METHOD FOR CONTROLLING THE ADDITION OF A FLOCCULANT TO A SLUDGE

Description

[0001] The invention relates to a method for controlling the addition of a flocculant to a sludge, wherein the flocculant is added to the sludge and the sludge is subsequently at least partially dewatered in a dewatering device, for example, a dewatering screw, a decanter, a sieve dewaterer, etc., wherein images of the dewatered sludge and / or of the liquid dewatered from the sludge are created using a camera system. The invention also relates to a dewatering device for sludge with a camera system and to a computer program product.

[0002] Sludge refers to solids dispersed in a liquid, where the solids are typically finely dispersed and very fine-grained, and the amount of liquid is comparatively small compared to the amount of solids. Technical processes often aim to further dewater the sludge and thus increase its solids content. This is usually achieved by adding a flocculant, typically a polymer, to the sludge, followed by further dewatering. The flocculant causes flocculation, meaning the fine-grained solids aggregate into larger flocs, making it advantageous to separate the liquid from the flocs. The amount of flocculant required for optimal flocculation depends on many parameters of the sludge—for example, the particle size distribution of the solids and, in particular, the nature of the solids (i.e., whether they are mineral, fibrous, or biological).If less than the optimal amount of flocculant is added, only a reduced flocculation is achieved, meaning that fine-grained, unflocculent solids remain in the liquid, leading to less than optimal dewatering. Conversely, if the optimal amount of flocculant is added, no greater than optimal dewatering is achieved. In fact, this may even result in a deterioration of dewatering. To complicate matters further, the achievable dewatering, and thus the dry matter content of the sludge, is not a fixed, predetermined target value, but rather strongly dependent on the sludge parameters themselves. Controlling the process based solely on the sludge dry matter content is therefore difficult. In current technology, flocculant dosage is often determined by operating personnel based on intuition and thus subjectively, which, in the interest of safe operation, tends to lead to overdosing.In general, the use of flocculants is costly. Efficient use is therefore desirable, and overdosing should be avoided for ecological reasons as well.

[0003] For example, EP3134354 B1 discloses a method for dewatering sludge on a screen. According to this method, the sludge is necessarily guided on a screen which is cleaned by washing nozzles before an inlet area, and the flow behavior of the sludge and the free screen surface are optically monitored in a control area. This method therefore requires the presence of a screen, screen conditioning, and the monitoring of the free screen surface, thus enabling the measurement of parameters independent of the actual sludge.

[0004] KR20130033148A discloses a method and a system for dewatering with controlled flocculant addition, wherein images of sludge agglomerated in a sedimentation tank are captured using a photographic device. To analyze the sludge properties and determine the flocculant dosage, the images are compared with previously stored reference images.

[0005] US5380440A describes a device and method for dewatering sludge, wherein, after filtration, images of the dewatered sludge are recorded with a video camera. The images are compared with images representative of a preferred dewatering process to assess the moisture content of the dewatered sludge.

[0006] The object of the invention is a method for controlling the addition of a flocculant to a sludge, wherein the control is carried out according to objective criteria by an assessment of the sludge and allows for the greatest possible ecological and economic efficiency.

[0007] According to the invention, this is achieved by evaluating the images of the dewatered sludge and / or the liquid dewatered from the sludge with a computer-implemented image analysis model, the model being previously trained with a training data set, the training data set comprising, on the one hand, training images of dewatered sludge and / or the liquid dewatered from the sludge, and, on the other hand, the training sub-images formed from the individual training images by subdivision, and the classification of the individual training sub-images, wherein the model subdivides the individual images into sub-images, classifies the sub-images of the individual image, and assesses the dewatering of the sludge based on the classification, wherein the addition of the flocculant is controlled based on the assessment of the dewatering of the sludge.According to the invention, images of the dewatered sludge and / or the liquid extracted from the sludge are captured using the camera system, and these images are evaluated using a computer-implemented image analysis model, designed as an artificial neural network running on a standard industrial PC. Crucially, for evaluation purposes, the individual images are first divided into sub-images, with the model then classifying these sub-images. Thus, the dewatering of the sludge is assessed based on the classification of these sub-images. Specifically, the image itself is not classified, for example, by comparing it to training images.Regarding the subdivision of the images into sub-images, it should be noted that the sub-images should be as small as possible, while still allowing for inferences about the surface texture, particularly the grain size, relief, or cracks, of the dewatered sludge, or about the bubbles or foam formation of the dewatered liquid. An evaluation based on a single pixel of the sub-image is therefore not possible, as this would not allow any conclusions to be drawn about, for example, the texture. It is particularly advantageous to identify an optimal subdivision based on the training images. For this purpose, training images with distinctive surface texture, grain size, relief, and cracks or bubbles and foam formation are first identified.The subdivision of these training images is then progressively increased, thus reducing the size of the sub-images. This increasing subdivision is continued as long as the sub-images still reveal the distinctive texture, grain, crack, etc., or parts thereof. This identification of the optimal subdivision is performed, for example, by a trainer of the computational model and is particularly efficient. Identifying the optimal subdivision during training by the computational model itself is also conceivable, although this entails a significant computational and resource expenditure. Unexpectedly, the evaluation and classification of the sub-images allows for more meaningful results with far less computational effort and lower demands on the computer equipment than would be possible with a comparison based on individual images. Further advantages arise during the generation of the training dataset.Thus, only a few training images are needed, as the subdivision process yields a multitude of training sub-images. In particular, the classification of these training sub-images can be carried out efficiently, since the training image as a whole is assigned to a class, with the resulting training sub-images essentially inheriting the class of the training image. The subdivision of the images into sub-images is analogous to the subdivision of the training images into training sub-images. Similarly, the creation of the images is also advantageously analogous to the creation of the training images, for example, with regard to the arrangement or the chosen perspective of the camera system in relation to the dewatered sludge or the liquid extracted from the sludge.

