Method and device for improving the efficiency and / or reduction of fine dust formation in a combustion

Abstract

The invention relates to a device (1) for analyzing the combustion of a combustible material in a fireplace, comprising a sensor device comprising an optical sensor, which is provided and configured to record an image and / or an image sequence of the combustion, and a computer device which is in data communication with the sensor device or to which a data connection can be formed, and which is in data communication with an output device or to which a data connection can be formed, wherein the output device is provided and / or configured to output a signal which correlates to a value which is characteristic of an efficiency of the combustion and / or of an amount of fine dust produced during the combustion.Furthermore, the invention relates to a method for analyzing the combustion of a combustible material in a fireplace and to machine-readable program code encoding this method or parts thereof. Furthermore, the invention relates to a machine-readable storage medium on which such machine-readable program code is stored.

Description

[0001] The present invention relates to a method and a device for improving the efficiency and / or reducing the formation of fine dust during combustion.

[0002] It is known from the prior art that certain Al-containing compounds, such as Al 2 O 3 , Al(OH) 3 , bauxite, talc, and / or kaolin or kaolinite (but also kaolin-containing clays), can be added as additives during the combustion of solids and reduce the formation of particulate matter. It is also known that combustion efficiency can be increased by controlling the air or oxygen supply.

[0003] For example, DE 10 2020 128 231 A1 discloses an incineration plant which comprises a kaolinite storage tank and a dosing device for adding kaolinite to the combustion process.

[0004] Furthermore, it is known from DE 10 2014 210 614 B4 that by adding an aluminum-containing inorganic compound as an ash and / or particulate matter reducing additive, raw materials such as chips from short rotation coppice (SRC), straw, grain and the like can be used for the production of pellets for heating purposes in small combustion plants.

[0005] Furthermore, EP 4 056 898 A1 discloses that parameters such as oxygen content, exhaust gas temperature, and combustion chamber temperature in a biomass heating system can be determined and used to control the biomass heating system. An AI model can be used for this control.

[0006] The applicant has determined in its internal, as yet unpublished studies that the effect of particulate matter reduction does not always correlate with the amount of particulate matter added. Particularly when using logs as fuel in single-room fireplaces, the results obtained regarding particulate matter development fluctuate greatly. The applicant attributes this to various effects, with fluctuations in the combustion process being considered particularly critical. These fluctuations in the combustion process occur particularly in single-room fireplaces. These are typically not equipped with sensors to measure the parameters relevant to combustion and to control or regulate the single-room fireplace accordingly.

[0007] There is therefore a need to provide a device and a method to evaluate a combustion process taking place in the fireplace with regard to its efficiency and / or the formation of fine dust and to be able to regulate the fireplace or the combustion process taking place in the fireplace based on an evaluation result.

[0008] If sensors are integrated into the stove, it would be preferable to include data from such sensors in the assessment. However, an assessment of the efficiency and / or particulate matter formation of the combustion process should preferably be possible regardless of whether sensors integrated into the fireplace are available to measure data from the combustion process taking place in the fireplace.

[0009] This object can be achieved by the subject matter of the independent patent claims. Preferred embodiments are described as such below and / or are the subject matter of the dependent claims.

[0010] A device according to the invention for analyzing the combustion of a combustible material in a fireplace is characterized in that it comprises a sensor device comprising a sensor, for example an optical sensor, which is provided and configured to record data from the combustion process, for example as a (preferably continuous) measurement of the amount of dust in the exhaust air within the furnace and / or by recording an image and / or an image sequence of the combustion. The sensor device can be arranged in the furnace or outside the furnace. If multiple sensors or sensor devices are present, data from one or more of these sensors or sensor devices can be used to analyze the combustion.

[0011] The device also comprises a computer device, which is in data communication with the sensor device and / or the sensor devices or to which a data connection can be established, and which is in data communication with an output device or to which a data connection can be established. The output device is provided and / or configured to output a signal that correlates to a value that is characteristic of a combustion efficiency and / or of a quantity of particulate matter produced during combustion.

[0012] Such a device makes it particularly easy to analyze combustion in terms of its efficiency and / or fine dust formation and, if necessary, to take measures to improve combustion, i.e. to increase efficiency and / or reduce the amount of fine dust produced.

[0013] In a first preferred embodiment of the device, the sensor device, for example the optical sensor, is not an integral part of the fireplace. Further preferably, the sensor device, for example the optical sensor, is arranged in a housing that is movable relative to the fireplace, in particular relative to a combustion chamber of the fireplace in which the combustion of the combustible material takes place.

[0014] This makes it possible to position the sensor device, such as the optical sensor, in such a way that a particularly precise analysis of the combustion is possible. Furthermore, the ability to move the sensor relative to the fireplace allows one sensor to be used to analyze multiple combustion events in different fireplaces. This results in a cost advantage, as not every fireplace needs to be equipped with such a sensor device.

[0015] The present invention is described below using the example of an optical sensor. An optical sensor is preferably part of a sensor device. Accordingly, the aspects of the invention disclosed for the example of the optical sensor are generally intended to apply to any type of sensor device. Such a sensor device can have one or more (possibly different) sensors.

[0016] Preferably, the optical sensor, and preferably the sensor device, are arranged in a common housing with a data transmission device. Such a data transmission device is provided and configured to provide a preferably at least partially wireless data connection to the computer device.

[0017] Such a data transmission device can therefore be used to transmit the data generated by the sensor to a computer. The sensor data can then be analyzed by the computer. This allows the unit comprising the optical sensor to be designed particularly compactly. This improves the mobility and transportability of the unit comprising the optical sensor. Furthermore, it is comparatively easy to scale a computer device located at a distance from the sensor according to requirements. This allows the computing power of the computer to be adjusted to any changing requirements.

[0018] In a further preferred embodiment of the device, a sensor, and preferably a sensor device, is an integral part of the fireplace. This makes it possible to position the (optical) sensor in such a way that a measurement adapted to the furnace and replicable is taken. Measurement accuracy can be increased and—particularly if the relative position of the sensor with respect to the combustion chamber remains unchanged—a change in the sensor data can be clearly attributed to changed conditions within the combustion chamber, and external influences due to a changed sensor position can be largely excluded.

[0019] Preferably, the (preferably optical) sensor is connected or connectable to the computer device via a data transmission device. The data transmission device can provide wireless and / or wired data transmission. The data transmission device can be arranged in a common housing with the sensor or arranged separately from it. It can be arranged (if necessary together with the sensor) within the furnace (or the fireplace), but is preferably arranged outside the combustion chamber, preferably also outside the furnace, in order to reduce thermal stress.

[0020] Such a data transmission device can preferably be used to carry out furnace-specific, permanent and selective optimizations.

[0021] Preferably, at least one piece of data recorded by a sensor from the combustion process is characteristic of a combustion parameter selected from a group comprising a flame pattern, a temperature, a (gas) pressure, a fuel quantity, a fuel condition (e.g., moist / dry), and a set gas or air supply. These parameters have proven particularly suitable for assessing combustion quality.

[0022] Preferably, a sensor is an exhaust gas sensor, which preferably determines a parameter of the exhaust gas selected from a group comprising a CO content, a CO2 content, a particulate matter content, a water (vapor) content, an oxygen content, a hydrocarbon content, a soot content, a tar content, a NOx content, an SxOy content, a (negative) pressure, and a (flow) velocity. These parameters have proven particularly suitable for characterizing the exhaust gas, on the one hand, but also for drawing conclusions about the quality of combustion, on the other.

