AI-controlled decanter

Abstract

The invention particularly relates to a computer-implemented method for optimizing the output of a decanter during operation using an enhanced artificial intelligence (AI) engine, the method comprising a. Operating the decanter according to a plurality of operating parameters; b. Treating the physical input comprising sludge and polymer by a decanter and producing a physical output comprising a centrifugal filtrate and a filter cake; c. Determining a plurality of substance parameters based on the physical output; d, transmitting the plurality of material parameters and the plurality of operation parameters to an enhanced AI engine; e. Determining, by the enhanced AI engine, a quality value for each of the plurality of substance parameters; f. Predicting, by the enhanced AI engine, a plurality of adjusted operating parameters; and further operating the decanter based on the plurality of adjusted operating parameters.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for optimizing the output of a decanter centrifuge, a related computer-implemented method for training a reinforcement / supervised learning artificial intelligence (AI) engine, a related trained reinforcement AI engine, an apparatus and a computer program. Background Art

[0002] Decanting centrifuges (also referred to herein as decanters) are widely used in the field of liquid treatment, in particular water treatment, such as, for example, wastewater reprocessing. Decanters are used to separate different phases and / or sediments contained in a fluid (e.g., a liquid substance such as sludge) from each other. For this purpose, a fluid can be filled into a rotating volume of a decanter, wherein the different phases / sediments can be separated from each other by means of centrifugal forces, which act differently on the different phases / sediments in the fluid, for example due to the different masses of the phases / sediments in the fluid. Thus, a phase / sediment with a higher mass can be displaced toward a radially outer region of the rotating volume of the decanter compared to a phase / sediment with a lower mass, while a phase / sediment with a lower mass can remain in a radial region closer to the central axis of the decanter, about which the rotation of the decanter is performed.

[0003] When operating a decanter in the field of wastewater treatment, it is envisaged that a purified output fluid (e.g., clean water, also referred to as centrate) is obtained by separating the suspended sediment from the water contained in the sludge and transporting the separated sediment away from the rotating decanter volume. The separated phase / sediment is generally referred to as a filter cake, which can be disposed of as waste. Since disposal is generally cost-intensive and the expense is generally proportional to the weight of the filter cake, it is preferred to maintain a low mass of the filter cake to be disposed of so that the disposal costs can be minimized. In addition, the higher the weight of the filter cake, the more complicated its transportation is, because more powerful transportation equipment is required. Therefore, the lighter the weight of the filter cake (depending on the moisture contained), the easier it is to transport. Therefore, it is desirable to reduce the mass and / or weight of the filter cake. This can be achieved at least in part by minimizing the amount of liquid (i.e., the moisture content in the filter cake), which can be achieved, for example, by setting the corresponding operating parameters of the decanter accordingly (e.g., the speed of the decanter bowl).

[0004] This adjustment is usually done manually by the operator of the decanter (and / or with the aid of a (manually set) PID controller). To this end, the operator usually needs to repeatedly collect samples of the filter cake to determine its moisture content and then may have to determine, based on the moisture content, which operating parameters should be used next, e.g. to further reduce the moisture content of the filter cake. However, due to the required labor effort, such a manual procedure for optimizing the moisture content of the filter cake can be cost-intensive and may lead to unsatisfactory results. The latter may lead to an even higher moisture content in the filter cake than expected and thus to a filter cake weight and / or mass that is too high. Therefore, a time-consuming iterative optimization by an operator procedure is required.

[0005] Furthermore, if the decanter is initially started up, it may take a certain amount of time (e.g., about 30 minutes or more) until a favorable range of optimal operating parameters is found and reached. This means that during this initial start-up time, the decanter is not operating under ideal operating conditions, potentially resulting in a filter cake having an undesirable moisture content, thereby unduly increasing the associated disposal costs.

[0006] If more than one parameter (e.g., filter cake thickness and water purity) is to be optimized, the situation becomes even more complicated. More specifically, in some cases, it may be desirable to minimize not only the moisture content of the filter cake. In some cases, it may also be seen that it is advantageous to also optimize, for example, the quality (e.g., the purity of the water purified in the decanter, also referred to as the centrate quality). These two aspects may result in an interaction between the moisture content of the filter cake and the centrate quality, making it necessary to adjust more than one operating parameter of the decanter to ensure that the desired optimized output of the decanter is obtained. This even further complicates the manual optimization procedure in terms of time effort and effectiveness.

[0007] The prior art only partially addresses these problems by suggesting the use of artificial intelligence based tools to determine the best operating parameters for the decanter and optimize the decanter output.

[0008] As an example, WO2019 / 150002A1 relates to a method for providing at least one input parameter of a sludge dewatering process of a wastewater treatment facility. The method comprises: obtaining data representing process and / or facility configuration data of the wastewater treatment facility; feeding at least a portion of the obtained data to at least one model, the at least one model being formed by at least historical process and facility configuration data collected from a plurality of wastewater treatment facilities in combination with the properties of chemicals applied in the wastewater treatment facility; and predicting at least one input parameter and / or at least one output parameter of the sludge dewatering process by means of at least one model for adjusting the sludge dewatering process of the wastewater treatment facility.

[0009] JP2021 / 102195 relates to a machine learning device for a centrifugal separation system, comprising: a learning data set storage unit, which stores a plurality of learning data sets including input data and output data, the input data including the slurry concentration of the treatment target liquid, the moisture content of the dehydrated solids, the concentration of the separation liquid, and the torque value of the screw conveyor, and the output data including control parameters of the centrifugal separation system, the control parameters being associated with the input data; a learning unit, which receives the plurality of learning data sets as inputs, and thereby learns a learning model that infers a correlation between the input data and the output data; and a learned model storage unit, which stores a learning model learned by the learning unit. The control parameters include at least one of a supply amount of an additive added to the treatment target liquid, a centrifugal force of a bowl, and a differential speed controlled by a differential speed generating device.

[0010] JP6994330B2 relates to a dehydration system, comprising: a dehydration device for generating suspended residue by removing liquid from a suspension; and an analyzer for storing a pre-built calculation model based on the following data, the following data including suspension data indicating the state of the suspension, operation data including a plurality of operation parameters of the dehydration device, and suspended residue data indicating the state of the suspended residue. The analyzer calculates a predicted value of the state of the suspended residue by inputting a current value of the suspension data and a current value of the operation parameter into the calculation model.

[0011] The AI-based methods known in the art do not provide the desired effectiveness for operating a decanter, i.e. the output of the decanter is not fully optimized in a satisfactory manner. In addition, these AI-based methods also lack flexibility, since it may not be easy to further train the corresponding AI model during operation. Therefore, there is a need for methods that provide improved optimization results and can be (further) optimized during operation. However, there is also a need to provide methods that achieve sufficient optimization results by requiring less input data at the same time, so that the complexity of the entire process is reduced. Summary of the invention

[0012] These shortcomings are at least partially addressed by one aspect of the present invention, namely a computer-implemented method for optimizing the output of a decanter during operation, which may use an enhanced artificial intelligence (AI) engine. The method may include a. operating the decanter according to a plurality of operating parameters, and may include b. processing a physical input that may include sludge and polymer by the decanter, and may include generating a physical output including a centrate and a filter cake. The method may also include c. determining a plurality of material parameters based on the physical output, and may include d. passing the plurality of material parameters and the plurality of operating parameters to the enhanced AI engine. The method may also include e. determining a quality value for each of the plurality of material parameters by the enhanced AI engine, and may include f. predicting a plurality of adjusted operating parameters by the enhanced AI engine. In addition, the method may include g. further operating the decanter based on the plurality of adjusted operating parameters.

[0013] The physical output may be understood as the substances originally contained in the physical input, which are separated from each other due to the operation of the decanter. The quality value may be understood as a numerical value indicating whether a corresponding substance parameter is within a predetermined acceptable parameter range.

[0014] The adjusted operating parameters may relate to operating parameters determined after it has been determined that previous operating parameters did not result in optimization of one or more of the plurality of material parameters.

[0015] By operating the decanter based at least in part on the above-mentioned computer-implemented method, the autonomous operation of the decanter can be promoted, because no or at least minimal input provided by humans is required to operate the decanter. The AI ​​control based on reinforcement of the decanter is particularly suitable for improving the AI-based control of the decanter, and can allow the decanter to be set to the optimal operating conditions more quickly (because the operator does not need a "trial and error method" method to find the optimal operating parameters of the decanter). This can at least contribute to the reduction of the disposal cost of the physical output of the decanter or even minimize the disposal cost. The autonomous operation of the decanter can further prevent the excessive resource consumption caused by the potential negligence of the operator (for example, if the operator does not quickly realize that a certain decanter is not operated under its optimized setting).

[0016] The plurality of operating parameters and the plurality of adjusted operating parameters may include one or more of: a differential speed of a vortex of the decanter, a speed of the decanter bowl, a feed flow of sludge; and / or a feed flow of polymer.

[0017] The speed of the decanter bowl may be understood as the rotational frequency (rpm) of the inner (process) volume of the decanter, in which the actual separation of the different phases (e.g., liquid and solid) and / or sediments may be performed. A low rotational frequency may result in correspondingly lower centrifugal forces (and a less pronounced separation between the different phases / sediments), and a high rotational frequency may induce higher centrifugal forces acting on the different phases / sediments contained in the inner volume compared to a low rotational frequency. As a result, heavier phases / sediments may accumulate at radially outer portions of the inner volume of the decanter bowl and may even leave the inner volume of the decanter bowl.

[0018] The decanter bowl may be at least partially surrounded by a scroll, wherein the scroll is adapted to discharge the separated phase / sediment that may have left the decanter bowl, such that the respective separated sediment can be processed from the decanter.

[0019] The differential speed of the decanter scroll can be understood as the absolute relative speed of the decanter scroll (e.g., in revolutions per minute (RPM)) and the decanter bowl (e.g., in RPM). A high differential speed of the scroll can be associated with a shorter retention time of the separated phases / sediment in the bowl (e.g., the period of time that the phases / sediment to be separated from each other may remain in the bowl) compared to a smaller differential speed at which the phases / sediment to be separated may have a higher retention time in the bowl.