[0008] In the embodiment of the method according to the invention, the computational model classifies the partial images of the dewatered sludge according to a surface texture, in particular a grain size, relief, or cracks of the dewatered sludge. According to the invention, the computational model is first trained with a training dataset, wherein the training images are subdivided into training sub-images, and the computational model is trained with the classified training sub-images. The subdivision of the training images into training sub-images is carried out such that the training sub-images have the smallest possible size, wherein the size of the training sub-images is chosen to be at least large enough to reproduce the surface texture, in particular the grain size, relief, or cracks of the dewatered sludge.After the computer model has been trained, the images of the dewatered sludge are analyzed using the computer-implemented image analysis model. The model divides and classifies the individual images into sub-images, thus assessing the dewatering of the sludge. If, in addition or alternatively, images of the liquid extracted from the sludge are created, the computer model analyzes the resulting sub-images for bubbles trapped in the liquid or foam forming on the liquid surface and classifies the sub-images accordingly. Again, the sub-images created by subdivision must be of sufficient size. Advantageously, the sub-images are also analyzed based on color properties, particularly hue, saturation, and brightness.

[0009] In a further advantageous embodiment of the method, the computer-implemented computational model classifies reduced partial images, having been trained on these reduced training images. Reduction of a partial image here refers to the averaging or summarizing of pixel information from a number of pixels. The reduced partial images thus represent a mosaic of the partial images. Unexpectedly, it was found that despite the reduction, the partial images still allow conclusions to be drawn about the surface texture, in particular the grain size, relief, or cracks, of the dewatered sludge, or about the bubbles or foam formation of the dewatered liquid, thus enabling successful classification with a further reduction in computational effort and resource expenditure.

[0010] An advantageous embodiment of the method involves classifying the partial images into at least two classes: a first class representing insufficient dewatering or flocculant addition, and a second class representing excessive dewatering or flocculant addition. The dewatering of the sludge is then assessed based on these classified partial images. This two-class classification—i.e., insufficient or excessive dewatering or flocculant addition—corresponds to the concept of a two-point control system. In this way, a state of optimal dewatering or flocculant addition can be achieved; however, according to the control concept, the system never reaches a state of equilibrium. Alternatively, the partial images can be classified into at least three classes. Each class represents satisfactory, insufficient, or excessive dewatering or flocculant addition.This corresponds to the concept of a three-point control system, whereby, in the range of optimal dewatering or flocculant addition, the amount of flocculant added to the sludge remains unchanged, while in the case of insufficient or excessive dewatering or flocculant addition, the amount of flocculant added to the sludge is increased or decreased, respectively. The control is thus more uniform. Advantageously, in the described configurations, the frequency distribution of the classes of the partial images is used to assess the dewatering of the sludge. For example, the flocculant addition can be controlled according to the most frequent class. Advantageously, the quotient of the frequencies of two classes is calculated, e.g., the classes for insufficient and excessive dewatering, and the assessment of the dewatering is then carried out according to this quotient.By specifying an "optimal quotient" for optimal dewatering or flocculant addition as a setpoint, the control system consequently attempts to establish this "optimal quotient." The "optimal quotient" is known, for example, from training the computational model with training images or training sub-images, which are classified as representing optimal dewatering. Advantageously, another class is introduced for capturing sub-images, with invalid sub-images assigned to this class. Sub-images are considered invalid if they do not depict dewatered sludge and / or liquid dewatered from the sludge. Sub-images might, for example, primarily represent elements of the dewatering device, such as the screw of a dewatering or transport screw or the screen of a screen dewatering system. These sub-images thus do not allow any conclusions to be drawn about the dewatering or flocculant addition.Results relating to the addition of the flocculant are deemed invalid and excluded from further evaluation, and are not included in the assessment of dewatering.