[0023] Sensors selected from a group that includes NDIR sensors (e.g. for determining CO2, O2, CO, CnH2n), optical sensors (flame image), CCD sensors, electrochemical sensors (O2, CO), absorption hygrometers, dew point mirror hygrometers, aspiration hygrometers (water, humidity), gravimetric sensors (particles & fine dust), light sensors / laser scattering / light refraction (fine dust and gas particles), semiconductor membranes (e.g. chimney negative pressure), thermocouples (temperature) and lambda sensors have proven to be particularly suitable for determining one or more of the above-mentioned parameters. Such sensors are sufficiently sensitive to detect the relevant parameters, are readily available and comparatively inexpensive to purchase and maintain.

[0024] In both embodiments described above, the computing device is preferably part of a computer network. This also facilitates scalability, for example, by outsourcing computing power to additional computing devices in the computer network. Furthermore, it is possible to use one computing device to analyze the sensor data from several different sensors, thereby increasing the utilization of the computing device. Furthermore, it is possible to assign other tasks to the computing device during periods of non-use in order to increase its utilization.

[0025] For example, if the computing device is connected to the optical sensor or other computing devices via the Internet, computing power does not need to be distributed across multiple computers in the computer network. This allows for load balancing and faster delivery of analysis results.

[0026] The sensor device preferably comprises a plurality of sensors, wherein at least two sensors can detect different physical parameters of the combustion. This makes it possible for combustion parameters to be detected by one sensor which are not perceptible to the other sensor. For example, it would be conceivable for one of the sensors to be an optical sensor, whereas another sensor is a temperature sensor or a gas sensor. With such a combination, a flame image could be detected by the optical sensor, whereas the other sensor detects a temperature profile of the combustion process or a composition of the exhaust gas. By combining different sensors in this way, the combustion process can be analyzed in more detail and an adjustment of the combustion parameters to optimize the combustion process and / or the formation of fine dust is easier.

[0027] Preferably, at least two sensors are optical sensors. Preferably, each of these at least two optical sensors is provided and configured to record an image and / or an image sequence of the combustion in a wavelength range that is different from the other of these at least two optical sensors. If an image or an image sequence of the combustion is recorded at different wavelength ranges, flame and / or ember images from different wavelength ranges can be used for analysis. Such an observation of different wavelength ranges can also provide information about the temperature distribution within a flame or the combustion chamber. Knowledge of a temperature profile, in turn, makes it possible to control parameters within the fireplace such that combustion takes place with greater efficiency and / or with less fine dust formation.In this regard, it would be conceivable, for example, that a gas flow is controlled in such a way that it is increased in the areas where a very low combustion temperature was detected, so that combustion is accelerated there.

[0028] A device is preferred in which the output device a display device which is preferably arranged in a common housing with the sensor and / or the sensor device and / or is provided and configured to display a recommended course of action to increase the efficiency of combustion and / or to reduce the amount of fine dust produced during combustion, or a control device by means of which a control command can be given to the fireplace which correlates to an increase in the efficiency of combustion and / or the reduction of the amount of fine dust produced during combustion.

[0029] If the output device is a display device, it can show the user which steps they should take to increase combustion efficiency and / or reduce the amount of particulate matter produced during combustion. The display device could, for example, be a screen that uses appropriate animated graphics to indicate which steps should be taken to achieve the desired effect.

[0030] In a particularly preferred embodiment, at least one sensor is part of a mobile device, for example, a camera of the mobile device, such as a smartphone. It can be connected to an external computing device via a data connection (for example, a mobile data connection (3G, 4G, UMTS, EDGE, 5G). Additionally or alternatively, a processor of the mobile phone could also serve as the computing device. A screen of the mobile phone could function as an output device.

[0031] It has proven particularly advantageous that, in one embodiment, the display device can be arranged in a common housing with the sensor and / or the sensor device. Since the sensor should be located near the fireplace anyway to detect combustion parameters, in this case the display device is also in the immediate vicinity of the fireplace. This allows the operator easy access to the fireplace and allows them to immediately implement the necessary instructions displayed on the display device.

[0032] If the output device is a control device that can issue a control command to the fireplace, this control device can, possibly without human involvement, control the fireplace in such a way as to increase combustion efficiency and / or reduce the amount of particulate matter produced during combustion. For example, a regulator for controlling an air supply flow could be controlled accordingly to achieve (possibly locally) faster combustion and a (locally) higher temperature.

[0033] The control device could also send a control command that triggers the addition of fuel to the combustion chamber of the fireplace. This is preferred if the computer calculates from the sensor data that additional fuel would be beneficial to maintain efficient combustion.

[0034] Furthermore, a method for analyzing the combustion of a combustible material in a fireplace is a solution to the underlying problem. This method is characterized by the following steps: Recording a data item from the combustion process and / or an image and / or an image sequence of the combustion by a sensor device comprising a sensor, transmitting the sensor data to a computer device, classifying the sensor data or data based thereon by the computer device, outputting a signal which correlates to a value which is characteristic of an efficiency of the combustion and / or a fine dust formation.

[0035] Such a method makes it particularly easy to analyze whether combustion is proceeding efficiently and / or whether particulate matter formation is within specified limits. This method is preferably divided into various process steps that run separately from one another. It is particularly preferred that the sensor records the data at a different location than the computer device classifies the (image) data. In particular, the computer device is not arranged in a common housing with the sensor. It is particularly preferred that the computer device is connected to the sensor via a computer network, for example, via the Internet.

[0036] Preferably, the (preferably processor-based) computer device receives combustion data relating to the combustion, in particular spatially, temporally, and / or thermally resolved, determined by (or by means of) the sensor device (particularly within the framework of a computer-implemented method step). "Receiving" can be understood to mean data transmission, for example, via a digital data exchange and / or via a (at least partially and preferably entirely) wired and / or wireless network. Furthermore, "receiving" can also be understood to mean forwarding and / or retrieving environmental data determined and / or recorded by the sensor device.

[0037] Furthermore, the (in particular processor-based) computer device is provided and configured to carry out a method (in particular in a computer-implemented method step), preferably processing data characteristic of the combustion, preferably using a, in particular trainable, machine learning model. The model preferably comprises a set of, in particular trainable, parameters which are set to values ​​learned as a result of a training process. Furthermore (hereby or through the processing) at least one notification variable is determined which is characteristic of an optical sensor signal, in particular of a significance of the optical sensor signal with regard to an efficiency of a combustion and / or a fine particle content in the exhaust gas.Preferably, the data characteristic of the combustion are the sensor data and in particular data derived from (raw) sensor data determined and / or generated by the at least one sensor device and / or data derived therefrom.

[0038] Preferably, the notification size is calculated using a processor device and / or data processing device, by applying at least one (computer-implemented) computer vision method in which (computer-implemented) perception and / or detection tasks are carried out, for example (computer-implemented) 2D and / or 3D object recognition methods and / or (computer-implemented) methods for semantic segmentation and / or (computer-implemented) object classification ("image classification") and / or (computer-implemented) object localization.

[0039] Preferably, the data acquired by the sensor and / or the sensor device are stored in a (temporary and / or local) storage device, preferably a storage device of the computer device. The storage device can be a ring buffer. Preferably, only the image data of a predetermined duration of a combustion process, a specific fireplace, and / or a predetermined storage location (preferably in the storage device of the computer device) are kept available. Particularly preferably, the image data are overwritten and / or deleted after the predetermined duration of a combustion process and / or the predetermined storage location have been exceeded.

[0040] The evaluation of the image data is preferably carried out in the evaluation device using an artificial neural network or artificial intelligence (AI).

[0041] Preferably, the artificial intelligence system is at least configured and trained to derive one, preferably several, parameters characteristic of the combustion process from the sensor data, wherein the characteristic parameter (or parameters) are selected from a group comprising a flame temperature, a flame pattern, a fuel quantity, and a fuel state.