[0020] Sludge can be understood as the process fluid that is injected into the decanter bowl. Sludge can include liquid (preferably water) and one or more different sediments that will be separated from the liquid by means of the decanter. In some cases, the sludge can be wastewater.

[0021] The feed flow of polymer may be understood as the rate (e.g. in [kg / s] or [l / s]) at which the polymer may be added to the sludge. The polymer may be adapted such that it supports the coagulation of sediment particles into larger and heavier compounds, so as to promote an improved separation of the formed compounds from other sediments and / or liquids.

[0022] This can advantageously aid in optimization of material parameters, which can be facilitated by adjusting operating parameters such that the material parameters converge toward preferred parameter ranges (eg, predetermined parameter ranges for certain material parameters that are considered to be optimized).

[0023] Determining the plurality of material parameters may further include: determining the dryness of the filter cake, the purity of the centrifugal filtrate and / or the dosage of the polymer in the centrifugal filtrate by one or more sensors.

[0024] The dryness of the filter cake can be understood as a measure of the liquid content of the filter cake. In other words, the dryness of the filter cake may be implicitly related to the moisture content of the filter cake. A high filter cake dryness may refer to a filter cake with a low moisture content compared to a low filter cake dryness with a correspondingly high moisture content.

[0025] The purity of the centrate can be understood as the cleanliness of the treated fluid (such as water).

[0026] The polymer dosage may refer to the amount of polymer added to the sludge.

[0027] Determining the plurality of material parameters allows for optimizing one or more of the plurality of material parameters based at least in part on the enhanced AI engine. This can facilitate closed-loop operation of the decanter by determining a plurality of operating parameters such that the plurality of material parameters can be optimized.

[0028] Determining the quality value of each of the multiple material parameters by the enhanced AI engine may further include: if the filter cake thickness is in the range of 10-19.99% dry matter (%DS), setting a low filter cake thickness quality value, or if the filter cake thickness is in the range of 20-35%DS, setting a high filter cake thickness quality value, and may include: if the purity of the centrifugal filtrate is in the range of 300-1000 nephelometric turbidity units (NTU), setting a low centrifugal filtrate purity quality value, or if the purity of the centrifugal filtrate is in the range of 50-299.99NTU, setting a high centrifugal filtrate purity quality value, and if the polymer dosage in the centrifugal filtrate is in the range of 2-9.99kg / tDS (kilograms per ton of dry matter), setting a high polymer dosage quality value, or if the polymer dosage in the centrifugal filtrate is in the range of 10-20kg / tDS, setting a low polymer dosage quality value.

[0029] A high filter cake quality value may indicate that the filter cake quality is acceptable or that the filter cake quality may be at least slightly reduced. A low filter cake quality value may indicate that the filter cake quality is insufficient and may therefore need to be increased. If the filter cake quality is preferably within 10-19.99% DS, more preferably within 12-18% DS, and most preferably between 14-16% DS, the filter cake quality value is considered low, while if the filter cake quality is within 20-35% DS, more preferably between 25-30% DS, and most preferably between 26-28% DS, the filter cake quality value is considered high.

[0030] The centrifugal filtrate purity quality value of high centrifugal filtrate can indicate that centrifugal filtrate purity is acceptable or can be at least slightly reduced. Low centrifugal filtrate purity quality value may indicate the need to increase centrifugal filtrate purity. Low centrifugal filtrate purity can be understood as the centrifugal filtrate with higher turbidity / more sediment compared with high centrifugal filtrate purity. If the centrifugal filtrate purity is preferably in 50-299.99NTU, more preferably between 100-250NTU, most preferably between 150-200NTU, then it can be considered that the centrifugal filtrate purity quality value is high, and if the centrifugal filtrate purity is preferably in 300-1000NTU, more preferably in 500-800NTU, most preferably in 600-700NTU, then it can be considered that the centrifugal filtrate purity quality value is low.

[0031] A high polymer dosage quality value may indicate that the polymer dosage is acceptable or that the polymer dosage may be slightly reduced. A low polymer dosage may be understood as providing less polymer to the sludge in the decanter centrifuge than with a high polymer dosage. If the polymer dosage is preferably within 2-9.99 kg / tDS, more preferably within 4-8 kg / tDS, and most preferably between 6-7 kg / tDS, the polymer dosage quality value may be considered high, while if the polymer dosage is preferably within 10-20 kg / tDS, more preferably within 12-18 kg / tDS, and most preferably within 14-16 kg / tDS, the polymer dosage quality value may be considered low.

[0032] By providing corresponding quality values ​​for filter cake thickness, centrate purity and polymer dosage, a quantitative estimation of the degree of optimization of material parameters associated with the output of the decanter can be facilitated. In addition, a quantitative metric can be supplied to the enhanced AI engine, based on which the predicted adjustments to the multiple operating parameters can be performed so that the multiple material parameters obtained subsequently converge towards the corresponding optimized values.

[0033] Predicting multiple adjusted operating parameters by the enhanced AI engine may further include determining the total energy consumption of the decanter based on the multiple operating parameters, and may include determining the optimized value of each of the differential speed of the decanter vortex, the speed of the decanter bowl, the feed flow of the sludge and the feed flow of the polymer by considering the balance between the dryness of the filter cake, the purity of the centrate, the total energy consumption of the decanter and the dosage of the polymer in the centrate.

[0034] The balance of the dryness of the filter cake, the purity of the centrate, the total energy consumption and the amount of polymer added can be understood as the interdependence of these parameters. In other words, as an example, a change in the dryness of the filter cake may also inevitably change the purity of the centrate and the total energy consumption of the decanter and the amount of polymer added.

[0035] Predicting the plurality of adjusted operating parameters may also facilitate convergence of one or more substance parameters toward optimal settings for the one or more substance parameters.

[0036] The determination of the optimization value may further include, if the filter cake quality value is high, reducing the optimization value of the feed flow of polymer, reducing at least one of the optimization value of the differential speed of the vortex of the decanter and the optimization value of the speed of the decanter bowl, and / or increasing the optimization value of the feed flow of sludge. The determination may further include, if the filter cake quality value is low, increasing the optimization value of the feed flow of polymer, increasing at least one of the optimization value of the differential speed of the vortex of the decanter and the optimization value of the speed of the decanter bowl, and / or reducing the optimization value of the feed flow of sludge.

[0037] If the filter cake density quality value is substantially high, at least two of the above-mentioned operating parameters of the decanter centrifuge may be adjusted such that the filter cake density may be reduced.

[0038] The reduction in the polymer feed flow may result in less coagulation of the sludge injected into the decanter bowl. As a result, coagulation of the sludge particles may be reduced, resulting in a sediment in the sludge having a lighter weight (which would be the case if the original polymer feed flow were considered). This may result in a reduced degree of separation of the phases / sediments included in the sludge. Thus, the filter cake thickness may be reduced and the filter cake may contain a higher moisture content compared to the exemplary case where the polymer feed flow is higher.

[0039] A reduction in the optimal value of the differential speed can result in a longer retention time of the phases / sediments to be separated. Due to the longer retention time, the time over which phase / sediment separation can occur is extended. This can result in an increased filter cake density and can comply with high filter cake density quality values.

[0040] A reduction in the speed of the decanter bowl may also reduce the differential speed of the decanter.

[0041] An increase in the feed flow of the sludge may reduce the time span over which the sludge passes through the decanter bowl. As a result, the time span over which the centrifugal force can act on the sediment in the sludge may be reduced. This may result in a less efficient separation of the phases / sediments from one another. As a result, the filter cake may contain a higher moisture content as it would have obtained if the feed flow of the sludge was higher.

[0042] Vice versa, the above-described adjustments of at least two operating parameters of the decanter centrifuge may also be applied in the opposite manner to increase the filter cake density.

[0043] By means of the above-described adjustment of at least two operating parameters of the decanter bowl, the optimization of the material parameters may be facilitated towards a desired optimization.

[0044] The determination of the optimized value may further include, if the centrate purity quality value is high, reducing at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or increasing the optimized value of the feed flow of the sludge. The determination may further include, if the centrate purity quality value is low, increasing at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or decreasing the optimized value of the feed flow of the sludge.

[0045] As described above, the reduction of the differential speed of the decanter can increase the retention time of the phase / sediment separated from the liquid of the sludge. Therefore, the degree of separation of the sediment from the liquid can be increased, which can lead to an increase in the purity of the centrate.

[0046] A reduction in the optimal value of the differential speed can result in a longer retention time of the phases / sediments to be separated. Due to the longer retention time, the time over which phase / sediment separation can occur is extended, which can result in an increase in centrate purity.

[0047] An increase in the feed flow of sludge can reduce the time span that the sludge passes through the decanter bowl. As a result, the time span that the centrifugal force can act on the sediment in the sludge can be reduced. This can result in a decrease in the purity of the centrate.

[0048] By means of the above-described adjustment of one or more operating parameters of the decanter centrifuge, the convergence of one or more material parameters towards respective optimized values ​​can be further supported.

[0049] Vice versa, the adjustment of the at least two operating parameters can also be applied to the purity quality of the reverse centrifugal filtrate.

[0050] The determination of the optimization value may further include that if the polymer dosage quality value is high, the optimization value of the feed flow of the polymer may be increased, and / or at least one of the optimization value of the differential speed of the vortex of the decanter and the optimization value of the speed of the decanter bowl may be increased, and / or the optimization value of the feed flow of the sludge may be reduced. If the polymer dosage quality value is low, the optimization value of the feed flow of the polymer may be reduced, and / or at least one of the optimization value of the differential speed of the vortex of the decanter and the optimization value of the speed of the decanter bowl may be reduced, and / or the optimization value of the optimization value of the feed flow of the sludge may be increased.

[0051] By means of the above-described adjustment of one or more operating parameters of the decanter centrifuge, the convergence of one or more material parameters towards corresponding optimal values ​​can advantageously be supported.

[0052] Predicting, by the enhanced AI engine, a plurality of adjusted operating parameters may further include deriving the plurality of adjusted operating parameters based on optimized values ​​for each of a differential speed of a decanter vortex, a speed of a decanter bowl, a feed flow of sludge, and a feed flow of polymer.