[0011] In an advantageous embodiment of the method, the creation of the training data set includes setting a desired dewatering or flocculant addition during operation of the dewatering device, wherein the training images of the dewatered sludge and / or the dewatered liquid are classified as representing satisfactory dewatering or flocculant addition. Equally advantageously, after setting the desired dewatering or flocculant addition during operation, the flocculant addition is reduced, resulting in insufficient dewatering or flocculant addition, or the flocculant addition is increased, resulting in excessive dewatering or flocculant addition, wherein the corresponding training images are then classified as representing insufficient or excessive dewatering or flocculant addition, respectively.For example, at one operating point of the dewatering device, the addition of flocculant is determined by a valve position, and a valve position of 45% results in satisfactory dewatering and flocculant addition. Reducing the valve position determining the flocculant addition to 40% or increasing it to 50% results in insufficient or excessive dewatering and flocculant addition, respectively. The training images generated in this way, representing optimal, insufficient, and excessive dewatering and flocculant addition, are then subdivided into training sub-images and appropriately classified to train the computational model. The described method for creating the training dataset is characterized by its efficiency, as training images for the classes optimal, insufficient, and excessive dewatering and flocculant addition can be generated and classified quickly.If necessary, for example because a different type of sludge needs to be dewatered, a new training data set can be created very quickly and the computer-implemented computational model can be trained accordingly.

[0012] In a further advantageous embodiment of the method, images of the dewatered sludge and / or the dewatered liquid are created in an area that includes an interface of the dewatering device, in particular a free screen surface of a screen on which the sludge is guided, for example, in an edge region of the screen or downstream of a barrier acting on a screen, a free surface of a dewatering or conveying screw, or a wall that is in direct contact with the sludge or the dewatered liquid. Surprisingly, it was found that in areas encompassing interfaces, the optimally dewatered sludge develops a particularly pronounced surface texture, especially grain structure, relief, or cracks. Insufficient flocculant leads to smooth, reflective structures, whereas excessive flocculant results in coarse, matte textures.Regarding the dehydrated liquid, marked turbidity, foaming and deposits can be observed in areas encompassing interfaces, depending on the dehydration or the addition of the flocculant.

[0013] The invention also relates to a dewatering device for sludge according to claim 11.

[0014] Advantageously, the camera system comprises a digital camera and at least one light source, with the digital camera having an optical axis and the light source being designed to illuminate along this optical axis. The camera system may include an aperture, which is positioned between the camera (or digital camera) and the dewatered sludge or liquid. The camera is surrounded by the light source, which is also positioned between the aperture and the dewatered sludge or liquid. The aperture protects the camera and the light source from contamination or condensation. Ideally, the at least one light source provides homogeneous illumination along the optical axis, thereby reducing or eliminating the interference of ambient light.

[0015] The invention further relates to a computer program product comprising commands that cause the device according to the invention to execute the method according to the invention.

[0016] The invention will now be described using the drawings as an example. Fig. 1 shows images or sub-images of dewatered sludge in a dewatering device. Fig. 2a und 2b Show differently dewatered sludge in a dewatering device. Fig. 3a, 3b und 3c further examples of differently dewatered sludge in a dewatering device are shown. Fig. 4 shows an area in a dewatering device with dewatered sludge and formed interfaces. Fig. 5a und 5b shows an image or sub-images of dewatered sludge in a further dewatering device.

[0017] Fig. 1 Images 2 and sub-images 3 show dewatered sludge in a dewatering device. The sludge is dewatered in a sieve-type dewatering device with the addition of a flocculant, and images 2 of the dewatered sludge are taken using a camera system. Fig. 1 This illustrates the relationship between image 2 and sub-image 3, where in Fig. 1 Two images 2 are shown. Each image 2 is subdivided into sub-images 3 by the computer-implemented computational model, whereby in Fig. 1 The respective sub-images 3 are represented as squares within the respective image 2. Generally, the sub-images 3 should be as small as possible, but their size must be large enough to reproduce the surface texture, particularly the grain size, relief, and cracks of the dewatered mud. The trained computational model then classifies the sub-images, which in Fig. 1 This is indicated by differently colored sub-images 3 or squares. The distribution of the frequency of the classes in sub-images 3 can be used to assess the dewatering of the sludge.