[0042] Preferably, the artificial intelligence system is configured and trained to detect a flame. It has been shown that detecting a flame based solely on color filtering is not sufficient to reliably identify a flame. Likewise, it is not possible to reliably detect a flame based on rigid rules regarding color and / or brightness changes, since, for example, color changes in LED lamps can also exhibit a similar pattern of color and / or brightness changes.

[0043] However, it has been shown that image classification by an artificial intelligence system (after training) is very capable of recognizing a flame in an image and, preferably, also determining its position. Such a system has proven to be particularly robust when video data is available. On the one hand, it is particularly easy to generate sufficient data sets to train the AI ​​with video data. However, due to the almost chaotically changing shape and color of a flame (at least to outsiders), an AI with video data is able to correctly recognize a flame (and, preferably, its position) in every frame / image of the video with a particularly high probability.

[0044] As described above, it is preferable for the artificial intelligence system to be able to determine the position of a flame. This makes it possible to define a so-called "region of interest" (ROI). This can then be analyzed in more detail. Limiting the analysis to such a region in an image or video reduces the workload and thus the required computing power, memory, and processing time.

[0045] Preferably, color filtering is performed in the ROI. For this purpose, the color information is preferably converted to HSV (hue, saturation, value) color space. It has been shown that for further analysis (possibly independently of each other), H values ​​are preferably between 19% and 95%, S values ​​are preferably above 65%, and V values ​​are preferably above 70%.

[0046] Preferably, the individual pixels are filtered according to a brightness value. It has been shown that it is advantageous to filter out pixels with a grayscale value of at least 230 out of 255 and use them for further analysis. This type of filtering makes it possible to detect a flame even under poor ambient conditions, for example, in low-light environments such as those with increased smoke or in dark images.

[0047] Preferably, after the pixels belonging to a flame have been identified based on the pixel selection described above, a signal and / or function is created from these pixels. For this purpose, pixels identified as belonging to a flame are assigned the value 1, while other pixels are assigned the value 0. When viewed in 2-dimensional space, the position of the flame is described by the x and y values ​​of the points (pixels) assigned the value 1.

[0048] Flame movement can be determined by comparing these values, which are calculated based on two images (frames) taken at different times, as described above. Preferably, the x and y values ​​determined from a first frame represent the original position of the flame. A change in the x and / or y values ​​determined based on two different frames indicates flame movement. The amplitude and frequency (if the time difference between the recording of the frames under consideration is known) of this movement can provide important clues to combustion properties. For example, parameters such as a high amplitude and / or frequency can be attributed to so-called "flickering" of the flame, which can be a sign of an unfavorable and / or excessive gas supply and thus inefficient combustion.

[0049] The training can, for example, be carried out using various image sequences, which are analyzed according to the above pattern and then assigned to (quality) classes of the combustion system based on expert opinion and / or (exhaust gas) measurements. For example, the frequency of flickering ("flicker degree") could be a relevant parameter for evaluation.

[0050] The flicker level is preferably determined using an autocorrelation function of the signal. This function is preferably performed for both the original (sensor) signal and the absolute values. The resulting signals differ depending on the flicker level. Example values ​​for a flame with only slight flame movement (i.e., low flicker level) and for a flame with strong and / or rapid flame movement (i.e., high flicker level) are shown in Table 1: Table 1 Visual assessment of flame movement Average value of an autocorrelation function of the original signal Average value of the magnitudes of an autocorrelation function of the original signal Average value of an autocorrelation function of the magnitudes of the original signal quiet / low 1.72e-06 8.96e-05 3.50e-04 fast / strong -3.01e-06 4.0e-4 1.99e-3

[0051] As can be seen from Table 1, in combustion with rapid / strong flame movements (i.e., a high degree of flickering), the fluctuation of the values ​​is very high, and therefore, considering the average value of an autocorrelation function of the original signal is not always useful. It has therefore been shown that using the absolute values ​​(magnitudes) of an autocorrelation function of the original signal is more meaningful. The average values ​​of an autocorrelation function of the magnitudes (or absolute values) of the original signal differ sufficiently to reliably distinguish the various flame patterns from one another.

[0052] Due to the specific characteristics of the signals, it is advantageous to consider additional parameters. In addition to the average values ​​of the various processed (or original) signals shown in Table 1, it has proven advantageous to also consider the variances and / or frequency spectra of the signals.

[0053] Preferably, the signals (newly acquired through processing) are further processed. This has been shown to improve the robustness of the results. In particular, it allows for generalization of the signals and better comparability. For example, influencing factors such as flame size and / or distance between flame and sensor can be reduced when evaluating the results. This also makes it possible to compare a close-up of a fire with a distance shot of a fire, although a close-up image shows much greater absolute area changes than a distance shot because the areas of the pixels classified as relevant are different.

[0054] Preferably, a large number of data items are provided for evaluation by the artificial intelligence system. These are preferably at least 3, preferably 4, more preferably 5, more preferably 6, especially preferably 7, most preferably all data from a group that includes a variance, a mean value, a frequency response, a frequency amplitude, a frequency spectrum, an output signal (fire / frame ratio), a change in an output signal, and an autocorrelation function. If an AI has several or even all of these data items available (for example, to determine an output), particularly precise analysis is possible and, for example, a change in the distance between the sensor and the flame can be correctly detected and its influence calculated.

[0055] Preferably, the AI ​​is trained on different furnaces. In this regard, it is conceivable, for example, that the AI ​​is trained on specific (preferably the most frequently occurring) furnaces. To do this, a flame pattern is forced to be trained, preferably for the furnace type(s) to be trained (for example, under laboratory conditions / on the test bench). This flame pattern is / are then saved as characteristic of certain combustion processes and / or furnace types. If a similar flame type is detected in such a furnace type during real operation, such a saved pattern signal can serve as a cross-correlation function for the detected signal (pattern) and provide a furnace operator with feedback on the extent to which the currently detected flame pattern deviates from the one detected in the laboratory / on the test bench.

[0056] Preferably, a noise level of the flickering of the detected flame is determined. This has proven advantageous because it can be used to create a signal filter. Convolution with the signal filter can result in the outlier(s), which in turn represent further AI features (i.e., AI features, namely features and / or input parameters that are suitable for training an AI or are actually used for training), from which further information about the flame image can be derived.

[0057] In a preferred variant, the color values ​​for the temperature designation are analyzed from the pixels classified as relevant for evaluating the combustion process. Preferably, the flame colors are determined in different color spaces, preferably in at least two color spaces selected from HSV, RGB, and LAB. The values ​​of the individual parameters are preferably used as AI features in training.

[0058] It is also possible, and in some embodiments preferred, to convert a (temporal) signal determined as above into a spectral function using a Fourier transformation. This enables a generalization of the signals, which are no longer dependent on the actually measured sensor signals or variables derived from them, e.g. the size of the flame. This also makes it possible to calculate out influences such as the distance of the (optical) sensor from the flame when determining the sensor data. As already explained above, for example, an analysis of a close-up of a fire (with the same flame pattern) would result in a much larger change in area (in terms of magnitude) than a long-distance shot, since the pixel surfaces are different.After Fourier transformation and consideration of frequency domains, these variables are normalized, allowing for a better comparison of sensor data recorded under different conditions (e.g., distances). Different signals can be represented, for example, in a spectrogram.

[0059] Once a fire is detected, further data relevant to the combustion process can be determined. For example, it is possible to determine the amount of fuel present in the combustion chamber from the ROI (e.g., determined as described above). It has been shown to be advantageous for the fuel in the combustion chamber (also referred to as the firebox) not to exceed half, preferably one-third, of the height of the firebox. If there is too much fuel in the combustion chamber, a warning signal can be issued.