[0053] In some cases, adjustment of a particular operating parameter may affect two or more of the material parameters due to their possible interrelationships.

[0054] In some cases, a plurality of operating parameters may be derived based on two or three or all four of the above-mentioned optimized values ​​for each of the decanter scroll differential speed, the decanter bowl speed, the sludge feed flow, and the polymer feed flow.

[0055] The prediction may be based at least in part on the quality values ​​assigned to one or more of the plurality of material parameters.

[0056] By deriving multiple adjusted operating parameters at one or more of the operating parameters, interrelationships of the material parameters may be taken into account and an improved optimization of the material parameters may be obtained (e.g., the material parameters may be optimized such that ideally all material parameters converge towards desired optimal values ​​and may therefore not be limited to individual material parameters).

[0057] The computer-implemented method may also include adjusting, by an agent, the augmented AI engine based on the plurality of material parameters and the plurality of operational parameters.

[0058] The agent may be adapted to determine whether to increase or decrease one of the one or more operating parameters of the decanter based at least in part on the one or more quality values ​​to ensure that the decanter is operated so that a desired optimal value of the one or more material parameters is achieved.

[0059] The agent can be configured to adjust the reinforcement AI engine so that the sum of the assigned quality values ​​is maximized.

[0060] The use of an agent to tune the augmented AI engine may allow the AI ​​engine to be trained during operation of the decanter. Thus, the prediction of one or more operating parameters of the decanter may be continuously further improved during operation of the decanter. In other words, the AI ​​engine may be further trained during operation of the decanter and may thus be adapted to the real world operating conditions of the decanter and, for example, be limited to simulation-based training of the decanter. This may at least contribute to the accurate operation of the decanter at its optimized multiple operating parameters.

[0061] The computer-implemented method may also include performing steps b to g (as described above) more than once.

[0062] Steps b to g may be performed continuously, repeatedly and autonomously in a closed loop. Alternatively, it is possible that steps b to g are repeated once per minute, once per hour, once per day, etc. Additionally or alternatively, it is possible that steps b to g are performed only upon user input.

[0063] By performing this step more than once, closed loop operation of the decanter can be facilitated and can ensure that the decanter is stably operated at its optimized multiple operating parameters, thereby ensuring an optimized output of the decanter.

[0064] Another aspect of the present invention relates to a computer-implemented method for training an enhanced artificial intelligence (AI) engine that can be used to operate a decanter centrifuge. The method may include the following steps: a. simulating the operation of the decanter centrifuge to generate a plurality of operating parameters, and may include b. determining a plurality of material parameters of the simulated decanter centrifuge. In addition, the method may include c. passing the plurality of operating parameters and the plurality of material parameters to the enhanced AI engine, and may include d. adjusting the enhanced AI engine by the agent based on the plurality of material parameters and the plurality of operating parameters.

[0065] The plurality of operating parameters obtained from the simulation may be the same as the operating parameters described above with reference to the computer-implemented method for operating a decanter. The plurality of operating parameters obtained from the simulation may be understood as training data for strengthening the AI ​​engine.

[0066] The plurality of material parameters determined by simulating the decanter may be the same material parameters described above with reference to the computer-implemented method for operating the decanter. The plurality of material parameters obtained from the simulation may be used as training data for strengthening the AI ​​engine.

[0067] In some cases, only the simulated data may be passed to the augmented AI engine for the purpose of training the augmented AI engine. However, in some cases, data obtained from one or more physically operated decanter centrifuges (e.g., a plurality of operating parameters) may additionally or alternatively be passed to the augmented AI engine for training purposes.

[0068] By means of a computer-implemented method for training an enhanced AI engine, improved operation of a decanter centrifuge by means of, for example, a computer-implemented method may be facilitated. This may be facilitated based, at least in part, on simulated data pairs of a plurality of operating parameters and a plurality of material parameters (related to the operating parameters). Due to the possibility of obtaining training data from simulations, it may be facilitated to provide a larger training data set to the enhanced AI engine in general, and in a more cost-effective manner, just as it may be possible to obtain training data from a physically operated decanter centrifuge. This may allow for more accurate and faster training of the enhanced AI engine, and based thereon, more accurate and reliable predictions of a plurality of operating parameters by the enhanced AI engine when the trained enhanced AI engine is used for operation of the decanter centrifuge. This may support close operation of the decanter centrifuge at desired optimized operating parameters.

[0069] Another aspect of the present invention relates to an enhanced artificial intelligence (AI) engine that can be trained according to the above-described computer-implemented method for training an enhanced AI engine. The enhanced AI engine can be used in the method described above.

[0070] The enhanced AI engine may be located at one or more decanter centrifuges (e.g., in a corresponding processing unit). Additionally or alternatively, the AI ​​engine may be located in a control unit that communicates with at least one or more decanters. In some applications, the AI ​​engine may be located at a remote entity (e.g., a server (cloud), etc.).

[0071] The AI ​​engine may be adapted so that it can be used to control the operation of more than one decanter centrifuge.

[0072] By providing the above-mentioned AI engine, an AI engine for optimizing decanter operation can be provided, which can be separated from a specific decanter centrifuge and can be compatible with different decanter centrifuges. Therefore, a universal AI-based control for the operation of a decanter centrifuge can be provided. Further, if the AI ​​engine communicates with more than one decanter, the AI ​​engine can be provided with training data from more than one decanter centrifuge, which can therefore lead to an increase in the amount of training data, thereby improving the AI-based prediction accuracy of multiple operating parameters of the centrifuge, and the operation of the decanter centrifuge can be further improved.

[0073] Another aspect of the invention relates to an apparatus which may comprise means for performing the method as described above.

[0074] In some cases, the device can be an entity physically separate from the decanter. Alternatively, the device can be integrated into the decanter. The device can be a computer. In some cases, the device can be part of the decanter (e.g., as a microcontroller, integrated circuit (IC)), etc.

[0075] In some cases, the device may be a remote entity. The remote entity may be adapted so that it can communicate with the decanter by means of, for example, a network connection.

[0076] The device may be adapted so that it controls and / or trains only a single decanter. Alternatively, the device may control and / or train more than one decanter. In a further case, the device may be adapted to control a grid of multiple decanters.

[0077] By providing an apparatus comprising means for performing the method as described above, a control unit adapted for controlling and / or training a reinforcement-based AI engine may be facilitated.

[0078] Another aspect of the invention relates to a computer program which may comprise instructions which, when executed by a processing system, may cause the processing system to perform the method as described above.

[0079] The processing system may be located in the decanter. Alternatively, the processing system may be located in an entity physically separated from the decanter. In the latter case, the processing system may be a remote entity (e.g., a remote server operating a cloud system) that communicates with the decanter.

[0080] Another aspect of the present invention relates to a computer-implemented method for optimizing the output of a decanter centrifuge during operation using a supervised artificial intelligence (AI) engine, the method may include one or more of the following: a. operating the decanter centrifuge according to a plurality of operating parameters, b. processing a physical input including sludge and polymer by the decanter and producing a physical output including centrate and filter cake, c. determining a plurality of material parameters based on the physical output, and d. passing the plurality of material parameters and the plurality of operating parameters to the supervised learning AI engine. The method may also include one or more of the following: e. determining a quality value for each of the plurality of material parameters by the enhanced AI engine, f. determining a quality value for each of the plurality of material parameters by the supervised learning AI engine, and g. further operating the decanter based on the plurality of adjusted operating parameters.

[0081] The AI-based operation of the decanter based on a supervised learning AI engine may be considered as a second option (in addition to the implementation of a reinforcement-based AI engine) to improve the AI-based operation of the decanter.

[0082] The plurality of operating parameters and the plurality of adjusted operating parameters may include one or more of a differential speed of a scroll of the decanter, a speed of the decanter bowl, a feed flow of sludge and / or a feed flow of polymer.

[0083] Determining the plurality of material parameters may also include determining, by one or more sensors, the dryness of the filter cake, the purity of the centrifugal filtrate, and the dosage of the polymer in the centrifugal filtrate.

[0084] Determining the quality value of each of the multiple material parameters by the supervised learning AI engine may also include: if the filter cake thickness is in the range of 10-19.99% dry matter, setting a low filter cake thickness quality value, or if the filter cake thickness is in the range of 20-35% DS, setting a high filter cake thickness quality value, if the purity of the centrifugal filtrate is in the range of 300-1000 nephelometric turbidity units, setting a low centrifugal filtrate purity quality value, or if the purity of the centrifugal filtrate is in the range of 50-299.99NTU, setting a high polymer feed quality value, or if the polymer dosage in the centrifugal filtrate is in the range of 2-9.99kg / ton dry matter tDS, setting a low polymer dosage quality value, or if the polymer dosage in the centrifugal filtrate is in the range of 10-20kg / tDS, setting a low polymer dosage quality value.

[0085] Predicting multiple adjusted operating parameters by the supervised learning AI engine may further include: determining the total energy consumption of the decanter based on the multiple operating parameters, and determining the optimized value of each of the differential speed of the decanter vortex, the speed of the decanter bowl, the feed flow of the sludge and the feed flow of the polymer by considering the balance between the dryness of the filter cake, the purity of the centrate, the total energy consumption of the decanter and the dosage of the polymer in the centrate.

[0086] Determination of the optimization value may further include: if the filter cake thickness quality value is high, reducing the optimization value of the polymer feed flow, reducing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or increasing the optimization value of the sludge feed flow; if the filter cake thickness quality value is low, increasing the optimization value of the polymer feed flow, increasing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or reducing the optimization value of the sludge feed flow.

[0087] Determination of the optimization value may further include: if the centrate purity quality value is high, reducing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or increasing the optimization value of the feed flow of the sludge; if the centrate purity quality value is low, increasing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or reducing the optimization value of the feed flow of the sludge.

[0088] Determination of the optimization value may further include: if the polymer dosage quality value is high, increasing the optimization value of the polymer feed flow, and / or increasing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or reducing the optimization value of the sludge feed flow; if the polymer dosage quality value is low, reducing the optimization value of the polymer feed flow, and / or reducing at least one of the optimization value of the differential speed of the vortex member of the decanter and the optimization value of the speed of the decanter bowl, and / or increasing the optimization value of the optimization value of the sludge feed flow.