[0018] Fig. 2a und 2b show differently dewatered sludge in a dewatering device, wherein Fig. 2a too low and Fig. 2b This reflects excessive dewatering or flocculant addition. The in Fig. 2a The insufficient addition of flocculant shown leads to smooth, reflective structures, whereas the in Fig. 2b The excessive addition of flocculant shown leads to coarse, matte textures. The trained computational model classifies the sub-images 3 and thus allows the assessment of the sludge dewatering. Unexpectedly, images of dewatered sludge that include an interface 4 of the dewatering device, in particular a free screen surface of a screen 6 on which the sludge is guided, for example in an edge region of the screen 6 or in the downstream area after a barrier 5 acting on a screen 6 that is in direct contact with the sludge, exhibit a particularly pronounced surface texture, especially grain size, relief, or cracks. The evaluation of these images is therefore advantageous and more informative. Fig. 2a und 2b The barriers 5, which act on a sieve 6, are clearly visible. Downstream of the barriers 5, the interfaces 4 and the sieve 6 are exposed, with the dewatered sludge in the vicinity of the interfaces 4 forming the particularly distinctive surface texture.

[0019] Fig. 3a, 3b und 3c show further examples of differently dewatered sludge in a dewatering device, wherein Fig. 3a too little Fig. 3b an optimal and Fig. 3c This indicates excessive dehydration or flocculant addition. Fig. 3a Insufficient addition of flocculant leads to smooth, reflective structures, lacking any distinctive surface texture, grain or relief. Fig. 3b Optimal addition of flocculant leads to the distinctive surface texture, grain size or relief, which in this example is accompanied by optimal drainage. Fig. 3c Adding too much flocculant leads to even coarser, matte textures, indicating an overdose. The images or training images are subdivided so that the sub-images or training sub-images are as small as possible, while the size of the training sub-images is chosen to be at least large enough to reproduce the surface texture, especially the grain size, relief, and cracks of the dewatered sludge. This instruction for the professional is well suited to Fig. 3a, 3b und 3c to understand. Since Fig. 3a Since it does not present a distinctive texture, the assessment of whether a subdivision was appropriately chosen cannot be based on the Fig. 3a This will take place. Fig. 3b und Fig. 3c In contrast, they exhibit a very distinctive texture, grain size, or relief. The size of the sub-image is chosen to be as small as possible, while still ensuring that the distinctive texture, grain size, relief, or crack is recognizable. Furthermore, it should be noted that images of dewatered sludge, which include an interface 4 of the dewatering device, in particular a free screen surface of a screen 6 on which the sludge is guided, for example, in the downstream area after a barrier 5 acting on a screen 6 that is in direct contact with the sludge, exhibit a particularly distinctive surface texture, in particular grain size, relief, or cracks. Fig. 3a, 3b und 3c The barriers 5, which act on the sieve 6, can be identified, whereby the interfaces 4 or the sieve 6 are exposed.

[0020] Fig. 4 Figure 1 shows an area in a dewatering device 1 with dewatered sludge and formed interfaces. The dewatered sludge is guided on a screen 6, with barriers 5 acting on the screen 6, which leads to the formation of distinctive surface textures in the downstream portion of the sludge after the barriers 5. Figures 2, which include an interface 4, thus allow for a particularly advantageous and informative evaluation of the partial figures 3.

[0021] Fig. 5a shows image 2 or Fig. 5b Figure 2 and sub-figures 3 show dewatered sludge. The dewatered sludge is transported by a screw conveyor. Fig. 5a The white frame indicates image 2 captured by the camera system. Fig. 5b Figure 2 shows sub-images 3, which are subdivided into sub-images 3 by the computational model. The sub-images 3 are classified by the trained computational model, and the class assigned to each sub-image is indicated by the coloring of the sub-image 3 or square 3. Advantageously, the classification includes a class for identifying invalid sub-images 3, which are not included in the assessment of the sludge dewatering. Sub-images are considered invalid if they do not depict dewatered sludge but instead show elements of the dewatering device. In Fig. 5a In image 2, for example, the screw conveyor can be clearly seen next to the dewatered sludge (see 5b).