[0060] To determine the amount of fuel present in the combustion chamber, the ROI is preferably first determined. This determination provides information about the location of the combustion chamber and the flame shape. By observing these parameters over a period of time, the height of the combustion chamber can be determined with great accuracy, as this usually corresponds to the position of the highest (detected) flame. At the same time, the combustion chamber is often brighter than the surroundings or an external combustion chamber boundary due to the flame located therein. This allows a fairly accurate determination of the dimensions of the combustion chamber and the amount of fuel it contains. Too much fuel and / or poor distribution of the fuel in the combustion chamber can have a negative impact on the efficiency and / or effectiveness of the combustion process. It can also cause increased soot formation and increased particulate matter emissions.

[0061] Preferably, a fuel condition is also determined. This is preferably done optically by an artificial intelligence system. An (optical) determination of the fuel condition is particularly well possible when a flame is detectable and thus particularly good conditions / lighting conditions exist for determining a fuel condition using visual AI. Preferably, the fuel and / or patterns detectable on the fuel are used for the evaluation. The patterns can preferably be determined in the form of rectangles and / or triangles by analyzing the darker corners on a fuel, for example a log, using visual AI. Depending on the strength and size of these (darker) boundary lines, it can be determined whether the fuel has a more compact or brittle shape.

[0062] It has been shown that assessing the condition of a fuel is particularly advantageous for determining when more fuel should be added to the combustion chamber. It can also provide indications of a potentially suboptimal air (or gas) supply / regulation. It has been shown that more air should be supplied to a more compact fuel than to a brittle fuel. This not only optimizes efficiency but also the efficiency of fuel consumption. Optimization has a direct impact on particulate matter emissions, because when less fuel is burned, less particulate matter is produced. For example, adjusted air regulation with brittle fuel ensures that the fuel burns more slowly, while the efficiency of the furnace remains the same or increases.

[0063] Preferably, at least one notification signal is provided for controlling the output device (for example, for displaying an instruction on a display device and / or outputting a control signal) depending on the at least one notification variable (particularly in a computer-implemented method step). It is conceivable that the notification signal is a signal to be transmitted to a user, such as a person authorized to control the fireplace and / or another person.

[0064] In a preferred variant of the method, the storage device is an external storage device, in particular a cloud-based and / or central storage device. This has the advantage that the storage capacity can be adapted to changing storage capacity requirements particularly easily and preferably independently of the fireplace.

[0065] Alternatively or additionally, a preferred variant of the method provides for the computing device to be a computing device spaced apart from the sensor, in particular a cloud-based and / or central computing device. This has the advantage that the computing capacity can be adapted particularly easily and preferably independently of the fireplace to changing computing capacity requirements. In particular, the consideration of high-resolution 3D (image) data of the combustion process as well as any meteorological data that may need to be taken into account (in particular air pressure and wind speed), as well as the forecast of meteorological data for a specific time period, can involve very large amounts of data and be computationally intensive, so that a computing device integrated into a common housing with the sensor device is not capable of doing this, or at least not within an acceptable time.Furthermore, a cloud-based and / or centralized computing facility has the advantage that potentially erroneous data from an individual fireplace can be identified as such and accordingly (not) taken into account when training a machine learning model.

[0066] Preferably, at least one control function of the fireplace is controlled and / or executed, in particular automatically, depending on the notification signal. For example, this could involve controlling the oxygen supply and / or the fuel.

[0067] Furthermore, the underlying problem is solved by a method for controlling the combustion of a combustible material in a fireplace. Based on the result of the combustion analysis according to the method for analyzing the combustion of a combustible material in a fireplace described above, information is provided. This information correlates to a control command, which in turn correlates to increased combustion efficiency and / or reduced particulate matter formation during combustion.

[0068] Preferably, a user is given a regulation recommendation for the fireplace and / or a recommendation for feeding an additive and / or a fuel into a combustion chamber on an output device, preferably an optical output device such as a screen.

[0069] The output device is preferably an optical output device, also referred to as a display device or screen. An instruction and / or control recommendation is preferably displayed on such a display as a graphic (image) element, in particular a symbol or in the graphic form of a symbol or icon. This offers the advantage that these can be intuitively, quickly, and reliably understood by both people, such as a person authorized to control the fireplace, and by machines.

[0070] Alternatively or in addition to issuing a control recommendation to a user, a control command can be sent to the fireplace. Preferably, such a control command results in the supply of a fuel and / or an additive and / or oxygen to the combustion chamber.

[0071] The additive preferably has an ash- and / or particulate matter-reducing effect during fuel combustion. The additive can also be introduced together with a fuel, for example in the form of a compact containing fuel and additive, such as a pellet or briquette. The additive is preferably selected from a group comprising kaolin, metakaolin, halloysite, dickite, and nacrite, thermally activated three-layer silicates, and / or combinations of these substances.

[0072] Preferably, the additive is provided together with a fuel in the form of a briquette or pellet. This has been found to be particularly advantageous because the additive can be continuously released during the combustion process, thus ensuring consistently low particulate matter formation even when combined with other fuels, such as logs. Even with mechanical ash resuspension due to the disintegration of burnt logs, the particulate matter-reducing additive is always available in a combustion chamber in sufficient quantities to achieve the desired particulate matter reduction. Packaging in the form of a pellet or briquette,

[0073] For the purposes of this invention, the term "briquette" should be understood in its usual definition, namely as a shaped piece (in a cuboid or egg shape) pressed from a specific fine-grained material. Briquettes differ from other pressed pieces, such as pellets, in their significantly larger size. While pellets have a very small shape, usually a straight circular cylinder with a diameter of less than approximately 2 cm, and are obtained by pelletizing, a briquette, even in its smallest dimension, usually has a size of well over 2 cm, for example, at least 3 cm, preferably ≥ 4 cm, and most preferably ≥ 5 cm. Along its largest dimension, an egg-shaped briquette is usually at least 5 cm. The smallest common commercial form of briquette is the so-called 3-inch briquette, which, as their name suggests, has a dimension of approximately 7.5 cm in its main dimension.

[0074] As mentioned above, the additive preferably comprises one or more substances selected from a group consisting of metakaolin, halloysite, dickite, and nacrite, thermally activated three-layer silicates, and / or combinations of these substances. These substances have proven particularly effective in reducing ash and / or fine dust. Preferably, at least one of these substances is thermally activated. In the thermally activated form, even greater effectiveness can be achieved. If the substance in question is already introduced in thermally activated form, thermal activation no longer needs to take place during the combustion process in the fireplace. This ensures that the particularly active form is available more quickly, and no thermal energy is required to convert the additive into the thermally activated form during the combustion process.This can increase the overall efficiency of combustion.

[0075] Thermal activation of the additive prior to briquette combustion has also proven advantageous, as it results in the conversion of, for example, kaolin to metakaolin, which is particularly suitable for binding particulate matter at low temperatures. It is assumed that, even with the known compositions, at least partial thermal activation occurs during the combustion process. However, this only occurs at comparatively high combustion temperatures, which may not be reached in some areas of the combustion chamber and / or the flame or fire. Therefore, for compositions without a thermally activated additive, particulate matter formation cannot be reliably reduced, depending on the flame temperature and the location of combustion in the combustion chamber.