[0089] Predicting, by the supervised learning AI engine, a plurality of adjusted operating parameters may further include deriving the plurality of adjusted operating parameters based on optimized values ​​for each of a differential speed of a decanter vortex, a speed of a decanter bowl, a feed flow of sludge, and a feed flow of polymer.

[0090] In some cases, the foregoing steps b to g may be performed more than once.

[0091] Another aspect of the present invention relates to a computer-implemented method for training a supervised learning artificial intelligence (AI) engine that can be used to operate a decanter, the method may include the following steps: a. simulating the operation of the decanter to generate multiple operating parameters, b. determining multiple material parameters of the simulated decanter, c. passing the multiple operating parameters and the multiple material parameters to the supervised learning AI engine, and d. adjusting the supervised learning AI engine based on the multiple material parameters and the multiple operating parameters.

[0092] Another aspect of the present invention relates to a supervised learning artificial intelligence (AI) engine, which can be trained according to the above method and can be used for any of the above methods.

[0093] Another aspect of the invention relates to an apparatus which may comprise means for performing a method according to one of the aforementioned methods.

[0094] Another aspect of the invention relates to a computer program which may comprise instructions which, when executed by a processing system, may cause the processing system to perform a method according to one of the aforementioned methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] The various aspects of the present invention will be explained in more detail below with reference to the accompanying drawings. These drawings show:

[0096] Figure 1 : Illustration of an exemplary decanter centrifuge;

[0097] Figure 2 : A schematic diagram of a decanter and exemplary inputs and outputs;

[0098] Figures 3A to 3C: An illustration of a decanter and potential sensors for monitoring inputs and outputs associated with the decanter;

[0099] Figures 4A to 4B : Illustration of the training and prediction cycle of the AI ​​engine for the operation of the decanter;

[0100] Figures 5A to 5B : Illustration of the reinforcement training of the AI ​​engine for the operation of the decanter;

[0101] Figure 6 : Performance vs. time of a decanter in different operating modes

[0102] Figures 7A to 7B : A plot of the performance of the decanter versus time when operated manually and when operated based on a trained AI engine (supervised learning);

[0103] Figure 8 : A plot of the performance of the decanter versus time when operated manually and when operated based on a trained AI engine (reinforcement learning);

[0104] Fig. 9 : Illustration of an exemplary supervised learning infrastructure;

[0105] Fig.10 : Exemplary training data can be used to train the AI ​​engine. DETAILED DESCRIPTION

[0106] Embodiments and modifications of the present invention will be described in more detail below.

[0107] Figure 1 A diagram is shown of an exemplary decanter centrifuge 100 in accordance with aspects of the present invention.

[0108] The exemplary decanter centrifuge 100 includes a decanter bowl 110, which may include an inner volume 110a and an outer volume 110b (the latter surrounding the former). The inner volume 110a can be filled with sludge through an inlet port 120. Since the decanter bowl 110 is driven by a motor M to rotate about its longitudinal axis, sediment in the sludge can be separated from the liquid in the sludge. The sediment can be transported along the radial direction of the decanter bowl 110 toward the inner wall of the inner volume 110a of the decanter bowl 110. The separated sediment can leave the inner volume 110a through the outlet port 130 and enter the outer volume 110b. The inner volume 110a can be further surrounded by a vortex 140, which is spirally wound around the outer wall of the inner volume 110a. Scroll 140 may be adapted to transport sediment exiting interior volume 110a through outlet port 130 toward outlet port 150 from which the sediment may be discharged from decanter 100. Liquid in interior volume 110a may exit decanter bowl 110 via centrate outlet port 160.

[0109] More specifically, the decanter bowl 110 provides a predominantly cylindrical volume (e.g., having a longer extension along the longitudinal axis of the decanter than along a radial direction perpendicular to the longitudinal direction). The predominantly cylindrical volume may be concave toward at least one of its ends (e.g., toward the right-hand side of the decanter 100, as shown in FIG. Figure 1 The decanter bowl 110 may include an inner volume 110a and an outer volume 110b. The inner volume 110a may be adapted to contain sludge to be processed by the decanter 100. The outer volume 110b may be adapted to surround and contain the inner volume 110a.

[0110] The decanter bowl 110 may be adapted to be driven by a motor M to rotate about its longitudinal axis at a bowl speed (e.g., a rotational speed in revolutions per minute (RPM)). Due to the rotation of the decanter bowl 110, a centrifugal force may be generated that may act on the liquid and sediment in the interior volume 110a of the decanter bowl 110. Due to the centrifugal force F Z Can be scaled to

[0111] (Equation 1)

[0112] Wherein, m represents the mass of a certain sediment in the sludge, v represents the tangential velocity of the sediment due to the rotation of the decanter bowl 110, and r represents the radial distance of the sediment from the longitudinal center axis in the decanter bowl 110. The heavier the mass m of different sediments in the sludge, the greater the centrifugal force F. z The stronger the centrifugal force F Z, heavier sediments are pushed in a radial direction toward the inner wall of the interior volume 110a of the decanter bowl 110 , away from the longitudinal axis of the decanter 100 , while light sediments remain adjacent the longitudinal (center) axis of the decanter bowl 110 .

[0113] The decanter bowl 110 is in liquid communication with an inlet port 120. The inlet port 120 may be adapted to serve as a feedthrough for sludge to enter the interior volume 110a of the decanter bowl 110. In some exemplary embodiments, the inlet port 120 may be provided with a valve so that the feed flow of the sludge may be controlled (e.g., the feed flow of the sludge may be reduced when the valve is at least partially closed, and vice versa).

[0114] The wall surrounding the inner volume 110a of the decanter bowl 110 may further be provided with a sediment outlet port 130, which provides fluid communication between the inner volume 110a of the decanter bowl 110 and the outer volume 110b of the decanter bowl 110. Due to the separation of the sediment from the liquid of the sludge, the separated sediment may accumulate at the inner wall of the inner volume 110a of the decanter bowl 110, as described above. These sediments may leave the inner volume 110a through the outlet port 130 into the outer volume 110b of the decanter bowl 110.

[0115] The inner volume 110a may be surrounded by a scroll 140. The scroll 140 may be arranged to rotate about the inner volume 110a within the outer volume 110b of the decanter bowl 110. The scroll 140 may be provided with a blade that is spirally wound about the inner volume 110a along the longitudinal direction of the decanter bowl 110. The height of the blade may be equal to the radial distance between the outer wall of the inner volume 110a and the inner wall of the outer volume 110b. The scroll 140 may be adapted to rotate in the same direction as the decanter bowl 110. In other words, if the decanter bowl 110 is adapted to rotate clockwise, the scroll 140 is also adapted to rotate clockwise, and vice versa.

[0116] The swirl 140 may be adapted to transport sediment that has exited the interior volume 110a of the decanter bowl 110 in the direction of the inlet port 120 to the outlet port 150 from which the separated sediment may exit the decanter 100 and be transported to final waste disposal.

[0117] On the opposite end of the decanter 100 (relative to its longitudinal extension) from the inlet port 120, the internal volume 110a and the decanter 100 itself may be provided with a centrate outlet port 160. Since the sediment in the originally injected sludge is separated from the liquid contained in the sludge, the final treated sludge may mainly contain liquid. The liquid may leave the internal volume 110a and the decanter 100 as a substantially clean liquid by means of the centrate outlet port 160.

[0118] Figure 2 The decanter 200 is shown (which may be referred to above Figure 1 100 ) and exemplary inputs (which may relate to a product inlet 210a and a polymer input 210b to be fed to the decanter 200 ) and an illustration of exemplary outputs 220a and 220b to be obtained as output of the decanter 200 .

[0119] More specifically, the product inlet 210a may be as described above with reference to Figure 1 The product inlet 210a can be fed by a solids rate (e.g., the amount of solid particles in the flowing liquid, expressed in %DS or kg / h) and a density (e.g., expressed in mass / volume (e.g., g / cm 3 、kg / m 3 ,t / m 3 It is parameterized by the mass of the product inlet (denoted by , etc.).

[0120] The polymer input 210b can be parameterized by the amount of polymer added (e.g., in kg / tDS or l / h) and the associated amount of water (input into the decanter together with the polymer). The latter facilitates the dilution of the polymer and can therefore allow the polymer concentration in the decanter to be precisely adjusted according to the needs of the situation (e.g., desired material parameters).

[0121] A decanter output may be obtained as a result of providing a product inlet 210a and a polymer input 210b to the decanter 200, and subsequently processing the inputs within the decanter 200. The decanter output may include a centrate 220a and a filter cake 220b.

[0122] The centrate 220a may be understood as the residue of the sludge (input into the decanter 200) after the sediment contained in the sludge has been separated from the liquid contained in the sludge. In other words, the centrate 220a may be referred to as a clean liquid, which may generally include a negligible amount of sediment compared to the amount of sediment originally included in the sludge.

[0123] The centrate quality may be parameterized in units of NTU, which may be associated with the quality / purity of the liquid after treatment in the decanter 200. Furthermore, the centrate 220a output from the decanter 200 may preferably be recycled to the environment as clean (cleaned) / purified liquid.

[0124] The filter cake 200b may be understood as a sediment originally contained in the sludge as solid particles and separated from the liquid contained in the sludge. The filter cake 220b may be parameterized by a solids ratio (e.g., in units of DS%). The filter cake 200b may be understood as waste generated by purifying the sludge by means of the decanter 200b and may be disposed of (i.e., not intended for further use within the scope of the present invention).

[0125] The inputs and outputs as previously described may be provided regardless of whether the decanter 200 is operated manually or based on an AI engine. If the decanter 200 is operated manually, the operator may be required to collect samples of the centrate and / or filter cake, determine their quality and dryness, respectively, and infer potential operating settings for the decanter so that the output of the decanter 200 (particularly their respective material parameters) is optimized. However, if the decanter 200 is controlled by an AI engine, the decanter 200 may be controlled as described elsewhere herein. The AI ​​engine may be provided with multiple parameters associated with the decanter 200 to predict potential one or more operating parameters of the decanter 200, such as one or more of the following: the differential speed of the vortex member 140 in the decanter 200, the speed of the decanter bowl 110, the torque of the vortex member 140 of the decanter 200, information on the vibration of the liquid side of the decanter 200, information on the vibration of the solid side of the decanter 200, current information of the main motor M of the decanter 200, current information of the secondary motor of the decanter 200 (the above-mentioned is used to drive the vortex member 140), and / or bearing temperature (wherein the bearing can be arranged between the motor M and the decanter bowl 110).

