with a greater distance to the pipeline (6), in particular an ambient temperature, and the second sensor data represent temperature measurements at a second position in and / or on the housing (7) with a smaller distance to the pipeline (6), in particular a surface temperature of the pipeline (6), wherein the computing unit (4, 4') comprises a processor (4b) which is designed to determine a fluid flow through the pipeline (6) based on the first sensor data and the second sensor data.The computing unit (4, 4') according to claim 9, characterized in that the processor (4b) is configured to determine a fluid flow through the pipeline (6) based on a temporal profile of the temperature measurements that are contained at least in the first sensor data and based on the second sensor data.The arithmetic unit (4, 4') according to claim 10, characterized in that the processor (4b) is configured to determine the fluid flow through the pipeline (6) based on temperature slopes or temperature slopes, in particular based on a slope of the temperature slopes or temperature slopes, within the time profile of the temperature measurements.The computing unit (4, 4') according to any one of claims 9 to 11, characterized in that the processor (4b) is configured to determine a predicted time profile of the temperature measurements at the second position based on a time profile of the temperature measurements at the first position and to compare the predicted time profile of the temperature measurements at the second position with at least one temperature measurement at the second position in order to determine a fluid flow through the pipeline (6), in particular wherein the processor (4b) is configured to compare a gradient of the predicted time profile of the temperature measurements at the second position with a gradient of a measured time profile of the temperature measurements at the second position in order to determine a fluid flow through the pipeline (6).The computing unit (4, 4') according to claim 12, characterized in that the processor (4b) is configured to determine the predicted time profile of the temperature measurements at the second position by means of a prediction machine learning model.The computing unit (4, 4') according to claim 13, characterized in that the processor (4b) is configured to adapt the prediction machine learning model based on the first and second sensor data, preferably by means of supervised learning, and / or characterized in that the prediction machine learning model is a time-series prediction machine learning model or a regression machine learning model.The computing unit (4, 4') according to any one of claims 9 to 11, characterized in that the processor (4b) is configured to determine a fluid flow through the pipeline (6) by means of a classification machine learning model based on the first and second sensor data, preferably wherein the classification machine learning model is trained on the basis of a temporal profile of the temperature measurements which are contained at least in the first sensor data and on the basis of the second sensor data to output a classification about a fluid flow through the pipeline (6).The computing unit (4, 4') according to any one of claims 13 to 15, characterized in that the processor (4b) is configured to provide the respective machine learning model with the time profile of the temperature measurements, which are contained at least in the first sensor data, as current input data, or characterized in that the respective machine learning model comprises a memory component, preferably a recurrent neural network, for example a long short-term memory, in order to keep the time profile of the temperature measurements, which are contained at least in the first sensor data, in the memory.The computing unit (4, 4') according to any one of claims 9 to 16, characterized in that the processor (4b) is configured to determine a constant temperature difference between the measured temperatures at the first position and at the second position based on a time profile of the temperature measurements, to detect an equilibrium state based on current temperature measurements and based on the constant temperature difference, and to determine a fluid flow through the pipeline (6) based on whether an equilibrium state is present.The computing unit (4, 4') according to any one of claims 9 to 17, characterized in that the processor (4b) is configured to estimate the volume flow of the fluid flowing and / or flown through the pipeline (6) based on a temporal profile of the temperature measurements, which are contained at least in the second sensor data, preferably by means of a regression machine learning model.The computing unit according to any one of claims 9 to 18, characterized in that the processor (4b) is configured to determine a leak based on a detected fluid flow through the pipeline (6).A system comprising the sensor device (1) according to any one of claims 1 to 8 and the computing unit (4') according to any one of claims 9 to 19, wherein the computing unit (4') is a remote computing unit arranged separately from the sensor device (1).Sensor device (1) according to one of Claims 1 to 8, comprising the arithmetic unit (4) according to one of Claims 9 to 19, wherein the evaluation device of the sensor device (1) comprises the arithmetic unit (4) as a local arithmetic unit.Method for determining a fluid flow through a pipeline (6), in particular for leak detection, using a sensor device (1) according to one of Claims 1 to 8 and a computing unit (4, 4') according to one of Claims 9 to 21, comprising: - obtaining (S1) first sensor data of a first temperature sensor (3); - obtaining (S2) second sensor data of a second temperature sensor (2), wherein the first temperature sensor (3) has a greater distance from the pipeline (6) than the second temperature sensor (2); - determining (S3) a fluid flow through the pipeline (6) on the basis of the first and second sensor data.

Images