[0076] It is conceivable, and in some embodiments preferred, for the briquette to comprise a further additive that reduces the formation of fine dust. This further additive is preferably selected from a group comprising Al 2 O 3 , Al(OH) 3 , bauxite, talc, kaolin, kaolinite and kaolin-containing clays, metakaolin, thermally activated three-layer silicates, fly ash, and microsilica, and / or combinations of these substances. It has been shown that the presence of a second component that reduces the formation of fine dust (hereinafter also referred to as the "second additive") can significantly reduce the formation of fine dust even at very low combustion temperatures, for example ≤ 650°C, ≤ 600°C, ≤ 550°C, or even ≤ 500°C, and / or in the case of incomplete combustion, for example due to a low oxygen supply.Even at high combustion temperatures and sufficient oxygen supply, the addition of a second additive has proven to be advantageous, as it improves ash formation and, by preventing clumping or at least reducing the tendency of the ash to clump, enables a good long-term oxygen supply through the fuel and the combustion chamber.

[0077] The briquette preferably contains Al 2 O 3 and / or aluminum hydroxide and / or aluminum oxide hydroxide and / or bauxite, however, in a weight fraction of ≤ 5%, preferably ≤ 3%, more preferably ≤ 2%, most preferably ≤ 1%, and particularly preferably no addition of Al 2 O 3 and / or aluminum hydroxide and / or aluminum oxide hydroxide and / or bauxite. In this context, "containing no additive" should be understood to mean that such a component is not actively added to a briquette (or to a mixture from which a briquette is pressed), but rather - if present at all - such a component is merely added to a briquette as an impurity to another component. This can prevent the ash from containing excessively high levels of aluminum and / or aluminum compounds from forming during combustion, which could potentially enter the exhaust gas.This should be avoided, particularly due to increased health concerns regarding some aluminum compounds.

[0078] The second additive preferably differs from the first additive in at least one property. The first and second additives can differ in their chemical composition by selecting different compounds from the group of chemical compounds defined above. Alternatively or additionally, however, it is also possible for the first and second additives to differ from one another in at least one other property. This could, for example, be a different (crystal) water content. Another exemplary difference between the first and second components could be the (average) particle size (preferably d 50 (Sedigraph)). Such differences between the first and second additives preferably result in them reaching their maximum activity at different times during the combustion of a briquette.For the above examples, it was shown that particles with higher (crystal) water content and / or particles with a larger (mean) particle size reach their activity maximum at higher temperatures in the combustion chamber and / or at later times during a combustion process.

[0079] The additive is preferably treated at a temperature in the range of approximately 600-900°C, preferably 650-800°C. At this temperature, for example, kaolin is converted to metakaolin. Preferably, this thermal treatment triggers a reaction that releases previously bound water of crystallization. In addition to or as an alternative to the treatment at the above-mentioned temperature, the thermal treatment preferably takes place over a period of 1-90 minutes, preferably 5-75 minutes, and particularly preferably 10-60 minutes. As can be seen from these values, it takes some time until sufficient or (almost) complete conversion to the reactive component has taken place, which can particularly effectively bind and / or reduce the particulate matter produced during combustion.

[0080] It inevitably follows that without thermal activation of the additive, at certain times (for example at the beginning of a combustion process and / or immediately after the addition of a briquette) there is not enough of the thermally activated, particularly reactive component present to effectively reduce the formation of particulate matter.

[0081] Experience shows that during uncontrolled combustion processes, temperatures can range from a very wide temperature range in a combustion chamber. This temperature range typically covers temperatures of around 100–1000°C. Within this temperature range, the temperature may be locally insufficient to form the reactive component (e.g., metakaolin), which is particularly suitable for reducing particulate matter, or it may be locally too high, leading to further conversion to silicates with reduced reactivity. The temperature distribution in a combustion chamber can only be controlled to a limited extent - if at all - so that, for example, the formation of metakaolin during combustion cannot be monitored or controlled. The same often applies to the residence time, which during combustion can last from a few seconds to several hours.If the residence time is too short, often not enough of the reactive component can be formed, while if the residence time is too long, deactivation of the reactive component (e.g. metakaolin) can occur.

[0082] Preferably, the additive is suitable for binding fine dust-forming sodium and / or potassium salts even at combustion temperatures ≤ 650°C. This enables the entire composition to reduce fine dust formation even at such low combustion temperatures.

[0083] Preferably, the pellet or briquette does not contain any added binder. It has been shown that in most cases, the compression pressure applied during pellet or briquette production is sufficient to form the product into a stable composite. A binder can then be omitted. This is not only advantageous for cost reasons, as binders are usually among the more expensive components (relative to their weight), but also prevents potentially harmful combustion products from the binder from being produced during combustion.

[0084] The particulate matter produced during the combustion of organic fuels, such as biomass, contains, in particular, alkali metal compounds, such as sodium or potassium compounds. Of particular importance are alkali metal hydroxides, chlorides, and sulfates. Although some of the substances produced during combustion are gaseous at the time of combustion, they can condense or precipitate as solids during the cooling of the combustion gases, aggregate into larger particles, and thus be released as particulate matter.

[0085] The fuel is preferably selected from a group comprising wood particles, for example wood chips, wood shavings or sawdust; stalk-like biomass, for example straw, hay, husks, grass, bran or hemp; peat; fruit or vegetable kernels or stones, for example cherry stones or olive stones; grains; nutshells, for example walnut or coconut shells; spent grain or pomace, for example fruit, vegetable or plant component pomace, preferably coffee bean pomace, malt grain or olive press pomace; biogenic residues; lignite, for example lignite coke or lignite powder; dry distillers' grains (DDGS); hard coal, for example hard coal coke or hard coal powder; charcoal, preferably charcoal powder; yeast cells; plant fibres; cellulose fibres; paper and petroleum coke.With these fuels, the formation of fine dust, and in particular the formation of the above-mentioned alkali metal compounds, especially sodium or potassium compounds, can be reduced particularly effectively. It is particularly preferred that the fuel be a cellulose-containing fuel. With such fuels, the additive has demonstrated particularly good effectiveness in reducing ash and / or fine dust.

[0086] The above-described process of alkali metal compound formation can be prevented or at least reduced by the controlled addition of an additive or additive-containing pellets or briquettes to a combustion process or as fuel based on the analysis result. This can be explained in particular by the fact that alkali compounds formed during combustion, as exemplified below using the reaction of metakaolin with KOH, can be captured from the exhaust gas and converted into low-volatility components.

[0087] In the example mentioned, metakaolin reacts with KOH according to the following reaction equation to form a low-volatility potassium aluminum silicate and water: Al 2 Si 2 O 7 + 2 KOH → 2 KAlSiO 4 + H 2 O

[0088] The potassium aluminum silicate is stable up to very high temperatures and can thus be removed from the combustion chamber as ash. Furthermore, the formation of potassium aluminum silicate particles—particularly during the cooling of the exhaust gas, for example, in a chimney or heat exchanger—promotes the deposition of other substances from the exhaust gas, such as additional potassium aluminum silicate particles. These particles grow and precipitate from the exhaust gas or can be easily separated from it (for example, by centrifugation).

[0089] Other substances such as potassium sulfate or chloride can also be converted into high-temperature stable substances and thus bound: 3 Al 2 Si 2 O 7 + 3 K 2 SO 4 + 6 KOH → (K 2 Al 2 / 3 SiO 4 ) 6 • Al 2 (SO 4 ) 3 + 3 H 2 O

[0090] The above-mentioned reactions are demonstrated using metakaolin as an example. However, binding of particulate matter is possible with any substance that produces reaction products that are sufficiently resistant to high temperatures and easily precipitate or separate from the exhaust gas. Silicates, aluminosilicates, and silicate-containing components have proven particularly suitable.