[0126] FIG. 3A to FIG. 3C An exemplary arrangement of sensors located at or adjacent to a decanter is depicted ( Figure 3A ) and an exemplary sensor for determining a plurality of material parameters and / or a plurality of operating parameters ( Figure 3B and Figure 3C ).

[0127] The sensors may be configured to provide quantitative information of a plurality of material parameters and / or a plurality of operating parameters, as described elsewhere herein.

[0128] More specifically, Figure 3A A decanter 300 is shown (which may be referred to above). Figure 1 and Figure 2100 or 200) and exemplary locations of one or more sensors to provide, for example, the above quantitative information of various substances and inlets. It is worth noting that the exemplary decanter 100 or 200 can be provided with a feed control 310, a centrate monitoring component 320 and / or a filter cake monitoring component 330.

[0129] The decanter 300 may be provided with a feed control 310 located at or proximate to the inlet port 120, as described above with reference to Figure 1 The feed control 310 may include a valve to control the flux of sludge into the decanter bowl 110 (as described above with reference to Figure 1 The feed control 310 may also include one or more sensors to measure the sludge flux (e.g., in l / h, m 3 / h, kg / h, etc.).

[0130] A sensor suitable for monitoring the feed may be an infrared double scattered light sensor, wherein a light source (e.g. a laser light source) is arranged on one side of the pipe to be monitored of the decanter 300 and a corresponding light sensor is arranged on the opposite side of the pipe of the decanter 300, for example. The light emitted from the light source may impinge on one or more sediments in the injected sludge and may be scattered from one or more of the sediments. The corresponding light sensor may be adapted to detect the light scattered from the sediment. This may then facilitate, for example, the deduction of the amount of sediment particles in the injected sludge and their size distribution. This sensing method of the injected sludge may be referred to as turbidity measurement according to DIN EN ISO 7027 and DIN 38414.

[0131] Additionally or alternatively, the decanter 300 may be provided with a centrate monitor 320. The centrate monitor 320 may include one or more sensors in a tube located near the decanter 300 that conveys the treated liquid in the decanter 300 or in a chute connected to the decanter 300. Monitoring of the centrate quality may be based on an infrared dual scattered light sensor as described above with reference to the feed control 310.

[0132] Additionally or alternatively, the centrate quality can also be measured based on fluoroscopy. The method also relies on a light source (e.g., a laser light source) arranged on one side of the tube and a corresponding light sensor arranged at the radially opposite side of the tube, which light sensor can be adapted to monitor the optical power of the light emitted from the light source, which has penetrated the tube of the decanter 300 and has reached the corresponding light sensor, thereby facilitating a light barrier-like arrangement. Any residual particles that pass through the light barrier may cause a temporary drop in the optical power detected at the corresponding light sensor, from which the centrate quality (i.e., the amount of residual sediment) can then be derived from the light sensor from the power difference between the optical power emitted by the light source and the optical power detected at the corresponding light sensor.

[0133] The decanter 300 may also be provided with a filter cake monitoring means 330. The filter cake monitoring means 330 may, for example, comprise a sensor for determining the moisture content of the filter cake (e.g. the filter cake thickness). The measuring principle may be based on time domain reflectometry. Time domain reflectometry is based on the concept of inserting one or more electrodes into the filter cake and transmitting an electrical pulse along the one or more electrodes and which may be reflected at the end of the one or more electrodes. Since the pulse propagation velocity along the one or more electrodes may be moisture dependent (due to moisture dependent changes in the dielectric properties of the filter cake), the moisture content in the filter cake surrounding the one or more electrodes may be determined.

[0134] Figure 3B A sensor 340 is shown illustratively, which may be used to sense the amount of sediment in the sludge entering the decanter bowl 110 via the inlet port 120, and which may be based on the above referenced Figure 3A The concept of infrared double scattered light. The sensor 340 may include a light source 340a and a corresponding light sensor 340b.

[0135] Figure 3C A sensor 350 for measuring the moisture content in a filter cake is exemplarily shown. The sensor 350 may include a first electrode 350a along which an electric pulse may propagate (in the exemplarily shown illustration, in a top-to-bottom direction). The electric pulse may be reflected at the end of the first electrode 350a. As a result of the reflection, the electric pulse may propagate along a second electrode 350b. Since the propagation speed may depend on the moisture in the surrounding environment, the moisture content may be derived from the length of the time period required for the electric pulse to travel along the first electrode 350a and the second electrode 350b.

[0136] Some of the above-mentioned sensors may be located only in the decanter 300, while additionally or alternatively, some of the sensors may be located in the control cabinet.

[0137] Other sensors used in the decanter (not shown here) may be a magnetic induction sensor (MID) configured to measure the polymer concentration in the decanter and / or a variable frequency converter (VFC) adapted to measure the torque of the scroll and / or the motor.

[0138] Figure 4A The basic training concept of the AI ​​engine according to various aspects of the present invention is exemplarily shown. Figure 4B A decanter according to aspects of the present invention is shown ( Figure 4B )’s AI engine-based operating concept.

[0139] Figure 4AAn iterative training method 400 of an AI engine (e.g., represented as a model to be trained) is exemplarily shown. The training method 400 can be understood as a cyclic process that starts with providing an AI engine 410 to be trained based on a supervised learning concept with input data 420. The input data 420 may include a labeled training data set. The labeled training data set may include one or more parameters to control a decanter (the speed of a decanter bowl, etc.), and the label may indicate output data corresponding to an output product of the decanter (e.g., filter cake degree, etc.), which is provided by the decanter when the decanter is operated according to one or more parameters of the labeled training data set. After the training cycle, the output data 430 is compared with the label of the training data set in process step 440. The result of the comparison may be provided to the AI ​​engine 410 for training. If the result of the comparison indicates that the output data corresponds to the label, the adjustment of the AI ​​engine 410 may not be performed. However, if the result of the comparison indicates that the output data 430 deviates from the label, the model may be adjusted with the aim of generating output data 430 corresponding to the label of another training data set for training in a subsequent training cycle. As indicated by the dashed arrows, the training data may be altered or extended by further training data sets (eg, by an operator during training) to improve the training.

[0140] The training data may be based on previously prepared data sets (eg based on data obtained from a simulated decanter and / or based on data collected from one or more physically operated decanters), and the labeling of individual sets of training data may be provided by (a human operator).

[0141] The training data may include labeled data. Labeled data may be understood as a plurality of known material parameters that can be obtained by a plurality of known operating parameters. Labeling may be performed, for example, by an operator of the decanter.

[0142] In some cases, the input data 420 may be pre-processed (eg, the input data 420 may be scaled, eg, with the aid of a robust scaler).

[0143] In one example, the training data may include data pairs of one or more operating parameters for operating a decanter and one or more corresponding material parameters generated by setting the decanter to the one or more operating parameters. The input data 420 may also include a flag (e.g., assigned to one or more of the plurality of operating parameters and / or the plurality of material parameters) indicating whether certain operating parameters are considered positive (e.g., the corresponding operating parameter may result in a change in one or more of the material parameters toward a desired optimized value) or negative (e.g., the corresponding operating parameter may not result in a change in one or more of the material parameters toward a desired optimized value).

[0144] The training of the AI ​​engine 410 may be based on adjusting the internal training model of the AI ​​engine 410 so as to achieve a mathematical mapping of multiple operating parameters to multiple material parameters, i.e., the trained AI engine 410 may be able to predict one or more operating parameters of the decanter, which operating parameters may achieve one or more parameters of the multiple material parameters. This may be achieved, for example, by minimizing an error function.

[0145] As a result of successful training, the AI ​​engine 410 is able to derive multiple operating parameters of the decanter when one or more of the multiple material parameters are provided to the AI ​​engine 410, such that the multiple material parameters are optimized. When the value of the calculated error function is below a predefined threshold, it can be assumed that the training is successful.

[0146] The above-mentioned supervised learning concept at the AI ​​engine 410 may also include maximization of optionally assigned feedback values ​​(e.g., if feedback values ​​in the interval of 0 (worst) to 1 (best) are assigned similarly to quality values ​​in reinforcement learning, as will be referenced below). Figure 5A and Figure 5B Further described). In some embodiments, the assigned feedback values ​​may include an aggregation of individual quality values ​​assigned for filter cake thickness, centrate quality, and polymer dosage. This aggregation may be understood as a quantitative measure for determining how tightly the decanter operates at its optimized settings. This may particularly allow monitoring of the operating costs of the decanter, such as the total cost of ownership (TCO costs). In general, it may be beneficial to collect as many parameter values ​​of multiple material parameters and multiple operating parameters as possible to obtain an accurate mapping of the current decanter operation. An accurate mapping of the decanter may advantageously contribute to the cost-effective operation of the decanter.

[0147] After providing one or more operating parameters (output data 430) to the decanter, feedback 440 can be determined, which indicates whether the predicted one or more operating parameters lead to an improvement in one or more material parameters (i.e., whether the current value of one or more material parameters is closer to the expected optimal value of one or more material parameters as in the previous iteration step). In some alternative cases, the predicted output data 430 can also be used for operating the decanter for training purposes instead of directly using the predicted output data 430, but manually assigning feedback 440, such as by an operator of the decanter. In some cases, the feedback can be provided as an error function. The error function can be configured to calculate, for one or more of the multiple operating parameters, the error between one or more of the multiple material parameters included in the training data set (assigned to the corresponding one or more of the multiple operating parameters) and the one or more of the predicted multiple material parameters, which can be obtained if the decanter is operated according to the predicted one or more of the predicted multiple operating parameters. The training can be configured so that the error function (also commonly referred to as an error or error metric) is minimized. The error function may be based on, for example, mean square error (MSE), its root variance (RMSE), mean absolute error (MAE), percentage variance (MAPE), Hilbert loss, and / or other suitable metrics and / or combinations thereof. The AI ​​engine model may be based on a random forest regressor alone or on a combination of a random forest regression with an adaptive boosting regressor, a kNeighbors regressor, a Gaussian naive Bayes, or a logistic regression. In the case of using multiple regressions, stacking regressions may be required.