[0022] The present invention offers numerous advantages. It allows for effective and objective control of the addition of a flocculant to a sludge, achieving optimal dewatering with minimal flocculant usage, which is both ecologically and economically relevant. The inventive method allows for quick and easy training of the computational model, making the method applicable to the dewatering of a wide variety of sludges. In particular, the division of the images into sub-images, with the evaluation of the sub-images being performed by the computer-implemented computational model, allows the computational model to be trained with a comparatively small number of training images. Furthermore, the evaluation of the sub-images is faster and less computationally intensive, with improved informative value, compared to evaluation based on individual images. This is achieved when the images are created in areas that, in addition to the dewatered sludge, also contain other materials.Since the dehydrated liquid also includes interfaces, the evaluation of the corresponding sub-images shows a further increased informative value. Reference sign

[0023] (1) Drainage device (2) Image (3) Partial image (4) Interface (5) Barrier (6) Sieve

Claims

1. Method for controlling the addition of a flocculant to a sludge, wherein the flocculant is added to the sludge and the sludge is subsequently at least partially dewatered in a dewatering device (1), for example a dewatering screw, a decanter, a wire dewatering device, etc., wherein images (2) of the dewatered sludge and / or of the liquid dewatered from the sludge are produced with a camera system, the images (2) are evaluated with a computer-implemented calculation model for image analysis, wherein the computer model has been previously trained with a training data set and the training data set comprises training images (2) of dewatered sludge and / or of the liquid dewatered from the sludge, characterised in that that the computer-implemented computational model for image analysis is formed as artificial neural network the training data set further comprises the training sub-images (3) formed from the individual training images (2) by a subdivision and the classification of the individual training sub-images (3), wherein for subdividing the training images (2) the size of the training sub-images (3) was reduced, the size of the training sub-images (3) being selected at least so large that the training sub-images (3) allow conclusions to be drawn about the surface texture, in particular the granulation, the relief, or the cracks, of the dewatered sludge or about the bubbles or foam formation of the dewatered liquid, the computational model dividing the individual images (2) into sub-images (3) analogously to the subdivision of the training images into training sub-images, classifying the sub-images (3) of the individual image (2), wherein the computational model classifies the sub-images (3) of the dewatered sludge according to a surface texture, in particular a granulation, a relief, or cracks, of the dewatered sludge and wherein the dewatering of the sludge is assessed on the basis of the classification of the sub-images (3) and the control of the flocculant dosage is effected on the basis of the assessment of the dewatering of the sludge.

2. Method according to claim 1, wherein the computational model classifies the sub-images (3) of the liquid dewatered from the sludge according to bubbles trapped in the liquid or a foam formed on the liquid.

3. Method according to claim 1 to 2, wherein the calculation model classifies the sub-images (3) according to colour properties, in particular a colour value, a colour saturation or a brightness value.

4. Method according to one of claims 1 to 3, wherein the classification of the sub-images (3) comprises at least two classes, a first class representing too little dewatering or flocculant dosage and a second class representing too much dewatering or flocculant dosage.

5. Method according to claim 4, wherein the classification of the sub-images (3) comprises at least a further third class, the third class being satisfactory dewatering or flocculant dosage.

6. Method according to claim 4 or 5, wherein a distribution of the frequency of the classes of the sub-images (3) of the image (2) is used to assess the dewatering of the sludge.

7. Method according to claim 4 to 6, wherein a further class detects invalid sub-images (3).

8. Method according to claim 1 to 7, wherein a desired dewatering or flocculant dosage is set to create the training data set and the training images (2) of the dewatered sludge and / or dewatered liquid are classified as representing satisfactory dewatering or flocculant dosage.

9. Method according to claim 1 to 8, wherein too little or too much dewatering or flocculant dosage is set to create the training data set and the training images (2) of the thus dewatered sludge and / or of the dewatered liquid are classified as representing too little or too much dewatering or flocculant dosage.

10. Method according to one of claims 1 to 9, wherein the images (2) of the dewatered sludge and / or of the dewatered liquid are made in an area comprising a boundary surface (4), in particular a free wire surface of a wire on which the sludge is guided, for example in an edge region of the wire or in the wake after a barrier (5) acting on a wire (6), a free surface of a dewatering or conveying screw or a wall which is in direct contact with the sludge or the dewatered liquid.

11. Dewatering device (1) for sludge, for example a dewatering screw, a decanter, a wire dewatering device, etc., for carrying out the method according to one of claims 1 to 10, comprising a camera system for producing images of the dewatered sludge and / or of the liquid dewatered from the sludge, means for controlling the addition of a flocculant to the sludge and a computer arranged to execute the computational model for image analysis according to the method of one of claims 1 to 10.

12. Device according to claim 11, wherein the camera system comprises a digital camera and illuminating means, the digital camera is associated with an optical axis and the illuminating means are configured for illumination in the direction of the optical axis.

13. A computer program product comprising instructions for causing the device of claim 11 or 12 to perform the method steps according to one of claims 1 to 10.