[0091] An embodiment in which the additive is used in particulate form or is present in this form in a pellet or briquette has proven particularly preferred. The particles of the additive preferably have a particle size (Sedigraph) d 95 ≤ 100 µm, preferably ≤ 50 µm, more preferably ≤ 20 µm, and particularly preferably ≤ 10 µm. Additionally or alternatively, particle sizes d 95 ≥ 1 µm, preferably ≥ 2 µm, more preferably ≥ 3 µm, and particularly preferably ≥ 5 µm are preferred. These minimum sizes can improve handling and, in particular, prevent these particles from entering the environment as fine dust even during handling.

[0092] The particle size is preferably adjusted by grinding and, if necessary, subsequent sieving and / or sifting.

[0093] In a preferred embodiment, the additive is used together with at least one other substance, for example an auxiliary substance. Both organic and inorganic auxiliary substances are conceivable as additional substances or auxiliary substances. Depending on the application, the (weight) proportion of the auxiliary substances in relation to the additive can be selected within a very wide range of 99.9 - 0.1, preferably 99 - 1, more preferably 75 - 25. In particular, if an auxiliary substance is also a fuel, its weight proportion can be well over 75%, for example over 90%, ≥ 95% or even ≥ 98% and can in particular be present in a very large excess compared to the additive. The auxiliary substance serves in particular to enable a delayed and / or continuous release of the additive over a long period of time. As a result, its ash- and / or particulate matter-reducing effect can be developed or maintained over a long period of time.

[0094] This effect, namely enabling a delayed and / or continuous release of the additive over a long period of time, can also be achieved by adding the additive as granules, rather than as a powder as described above. Such granules can be mixed with the above-mentioned materials, but can also be used as a pure substance, the latter with or without a binder, dry or moist, preferably as a pure additive and / or preferably with a residual moisture content of < 30 wt.%, more preferably < 20 wt.%, particularly preferably < 13 wt.% and most preferably < 5 wt.% moisture. The (average) grain size d 50 (e.g. by means of a) dry sieving with a sieve tower or b) laser diffraction particle size analysis) of the granules is > 0.5 mm, preferably > 3 mm, most preferably > 5 mm and < 50 mm. The granules can come from extrusion, spray drying, build-up granulation or a comminution process.The shape can be round, cylindrical, or angular. Mixtures of different (average) grain sizes are advantageous.

[0095] The additive is preferably fed by the control device to the combustion chamber in a quantity such that the additive has a (weight) proportion based on the total mass used for firing (i.e. including the fuel) preferably in the range of 0.1 - 5 wt.%. This proportion has been found to be particularly effective in efficiently reducing the amount of particulate matter that is otherwise usually produced during fuel combustion. In particular, a proportion of 0.5 - 4 wt.%, more preferably 0.75 - 3 wt.% and particularly preferably 1 - 2.5 wt.% has proven to be particularly efficient. Using the example of a pellet or briquette as described above with, for example, 40 wt.% of the additive, this means that approximately 1 weight part of pellets or briquettes should be used for 20 weight parts of the remaining fuel in order to achieve a (weight) proportion of the additive based on the total mass used for firing of 1.9 wt.%.-%.

[0096] With regard to the fireplace, there are no restrictions whatsoever, either in the use of a device as described above or in the implementation of the method described above. One advantage of the device - particularly when the computer device is arranged at a distance - is that it is extremely flexible. It can be used, for example, in individual fireplaces, single-room heating systems (e.g., a fireplace in the living room), or boiler systems of any size (e.g., central heating systems for single-family or multi-family homes), and in larger heating or power plants. Such a device can also be used, and the method is applicable and advantageous, for outdoor fireplaces, in particular those fired with wood, coal, briquettes, pellets, or other (preferably solid) fuels.Such a fireplace can, for example, be a fireplace from a group that includes a barbecue fireplace of any kind, a fire bowl, a campfire site, a fire basket, a fire table and an outdoor fireplace.

[0097] By means of a method and / or a device as described above, it is possible to reduce the amount of particulate matter produced during combustion of a solid by at least 10%, preferably at least 25%, more preferably at least 50%, and in many particularly preferred cases even by 75% or more. The percentages mentioned in this regard (like all other percentages mentioned in this disclosure, unless otherwise defined) are in each case percentages by weight. The base value in this case is the amount of particulate matter produced during combustion of a combustible material without the addition of an additive as described above.

[0098] In many cases, the addition of an additive as described above also leads to an increase in the efficiency of fuel combustion and thus to a reduction in the CO content in the exhaust air. This can be attributed, among other things, to the improved ash sintering behavior and the associated better supply of fresh air and / or oxygen, resulting in more complete combustion and thus higher energy efficiency.

[0099] Preferably, a device and method as described above can achieve cost savings because efficiency is increased and exhaust gas aftertreatment or exhaust gas purification can often be dispensed with. Even if exhaust gas aftertreatment or exhaust gas purification is necessary, the required systems and equipment can be designed to be smaller or less complex.

[0100] Preferably, a device as described above is configured, suitable, and / or intended to carry out a method as described above and all method steps described in connection with the method individually or in combination with one another, or individual method steps using the device described above. Conversely, the method can be carried out with all features described in the context of the device individually or in combination with one another.

[0101] Optionally, a device as described above can be configured, suitable and / or intended to combine a method as described above and all method steps described in connection with the method with other methods for reducing fine dust from combustion processes, such as filtration methods, cyclone methods or electrostatic methods, in particular electrostatic precipitators installed in the exhaust air stream.

[0102] Furthermore, the invention is directed to a machine-readable program code which encodes a method as described above or parts thereof or encodes an instruction for carrying out a method as described above or parts thereof.

[0103] Furthermore, the invention is directed to a machine-readable storage medium on which a machine-readable program code as described above is stored. The storage medium can be, for example, a magnetic storage device, a magneto-optical storage device, an optical storage device, or a solid-state storage device, for example, a flash memory.

[0104] Further advantages and embodiments can be seen from the attached figures: Fig. 1A shows a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace in a first state; Fig. 1B shows the result of an analysis of the Fig. 1Ashown combustion process; Fig. 2A a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace in a second state; Fig. 2B result of an analysis of the Fig. 2A shown combustion process; Fig. 3A a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace in a third state; Fig. 3B result of an analysis of the Fig. 3A illustrated combustion process; Fig. 4A a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace in a third state; Fig. 4B result of an analysis of the Fig. 4A combustion process shown; Fig. 5A a representation of a computer keyboard 30 with the backlight switched on, taken under analogous conditions to Fig. 1A, 2A and 3A ; Fig. 5BResult of an analysis of the Fig. 5Ashown illustration of the computer keyboard 30; Fig. 6A a diagram illustrating the changes in the data associated with a flame during a quiet combustion; Fig. 6B a diagram illustrating the changes in the amounts of the data associated with a flame during a quiet combustion; Fig. 7A an autocorrelation function of the Fig. 6A shown function; Fig. 7B an autocorrelation function of the Fig. 6B shown function; Fig. 8A a diagram showing the change in the magnitudes of the data assigned to a flame during unsteady combustion with rapid flame movements; Fig. 8B a diagram showing the absolute change in the data assigned to a flame during unsteady combustion with rapid flame movements; Fig. 9A an autocorrelation function of the Fig. 8A shown function; and Fig. 9B an autocorrelation function of the Fig. 8B function shown.

[0105] Figure 1A shows a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace 10 in a first state. To record the Figure 1A An optical sensor was used in the illustration shown. The fireplace 10 with the combustion chamber 30 located therein can be seen. A partially glowing fuel 40 is located in the combustion chamber.

[0106] Figure 1B shows an analysis result which is based on the Figure 1A The analysis data shows brightly those areas of the sensor-detected area that have been classified as relevant for combustion.

[0107] To generate such an analysis result, the position of the flame is first determined. This allows the definition of a region of interest (ROI), which roughly represents the combustion chamber. This can significantly reduce the amount of data, as data outside the ROI can be excluded from further analysis.