[0148] The predicted output 430 may be fed back to the AI ​​engine 410 to be trained, and the determined feedback 440 may be fed back to the AI ​​engine 410 and optionally combined with the predicted output 430 to further train the AI ​​engine 410 in subsequent iteration steps of the training method 400 .

[0149] In some embodiments, a data pair including a parameter in a plurality of material parameters and a parameter in a plurality of operating parameters may be defined. Such a data pair may be referred to as a regressor and may be trained according to the aforementioned supervised learning method. If the first regressor is successfully trained, the second regressor may be trained in the same manner. Once at least two regressors have been trained, a combination of at least two regressors may be trained. As an example, supervised learning at the AI ​​engine 410 may obtain one or more parameters in a plurality of operating parameters (e.g., included in the input data set 420), including a decanter differential speed, a decanter bowl speed, a decanter vortex member torque, a liquid side vibration, a solid side vibration, a main motor current, a secondary motor current, a bearing temperature, a feed flow, a feed dry matter, and / or a polymer flow. As labeled data (a plurality of material parameters), one or more of a filter cake thickness, a polymer dosage, and / or a centrifuge filtrate quality may be provided to the AI ​​engine 410.

[0150] During training, the AI ​​engine 410 may adjust its internal algorithm so as to facilitate the mapping of multiple operating parameters. In other words, if a set of one or more of the multiple operating parameters is provided to the AI ​​engine 410, the AI ​​engine 410 may be able to reproduce the associated multiple material parameters. During training, the AI ​​engine 410 may predict certain multiple material parameters (output data 430) and may feed it to the feedback 440, which may calculate the current deviation of the prediction from the labeled data. The feedback 440 may be provided to the AI ​​engine 410. The AI ​​engine 410 may predict subsequent multiple material parameters (output data 430) in subsequent iteration steps of the training method 400, which are again evaluated by the feedback 440. The process may be iterated until the difference between the predicted multiple material parameters and the labeled data is below a predetermined threshold.

[0151] Figure 4B The operation method 450 of the decanter based on the trained AI engine 470 is exemplarily shown. The operation method can preferably be performed iteratively and can be the same as the above reference Figure 4A The described training method 400 is the same, however, the AI ​​engine 470 may not be further trained in some cases and may only be used to predict output data 480 (even though further training of the AI ​​engine 470 is not excluded).

[0152] The operating method 450 can be based on an input data set 460 and preferably includes a plurality of material parameters that can be obtained by analyzing (e.g., with the aid of a filter cake monitor 330 and / or a centrate monitor 320) the output of the decanter. In addition, the input data set 460 can include a plurality of operating parameters (output data 480).

[0153] The input data set 460 is optimized by the trained AI engine 470. The trained AI engine 470 (e.g., as described above with reference to Figure 4A The trained AI engine described above, or any other suitable AI engine or any suitable AI engine) can be provided with an input data set 470 and can predict an output data set 480 based on its training. In some cases, the input data set 460 can be pre-processed before being provided to the trained AI engine 470.

[0154] The output data set 480 may include a plurality of operating parameters to be provided to the decanter. If the decanter is operated according to the plurality of operating parameters, the decanter may generate an output (e.g., centrate 220a and / or filter cake 220b and associated plurality of material parameters) that may be analyzed to determine whether the output data set 480 predicted by the trained AI engine 470 has resulted in optimization of the plurality of material parameters.

[0155] If the decanter is operated according to the predicted output data 480, the output produced by the decanter may result in a change in the input data 490. This new input data 490 may then be used as input data 460 and may be provided to the trained AI engine 470 to obtain new output data 480 in a subsequent iterative step.

[0156] Figure 5A and Figure 5B A training method 500 for an AI engine 510 based on reinforcement learning according to one aspect of the present invention is shown. Figure 5A ) is a diagram of another exemplary embodiment, and as Figure 5A Detailed illustration of an exemplary agent interacting with an environment as part of the reinforcement learning concept shown ( Figure 5B ).

[0157] like Figure 5A The training method 500 shown can generally be used in conjunction with the above-referenced Figure 4A The described training method 400 (supervised learning) is implemented identically, except that the actual training algorithm used to train the AI ​​engine 510 is based on reinforcement learning, which will be described in further detail below.

[0158] Figure 5A A reinforcement training method 500 is shown. The training method 500 may include an input data set 520, which may include training data. The training data may include a plurality of operating parameters for a decanter and, optionally, a plurality of material parameters associated with the plurality of operating parameters. In addition, the training data may include one or more quality values, which may be assigned to each of the plurality of operating parameters and will be further described below.

[0159] An input data set 520 may be provided to an AI engine 510 to be trained. Based on the input data set 520, i.e., a plurality of operating parameters (and corresponding quality values), the AI ​​engine 510 may be adapted (e.g., based on a reinforcement learning algorithm) such that, when presented with one or more parameter values ​​of a plurality of material parameters, the AI ​​engine 510 may predict one or more parameter values ​​of the plurality of operating parameters (i.e., may determine new plurality of operating parameters for the decanter), with the goal of ideally improving one or more parameters included in the plurality of material parameters. Improvement in this case may be understood as one or more parameter values ​​of the plurality of material parameters in the current iteration of the method 500 being closer to the desired optimized values ​​of the one or more parameter values ​​of the plurality of material parameters than the one or more parameter values ​​of the plurality of material parameters in the previous iteration step of the method 500. In other words, the plurality of operating parameters may be predicted such that, when the decanter is operated according to the predicted plurality of operating parameters, the plurality of material parameters may converge towards the desired optimized values. The training of the AI ​​engine 510 may include interaction with an intelligent agent, as will be described below with reference to Figure 5B Further described.

[0160] Based on the initial training step of the AI ​​engine 510 and the input data 520 provided to the AI ​​engine 510, the AI ​​engine 510 can predict an output data set 530. The output data set 530 can include a plurality of operating parameters that can be provided to the decanter. Based thereon, the decanter can generate an output (as described above) from which a new, subsequent plurality of material parameters can be derived. The output data set 530 can, for example, include one or more of the differential speeds of the scroll.

[0161] In subsequent method steps of method 500, corresponding quality values ​​540 may be assigned to multiple operating parameters based on the obtained multiple material parameters (i.e., in a preferred embodiment, corresponding quality values ​​540 may be assigned to each parameter value of the multiple operating parameters according to one or more parameter values ​​of the multiple material parameters).

[0162] The quality value may indicate whether one or more parameter values ​​of the plurality of operating parameters may be associated with an improvement of one or more parameter values ​​of the plurality of material parameters toward the corresponding desired optimized material parameters, or whether one or more parameter values ​​of the plurality of operating parameters may be associated with a degradation of one or more parameter values ​​of the plurality of material parameters (e.g., whether the predicted plurality of operating parameters have resulted in the plurality of material parameters being further away from the corresponding optimized material parameters as compared to the previous iteration of the method 500), or one or more parameter values ​​of the plurality of material parameters may be scored as neutral (e.g., the one or more parameter values ​​of the plurality of material parameters are neither considered an improvement nor a degradation as compared to the previous iteration of the method 500). In some cases, the quality value may be passed to an agent of the AI ​​engine 510. The agent may be further described with reference to the agent model 550 below, and the agent may be adjusted such that a highest priority is to achieve a relatively high filter cake density (e.g., a filter cake density of more than 20% DS). If an optimized cake density is achieved, the AI ​​engine 510 can attempt to optimize the centrate quality (e.g., a value below 300 NTU), followed by the goal of keeping the polymer dosage low (e.g., the polymer dosage is less than 10 kg / tDS) and minimizing the energy consumption of the decanter. Considering the energy consumption (e.g., if energy consumption can be a priority), for example, by reducing the cake density by 0.1% to 1.2%, preferably 0.2% to 0.8%, the energy consumption of the decanter can be reduced by, for example, 25% to 35%, preferably 30%. This reduction in cake density can be achieved, for example, by reducing the bowl speed of the decanter.

[0163] The assigned quality value 540 (optionally, the output data 530 and a plurality of material parameters obtained by providing the output data 530 to the decanter) can be incorporated as input data 520 for a subsequent iterative step of the method 500. Based on this new subsequent input data set 520, a subsequent training step of the AI ​​engine 510 can be performed.

[0164] Through reinforcement learning, the AI ​​engine 510 will be more efficient with respect to the assigned quality values, because it learns to maximize the sum of all assigned quality values. In addition, reinforcement learning (compared to supervised learning) provides greater flexibility with respect to the training data used. As an example, supervised learning requires data pairs of operating parameters and material parameters, while reinforcement learning may only require operating parameters. Therefore, continuous further training of the reinforcement AI engine 510 can be facilitated even when the decanter is in the operating phase.

[0165] Furthermore, in this regard, the training data used to train the reinforcement-based AI engine 510 may not require any labeling.

[0166] In some cases, the AI ​​engine 510 may include a neural network that includes one or more layers, and each layer includes one or more nodes. The neural network may be a feedforward neural network or a convolutional neural network or any combination thereof. The training of the AI ​​engine 510 may then include determining one or more weighting coefficients associated with the connection of nodes in two adjacent layers.

[0167] Training can be further based on a Dual Deep Q-Network (DDQN) or Deep Deterministic Multi-Gradient (DDPG) architecture.

[0168] During training, training data (e.g., including inflow, time steps defining the time) is provided to the AI ​​engine 510. In addition, control parameters for time step t+1 (e.g., speed of the decanter bowl, differential speed, etc.) can be part of the training data. The model then predicts the remaining parameters for time step t+1 that are not part of the training data. The DDQN-based agent can provide all possible actions (e.g., increase or decrease the speed of the decanter bowl) as output, from which the action with the highest value is selected. The DDPG-based agent outputs a magnitude for each parameter. In particular, t represents the time step.

[0169] Figure 5B A diagram of an exemplary interaction method 550 of an agent 560 and an environment 570 is shown to further illustrate the concept of the reinforcement training method 500, and in particular the effect of the training method 500 on the AI ​​engine 510 (described above with reference to Figure 5A describes the latter). Figure 5B and the corresponding description use terminology commonly used in the context of reinforcement learning in the art, however, unlike the above reference Figure 5A No new entities are introduced compared to the description provided.

[0170] Agent 560 may be constructed by AI engine 510, as described above with reference to Figure 5A The agent 560 may provide an action A to the environment 570 in which the agent 560 is located.

[0171] Action A can be compared with the above reference Figure 5A The output data 530 described is the same.

[0172] Environment 570 may include one or more decanters that may operate according to action A. Thus, environment 570 may provide state information S (which may be similar to that described above with reference to Figure 5A The same operating parameters as described above) and the quality value R (which can be compared with the above reference Figure 5A The quality value R and the state information S may be understood to be included in the input data set 520, as described above with reference to Figure 5A As described.

[0173] The state information S and the quality value R may be provided to the agent 560 , and the agent 560 may predict new actions A in subsequent iteration steps of the interaction method 550 based on the state information S and the quality value R.

[0174] In the following, the interaction between the assignment of a quality value R and the prediction of an action A, which is predicted by the agent 560 at least partially based on the assigned quality value R and the state information S, is described.

[0175] In the following, exemplary assignments of quality values ​​R to a plurality of operating parameters are described based on one or more of the parameter values ​​of the substance parameters. As outlined elsewhere herein, the plurality of substance parameters may include polymer dosage, filter cake thickness, and centrate quality.

[0176] In general, the assignment of the quality value R may be configured such that a quality value R between 0 and 1 may be assigned (however, any other suitable interval may also be applicable). As an example, if one or more parameter values ​​of the plurality of predicted operating parameters have resulted in an improvement in one or more parameter values ​​of the plurality of material parameters, a quality value of 1 may be assigned. The quality value R may be assigned continuously within the interval between 0 and 1.

[0177] The quality value R can be assigned relative to the centrate quality, so that when the centrate quality is in the range of 150 to 250, if the centrate quality is 200, the maximum quality value R can be assigned. In this case, the maximum quality value can be assigned to be 1. This can be understood as indicating that the desired optimized centrate quality is 200. However, if the centrate quality is 225, the assigned quality value R can be 0.5. This can be understood as, relative to the centrate quality of 200, it is not expected that a higher centrate quality value (i.e., a centrate with a lower purity is substantially). This may be due to the fact that a higher centrate quality value may also affect, for example, the cake thickness and / or the polymer dosage, and therefore may only result in partial optimization of multiple material parameters during the operation of the decanter (e.g., with respect to the centrate quality), while degradation of material parameters may occur. If the centrate quality is 150 or lower, or 250 or higher, the assigned quality value may be 0, which may indicate that these material parameters are to be avoided.

[0178] A similar assignment of quality values ​​R may apply to the polymer dosage. The polymer dosage may be in the range of 8 kg / tDS to 12 kg / tDS. In this embodiment, if the polymer dosage is 10 kg / tDS (the desired optimal value), a quality value of 1 may be assigned, whereas if the polymer dosage is 9 kg / tDS or less, or 11 kg / tDS or more, a lower quality value such as 0.5 may be assigned. If the polymer dosage is 8 kg / tDS or less, or if the polymer dosage is 12 kg / tDS or more, an even lower quality value R of 0 may be assigned.

[0179] A similar assignment may be performed with respect to the cake thickness. In one example, the maximum (theoretical) cake thickness may be 25% DS (however, this exemplary theoretical assumption should not be understood as a limitation on the empirically achievable cake thickness, which may exceed the theoretical dryness and may be as high as 35%, and in some exemplary cases, the maximum (theoretical) cake thickness may also be selected to be as high as 35%). In this case, a quality value of 1 may be assigned. If the cake thickness is 20% DS or less, or if the cake thickness is 30% DS or more, a quality value of 0.5 may be assigned. Also in this regard, if the cake thickness is 15% DS or less, or if the cake thickness is 35% or more, a quality value of 0 may be assigned.

[0180] It should be understood that the above-mentioned distribution of quality values ​​to multiple material parameters is only understood as exemplary, and any other suitable distribution is also possible. More importantly, the agent 560 may also not be suitable for maximizing the sum of the allocated quality values ​​R, but the agent 560 is suitable for minimizing the sum of the allocated quality values ​​R.

[0181] The agent may be adapted so that it attempts to maximize the sum of the individual quality values ​​R assigned to each of the plurality of material parameters. More specifically and as an example, based on assigning a corresponding quality value R to each individual parameter included in the plurality of operating parameters (action A), the agent 560 may be informed of which operating parameter values ​​are considered an improvement in the operation of the decanter (i.e., one or more parameters included in the plurality of material parameters may be closer to the desired optimized parameter value), and which operating parameter values ​​are considered a deterioration in the operation of the decanter (e.g., one or more parameters included in the plurality of material parameters may be further away from the desired optimized parameter value). Based on this, the agent 560 may establish a winning function, which may include a calculation of the sum of the assigned quality values, with the goal of optimizing the sum. Based on the goal of maximizing the sum of the assigned quality values, the agent 560 may predict one or more of the plurality of operating parameters (action A), such that based on the experience gained by the agent 560 in a previous iteration of the method 550, the determined one or more of the plurality of operating parameters (when provided to the decanter) may result in an improvement in the plurality of material parameters.

[0182] Figure 6 Depicted qualitatively illustrates a decanter (e.g., as described above with reference to Figure 1 and Figure 2 A graph 600 of exemplary performance of the device 100 of the invention is shown with respect to time.

[0183] The performance of a decanter may be understood as its ability to separate sediment contained in the sludge from liquid contained in the sludge. In this case, a high performance may be understood as a relatively high degree of separation (e.g., almost all sediment is separated from liquid in the sludge) compared to a low performance with a corresponding relatively low degree of separation (e.g., almost no sediment is separated from liquid in the sludge).

[0184] More specifically, Figure 6 The performance of an exemplary decanter with respect to time is shown while manually operated in the manual operation section 610 , after stabilizing the decanter operation in the stabilization section 620 , and after optimizing the decanter in the optimization section 630 .

[0185] More specifically, when the decanter is started, a plurality of operating parameters may be initially set in the manual operation section 610 by the operator of the decanter, for example, based on experience. After the initial start-up of the decanter, the operator may, for example, regularly collect samples of the output of the decanter (e.g., centrate 220a, filter cake 220b), may analyze the output (e.g., to determine the corresponding plurality of material parameters, such as filter cake thickness, centrate quality, polymer dosage, etc.), and may derive therefrom a new plurality of adjusted operating parameters of the decanter, so that the desired optimized values ​​of the material parameters may be obtained. Since the concept is essentially based on a "trial and error method" - trying to find a plurality of operating parameters, which may result in the optimization of a plurality of material parameters, the performance of the decanter achieved by means of manual operation may fluctuate around a certain average performance baseline M.

[0186] In the potential (subsequent) stable segment 620, a number of operating parameters for the decanter may have been found, and the decanter may have been set to these operating parameters, which may have resulted in a reduction in the peak-to-peak fluctuations of the performance versus time evolution in the graph 600. However, the reduced peak-to-peak fluctuations may still fluctuate around the same average performance baseline M (as described above with reference to the manual operation segment 610), which may be lower than the desired optimized value O of the decanter performance. In other words, even though the peak-to-peak fluctuations may have been reduced in the stable segment 620, the desired optimized performance value O has not yet been reached.

[0187] The optimization section 630 shows how the performance of the decanter evolves with respect to time if a number of operating parameters are set to their optimal values. In this case, the peak-to-peak fluctuations in performance with respect to time can be minimized, and the mean value where the fluctuations occur can be shifted to an optimized performance value O that is only slightly below the expected performance limit L.

[0188] One aspect of the present invention is to reach the optimized zone 630 shortly after initial startup of the decanter so that the period of time during which the decanter is not operating at its optimized performance level is kept as short as possible.

[0189] Fig. 7A and Figure 7B shows the manual operation of the decanter ( Fig. 7A ) and trained AI engines (based on supervised learning training, Figure 7B )'s operation of a decanter showing exemplary performance of a decanter versus time.

[0190] like Fig. 7AAs shown, when manually operated, the performance of the decanter fluctuates significantly around a certain average performance value M. Each position where the slope of the performance curve changes suddenly can be associated with a manual readjustment of one or more of the multiple operating parameters of the decanter. This can be based on, for example, sample collection (e.g., every 45 minutes) performed by an operator and subsequent readjustment of one or more parameters included in the multiple operating parameters. Due to the fluctuation of performance relative to time, the decanter may not be able to provide a constant and effective separation of sediment in the sludge from the water in the sludge, thereby producing a corresponding adverse effect on the centrate quality (fluctuating centrate quality and / or overall poor centrate quality) and / or filter cake density (e.g., the filter cake may be too wet, thereby increasing the waste disposal cost of the filter cake), etc.

[0191] In contrast, Figure 7B The performance of a decanter versus time is shown when the decanter is operated based on a trained AI engine, where the trained AI engine is trained based on supervised learning.

[0192] It can be seen that the peak-to-peak fluctuation of the performance curve is Fig. 7A The performance curve shown in Figure 2 is significantly lower than that in Figure 2. In addition, the mean performance baseline around which the performance fluctuates has been Fig. 7A Thus, when the decanter is operated based on a trained AI engine, the decanter can not only provide more constant performance over time, but can also provide generally higher performance. It should also be noted that, respectively, Fig. 7A and Figure 7B The initial ramp-up slope depicted in each of the graphs shown in can arise from the initial ramping of the decanter bowl 110 and / or the scroll 140 from a stationary state (e.g., a state in which the decanter is closed) to an operational state (e.g., a state in which sludge processing can occur, such as, for example, separation of, for example, centrate 220a and filter cake 220b). During this ramp-up process, the speed and / or differential velocity of the decanter bowl can be increased to a desired value, wherein the separation capacity of the decanter can increase with the increase in the speed and / or differential velocity of the decanter bowl, and can preferably peak at its desired (average) efficiency.