[0108] Once the ROI has been defined, color filtering is performed. For this purpose, the sensor data recorded at different times—in the example shown, image data and also referred to as frames in this description—is first converted to the HSV (hue, saturation, value) color space. Initial parameter optimization showed that particularly good results for subsequent feature extraction are achieved when the H value is between 19% and 95%, the S value is above 65%, and the V value is above 70%.

[0109] Furthermore, the individual pixels are filtered according to brightness. For this purpose, each pixel previously selected by the filtering process must meet a specified grayscale condition. In the example shown, pixels with a grayscale value of at least 230 out of 255 are passed through. This filtering makes it possible to detect a flame relatively reliably even in poorly lit environments, such as in dark images or with heavy smoke.

[0110] Further evaluation of this in Figure 1BThe brightly displayed areas allow conclusions to be drawn about the quality of the combustion taking place in the combustion chamber 30. It is particularly helpful to compare such analysis results at different points in time. From the changes in the sensor data classified as relevant for the evaluation of the combustion process, it can be determined, for example, whether the brightly displayed areas represent a glowing fuel or are more likely to be attributed to a flickering fire. This part of the evaluation will be discussed in connection with the Figures 6A - 9B discussed in more detail.

[0111] Figure 2A shows a further representation of the fireplace with a combustion process 20 taking place in the fireplace 10 in a second state. In this representation, which, as in Figure 1A in a visible wavelength range, it can be seen that the fireplace 10 is identical to the one shown in Figure 1A is shown. Here, too, the combustion material 40 is arranged in the combustion chamber 30. In comparison to the illustration from Figure 1A However, it can be seen that some of the Figure 1A very bright areas in Figure 2A are less bright. This is due to a less intense glow in these areas.

[0112] This visual impression is also enhanced by the Figure 2B shown result of an analysis of the Figure 2A The number and area of ​​the combustion process shown in Figure 2B The areas also shown in bright colours, which were classified as relevant for the evaluation of the combustion process in the combustion chamber 30, are considerably smaller than the corresponding areas in Figure 1B .

[0113] A like in Figure 2BThe analysis result shown could indicate that a combustion process has just begun and the embers have not yet reached the entire fuel material 40. Non-glowing fuel material 40 could be located between the sensor and the glowing fuel material, thus preventing the hidden embers from being recognized as relevant for the evaluation of the combustion process. However, the widespread distribution of the sensor data classified as relevant for the combustion process speaks against such an interpretation. At the beginning of a combustion process, such a widespread distribution of embers is usually not to be expected. It is more likely that - especially if an analysis result has already been obtained, such as in Figure 1B was present - that the combustion of the fuel is no longer taking place under optimal conditions. For example, the fuel could be running low or the oxygen supply could be too low.

[0114] Figure 3Aalso shows a representation of a fireplace 10 with a combustion process 20 taking place in the fireplace 10 in a third state. As in Figure 3A As can be seen from the representation of the data recorded by a sensor operating in the visible range, the combustion process 20 takes place with a bright blazing flame.

[0115] This visual impression is also enhanced by the Figure 3B analysis result of the Figure 3A The combustion process described in the Figures 1B and 2B are also in Figure 3B , the sensor data classified as relevant for the evaluation of the combustion process are highlighted. This is particularly evident in comparison with the representation in Figure 1B As can be seen, the brightly displayed data points are distributed over a very wide area. As the underlying sensor data from Figure 3AAs can be seen, in the present case a bright blazing flame is the cause of the wide distribution of the data points relevant to the combustion process.

[0116] However, an alternative cause for such a pattern could also be a very high fuel level 40 in the combustion chamber 30. Even in such a case, a comparable data pattern could be obtained with glowing fuel. However, as already explained above, to distinguish such analysis results, it is advisable to compare the analysis data at two different points in time, preferably two points in close succession. While bright flames lead to a rapidly changing analysis data pattern, this is a significantly slower process with embers.

[0117] In particular, a different flame pattern becomes apparent when analyzing image sequences, such as videos. Flickering fires with rapidly changing high-temperature centers can be easily distinguished in image sequences from glowing fuel, in which the particularly high-temperature centers remain essentially static.

[0118] Figure 4A shows another representation of a fireplace 10 with a combustion process 20 taking place in the fireplace 10, but in a fourth state. Again, the sensor operating in the visible range is used as well as for detecting the Figures 1A, 2A and 3A. The sensor data shown in Figure 4A The combustion process 20 shown also takes place with flame formation, but with less high and blazing flames than in Figure 3B shown.

[0119] This visual impression is also enhanced by the Figure 4BThe analysis result shown is confirmed. The sensor data classified as relevant for evaluating the combustion process are again highlighted and cover a smaller area compared to the representation in Figure 3B.

[0120] It has been shown that artificial intelligence (AI) systems trained on this can detect differences in the flame pattern after training and derive information about the status of the currently ongoing combustion process 20 from the flame pattern. This information can be used to draw conclusions about the efficiency or effectiveness of the combustion process 20. It can also be used to determine the amount of particulate matter to be expected in the exhaust gas from the currently ongoing combustion process 20.

[0121] After training an artificial intelligence system as described above, this system was able to identify each of the four Figures 1B, 2B , 3B and4B to identify the analysis results presented as belonging to a combustion process. To verify the artificial intelligence system, a negative test was carried out. For the negative test, the method used to create the Figures 4A used sensor a computer keyboard 60 with backlight. In Figure 5A is a picture of such a keyboard 60 with the backlight switched on in the visible area.

[0122] Even if the backlight is switched on in the Figure 5A computer keyboard 60 shown in analogous conditions to the Figures 1A, 2A , 3A and 4A is not clearly recognizable, the resulting and in Figure 5B The result shown is an analysis of the Figure 5AThe sensor data shown in the figure shows bright areas that could possibly be classified as relevant for detecting a combustion process. Even if the Figure 5B The arrangement of the bright areas shown is similar to that shown, for example, in Figure 2B are shown, the artificial intelligence system was nevertheless able to reliably recognize that the Figure 5B The system can therefore, based on the analysis of the sensor data as shown in the Figures 4A and 4B or 5A and 5B, it is possible to determine whether the detected object is actually a fire 20.

[0123] The Figures 6A and 6Bshows a diagram illustrating the changes in the data associated with a flame during steady combustion. As described above, in many cases it is advantageous not only to determine data at a single point in time during the combustion process, but also to record data at different points in time during the combustion process. This data is preferably recorded at fixed intervals. In the Figures 6A - 9B In the diagrams shown, values ​​are plotted against a frame number. Each frame represents a data set acquired by the sensor device at a specific point in time during the combustion process. The x-axis shows the consecutive number of the frames recorded at different points in time. In this example, each frame was recorded at a fixed time interval, so the consecutive number correlates to a specific time.

[0124] On the vertical axis in the Figure 6A The diagram shows values ​​that represent the ratio of the change in the position and / or area of ​​the data points considered relevant for the evaluation of combustion. The determination of the data points considered relevant is preferably carried out as described in the Fig. 1A and 1B described.

[0125] To generate the Fig. 6A In the graphs shown, the data obtained from the fireplace at a first point in time are processed and compared with the analogously processed data obtained at a second point in time. For this purpose, the positions of the pixels considered to belong to a flame are determined analogously to the Figures 1B, 2B , 3B and 4B at different times. From the displacements of the pixels belonging to the flame along the X and / or Y direction, a Figure 6A generated function shown.

[0126] In Fig. 6A A graph is shown, as it would be obtained from the analysis of a fire whose flames move comparatively little. This represents a relatively quiet combustion, which is also referred to as a "quiet fire." As can be seen in Fig. 6A can be detected, the plotted values ​​fluctuate around a mean value. This behavior is to be expected, since a flame, even if it is moving, essentially blazes where the combustible material is present. Even if the flame is moved, for example, by a gas stream, a flame will be detectable again at a later time near the combustible material (which is not moved by the gas stream). Such a return is indicated by negative values.

[0127] Overall, especially in comparison to the following described Figure 8A - in Fig. 6Asmall changes in the flame position. This is typical for a steady burning flame. The variance between the values ​​at different times (or frames) is Figure 6A shown graphs are comparatively small.

[0128] In Figure 6B Another diagram is shown to illustrate the change in the data associated with a flame during a steady combustion process. In contrast to the diagram in Figure 6A shown illustration are in Figure 6B However, the magnitudes of the changes (also called absolute changes) are shown. Therefore, the values ​​plotted along the vertical axis are all positive. The magnitudes of the changes are usually below 0.008 units, which is characteristic of smooth combustion.

[0129] The Figures 7A and 7B show graphs of autocorrelation functions of the Figures 6A and 6BThe graph shows the deviation between the average value of all changes up to one frame and the average value of all changes across all frames. As expected, this graph therefore approaches the value 0.

[0130] However, since the combustion characterized by these graphs is a smooth combustion, the changes are small across the entire graph and are significantly below 0.001.

[0131] As already mentioned above in connection with the Figures 6A - 7B As indicated, the Figures 8A - 9B Analogous representations, but for a combustion process in which strong changes in the position of the flame and the sensor area covered by flames occur. A combustion process in which blazing flames move vigorously and / or rapidly is also referred to as "rapid fire."

[0132] In the in the Figures 8A - 9BThe diagrams shown are along the horizontal axis analogous to the Figures 6A - 7B the number of frames. This axis thus forms a time axis.

[0133] Essentially, the representation of the Figure 8A to that in Figure 6A , the representation of the Figure 8B to that in Figure 6B , the representation of the Figure 9A to that in Figure 7A and the representation of the Figure 9B to that in Figure 7B .

[0134] However, while in the Figures 8A and 8B the changes compared to the Figures 6A and 6B are not clearly visible in the course of the graph, the changes become clear when looking at the amounts of the changes: In contrast to the Figures 6A and 6B In the graphs shown, the maximum changes are < -0.01 and + almost 0.02, about twice as strong as in the Figures 6A and 6BThese values ​​alone reveal an uneven course of the combustion processes represented by such a graph.

[0135] However, this difference is even more evident when comparing the Figures 9A and 9B graphs shown with those shown in the Figures 7A and 7B The autocorrelation functions shown show, at least for low frame numbers, changes that are approximately 4 times the values ​​shown in the Figures 7A and 7B correspond to the changes shown.

[0136] Using the signal data presented above, an algorithm was developed to determine and evaluate the flame pattern 20 and the temperature. Based on the flame pattern 20 and the temperature, an evaluation can be performed. For example, if the flame pattern is unsteady and the high-temperature areas are distributed over a large area, this indicates a poor flame pattern and an excessively high temperature. In this case, it can be assumed that too much air is being supplied to the fireplace, and combustion is inefficient due to the excess oxygen available.

[0137] To train the system, training videos were recorded, each depicting conditions classified by domain experts, i.e., assigned to different combustion types. The algorithm was trained based on these training videos and the class assigned by the domain experts. eXtreme Gradient Boosting (XGBoost) was used for classification.

[0138] In an exemplary training, the four parameters flame pattern (flickering: high, okay, little), temperature (good, high, low), fuel quantity (combustion chamber overfilled, okay, little), and fuel condition (solid, ash-like, etc.) were selected for classification. For temperature and flame pattern, labels and / or classifications created by domain experts were associated with the extracted features using an XGBoost algorithm. Preferably, the labels are analyzed individually. The result is two decision trees that allow an estimation of the temperature and fire pattern labels based on the defined features.

[0139] Other algorithms are also conceivable. However, XGBoost has proven particularly suitable, as it delivers the best results of the algorithms considered, especially for the underlying dataset. Other algorithms, such as k-nearest neighbor, support vector machine, and others, could be advantageous, especially for larger datasets.

[0140] After an initial training with the manually classified training videos, the training method was changed and the algorithm was further trained with new, unclassified training videos. The artificial intelligence system was further improved with this additional training data. Thus, relationships between different parameters were recognized and the states of new / unknown sensor data were correctly determined.

[0141] The applicant reserves the right to claim all features disclosed in the application documents as essential to the invention, provided that they are new, individually or in combination, compared to the prior art. List of reference symbols

[0142] 10Fireplace, oven 20Combustion process, flame, embers 30Combustion chamber, combustion space 40Combustion material, fuel 50Area 60(Computer) keyboard

Claims

1. Device (1) for analyzing the combustion of a combustible material in a fireplace, characterized by a sensor device comprising a sensor, which is provided and configured to record data from the combustion process and / or an image and / or an image sequence of the combustion, and a computer device which is in data connection with the sensor device or to which a data connection can be formed, and is in data connection with an output device or to which a data connection can be formed, wherein the output device is provided and / or configured to output a signal which correlates to a value which is characteristic of an efficiency of the combustion and / or of an amount of fine dust produced during the combustion.

2. Device (1) according to claim 1, characterized in thatthe sensor, and preferably the sensor device, is not an integral part of the fireplace, but is preferably arranged in a housing which is movable relative to the fireplace, in particular relative to a combustion chamber of the fireplace in which the combustion of the combustible material takes place.

3. Device (1) according to one of the preceding claims, characterized in that the sensor, and preferably the sensor device, is arranged in a common housing with a data transmission device which is provided and configured to provide a preferably at least partially wireless data connection to the computer device, wherein the computer device is preferably part of a computer network.

4. Device (1) according to one of the preceding claims, characterized in thatthe sensor device comprises a plurality of sensors, wherein different physical parameters of the combustion can be detected by at least two sensors, wherein preferably at least two sensors are optical sensors, wherein each of these at least two optical sensors is provided and configured to record an image and / or an image sequence of the combustion in a wavelength range different from the other of these at least two optical sensors.

5. Device (1) according to one of the preceding claims, characterized in thatthe output device - is a display device which is preferably arranged in a common housing with the sensor and / or the sensor device and / or is provided and set up to display a recommended course of action to increase the efficiency of combustion and / or to reduce the amount of fine dust produced during combustion, or - is a control device by means of which a control command can be given to the fireplace which correlates to an increase in the efficiency of combustion and / or the reduction in the amount of fine dust produced during combustion.

6. Method for analyzing combustion of a combustible material in a fireplace, characterized bythe steps: - recording a data item from the combustion process and / or an image and / or an image sequence of the combustion by a sensor device comprising a sensor, - transmitting the sensor data to a computer device, - classifying the sensor data or data based thereon by the computer device, - outputting a signal which correlates to a value which is characteristic of an efficiency of the combustion.

7. Method for controlling the combustion of a combustible material in a fireplace, characterized in that based on a result of the analysis of a combustion according to claim 8, information is provided which correlates to a control command which correlates to a higher efficiency of the combustion and / or a reduced formation of fine dust during combustion.

8. Method according to claim 7, characterized in thata user is shown a control recommendation for the fireplace and / or a recommendation for adding an additive and / or fuel to a combustion chamber on a screen.

9. Machine-readable program code which encodes a method according to one of claims 6 - 8 or parts thereof or encodes an instruction for carrying out a method according to one of claims 6 - 8 or parts thereof.

10. Machine-readable storage medium (100) on which a machine-readable program code according to claim 9 is stored.

Images