[0193] Figure 8 An exemplary simulated performance of a decanter with respect to time is shown if the decanter is operated based on a trained enhanced AI engine.

[0194] from Figure 8 As can be seen from the figure, if the enhanced AI engine is used to operate the decanter, the peak-to-peak fluctuation of the decanter is similar to Fig. 7A In addition, the operation of the decanter based on the trained enhanced AI engine further shows fluctuations around the average performance baseline M, which is higher than Fig. 7A Average performance baseline M (decanter operated in manual mode).

[0195] Fig. 9 A schematic diagram of an exemplary system 900 (infrastructure) that can be used to operate a decanter is shown. System 900 can include an environment 910 that can communicate with a human-machine interface 920, an IoT edge device 930, and a cloud service 940. Cloud service 940 can communicate with a computer 950, and the computer can communicate with the human-machine interface 920.

[0196] Environment 910 can be combined with environment 570 (as referenced above) Figure 5B The environment 910 may be the same as described above and may include one or more decanters. The environment 910 may be in communication with a human machine interface (HMI) 920 (wireless and / or wired).

[0197] HMI 920 may be physically separate from environment 910, or may be part of environment 910 (eg, part of a decanter as a display). HMI 920 may provide components for displaying decanter-specific settings to an operator and receiving commands from the operator.

[0198] HMI 920 can communicate with IoT edge device 930. IoT edge device 930 can be configured to provide access to a network and / or the Internet. Based on this, IoT edge device 930 can communicate with cloud service 940.

[0199] The cloud service 940 may be accessed via a (local) network and / or via the Internet. The cloud service 940 may implement, at least in part (based on supervised and / or reinforcement learning), a trained AI engine such that a plurality of material parameters may be provided to the AI ​​engine, which may then predict a plurality of operating parameters for one or more decanters located in the environment 910. In some cases, the cloud service 940 may be implemented as a database storing a plurality of material parameters and a plurality of operating parameters (and optionally feedback and / or quality values), such that a potential training data set is generated when operating one or more decanters in the environment 910. The training data set may, for example, be used to train an AI engine that may be used for the operation of one or more other decanters.

[0200] The cloud service 940 can also communicate with a computer 950. The computer 950 can be used to monitor and control the prediction of multiple operating parameters of one or more decanters. In addition, the computer 950 can communicate with the HMI 920.

[0201] Fig.10 An exemplary training data set that can be used to train an AI engine is shown, as described elsewhere herein. The training data set can be particularly suitable for reinforcement learning. Fig.10Each column shown in the training data set may relate to a particular operating parameter among a plurality of operating parameters that may be particularly relevant to the operation of a decanter and suitable for training. Fig.10 Each of the rows shown in the training data set may relate to a distinct set of operating parameters determined simultaneously, while the operating parameters in different rows may have been determined at different times.

[0202] An exemplary training data set may include one or more of the following: bearing liquid temperature (bearing_liquid_temp), bowl speed of the decanter (Bowl_Speed), XX of the inlet (TS_Inlet), differential speed of the decanter (Diff_Speed), TRB of the centrate (TRB_centrate), current of the motor (current), inlet temperature of the solid (Ti_Solid), bearing solid temperature (Bearing_Solid_Temp), temperature of the solid (TS_Solid, typically in the range of 15°C-45°C, preferably between 20°C-40°C, depending on process and weather conditions), feed flow (feed_flow), torque of the scroll (Torque, measured in % and typically can be between 40%-90%, preferably between 45%-85% depending on the process conditions of each wastewater treatment plant), bowl speed setting (Bowl_Speed_Setting), and differential speed setting (Diff_Speed_Setting).

[0203] As referenced above Figure 5B As explained, a quality value may be provided for each operating parameter included in the training data set to provide a basic score as to whether a particular operating parameter in the operating parameter set can positively affect (i.e., improve) one or more of the plurality of material parameters, or whether degradation of one or more of the plurality of material parameters occurs.

[0204] In the following, a method including an agent (such as the one referenced above) is described. Figure 5B Exemplary training of an enhanced AI engine of an agent 560) described above.

[0205] As an example, if the agent determines a bowl_speed of 37.49 au ( Fig.10An increase in the filter cake thickness (e.g., the first row of the column “Bowl_Speed” of the training dataset) to 65.729 au may positively affect the filter cake thickness, for example, a quality value of 1 may be assigned to the “Bowl_Speed” parameter of 65.729 au. This may indicate to the agent that an increase in “Bowl_Speed” may be advantageous in order to further increase the output value of the agent's winning function. Therefore, the agent may predict an even higher “Bowl_Speed” (e.g., 97.789 au) to potentially increase the value of its winning function even further. In a subsequent iteration step, it may be determined that the filter cake thickness has increased further. Therefore, a quality value may be assigned to the “Bowl_Speed” of 97.789 au. (e.g., a quality value of 1 may be assigned), indicating that a further increase in “Bowl_Speed” is considered advantageous in view of, for example, the optimization of the filter cake thickness. Based on this, the agent may further attempt to increase the value of the winning function by again predicting an even higher “Bowl_Speed” of 124.325 au. In the event that it is determined that further increase in "Bowl_Speed" results in degradation of one or more of the multiple material parameters, "Bowl_Speed" may be initially reduced so that not only individual parameters of the multiple material parameters are optimized, but the multiple material parameters are optimized as a whole. Fig.10 Training is performed similarly for the additional training parameters included in the training data set shown.

[0206] Based on the above exemplary training method, the intelligent agent can obtain knowledge of which parameters of one or more parameters of multiple operating parameters are to be adjusted and how to achieve optimization of multiple material parameters.

[0207] The method according to the present invention can be implemented according to a computer program, which can be run on any suitable data processing device including components configured accordingly (e.g., a memory and one or more processors operably coupled to the memory). The computer program can be stored as a computer executable instruction on a non-transitory computer readable medium. The embodiments of the present disclosure can be implemented in any of various forms. For example, in some embodiments, the present invention can be implemented as a computer-implemented method, a computer-readable storage medium, or a computer system.

[0208] In some embodiments, a non-transitory computer-readable memory medium may be configured such that it stores program instructions and / or data, wherein the program instructions, if executed by a computer system, cause the computer system to perform a method, such as any one of the method embodiments described herein, or any combination of the method embodiments described herein, or any subset of any of the method embodiments described herein, or any combination of such subsets.

[0209] In some embodiments, a computing device may be configured to include a processor (or a group of processors) and a memory medium, wherein the memory medium stores program instructions, wherein the processor is configured to read and execute the program instructions from the memory medium, wherein the program instructions may be executed to implement any of the various method embodiments described herein (or any combination of the method embodiments described herein, or any subset of any method embodiments described herein, or any combination of such subsets). The device may be implemented in any of a variety of forms.

[0210] Although specific embodiments have been described above, these embodiments are not intended to limit the scope of the present disclosure, even in the case where only a single embodiment is described with respect to a particular feature. Unless otherwise stated, the examples of features provided in the present disclosure are intended to be illustrative rather than restrictive. The above description is intended to cover such substitutions, modifications, and equivalents that are obvious to those skilled in the art having the benefit of the present disclosure.

[0211] The scope of the present disclosure includes any feature or combination of features disclosed herein (explicitly or implicitly), or any generalization thereof, whether or not it mitigates any or all of the problems addressed herein. In particular, with reference to the appended claims, features from dependent claims may be combined with features of the independent claims, and features from respective independent claims may be combined in any appropriate manner, and not merely in the specific combinations listed in the appended claims.

Claims

1. A computer-implemented method for optimizing the output of a decanter during operation using an enhanced artificial intelligence (AI) engine, the method comprising: a. operating the decanter according to a plurality of operating parameters; b. processing the physical input comprising sludge and polymer by the decanter and producing a physical output comprising centrate and filter cake; c. determining a plurality of material parameters based on the physical output; d. passing the plurality of material parameters and the plurality of operating parameters to the enhanced AI engine; e. determining, by the enhanced AI engine, a quality value for each of the plurality of material parameters; f. predicting a plurality of adjusted operating parameters by the enhanced AI engine; and g. further operating the decanter based on the plurality of adjusted operating parameters.

2. A computer-implemented method according to the preceding claim, wherein: The plurality of operating parameters and the plurality of adjusted operating parameters include: The differential speed of the scroll of the decanter; decanter bowl speed; a feed stream of said sludge; and / or A feed stream of said polymer.

3. The computer-implemented method of claim 2, wherein: Determining the plurality of material parameters further comprises: Determined by one or more sensors: The dryness of the filter cake, the purity of the centrate, and The dosage of the polymer in the centrifugal filtrate.

4. The computer-implemented method of claim 3, wherein: Determining, by the enhanced AI engine, the quality value of each of the plurality of material parameters further comprises: If the cake density is in the range of 10-19.99% dry matter %DS, a low cake density quality value is set, or if the cake density is in the range of 20-35% DS, a high cake density quality value is set; If the purity of the centrate is in the range of 300-1000 nephelometric turbidity units (NTU), a low centrate purity quality value is set, or if the purity of the centrate is in the range of 50-299.99 NTU, a high centrate purity quality value is set; and If the polymer dosage in the centrifugal filtrate is in the range of 2-9.99 kg / ton dry matter tDS, a high polymer dosage quality value is set, or if the polymer dosage in the centrifugal filtrate is in the range of 10-20 kg / tDS, a low polymer dosage quality value is set.

5. The computer-implemented method according to any one of claims 2 to 4, wherein: Predicting, by the enhanced AI engine, a plurality of adjusted operating parameters further comprises: determining a total energy consumption of the decanter based on the plurality of operating parameters; By considering the balance between the dryness of the filter cake, the purity of the centrate, the total energy consumption of the decanter and the dosage of the polymer in the centrate, the optimal value of each of the differential speed of the vortex of the decanter, the speed of the decanter bowl, the feed flow of the sludge and the feed flow of the polymer is determined.

6. The computer-implemented method of claim 5, wherein: Determining the optimization value further includes: if the filter cake thickness quality value is high, reducing at least one of the optimized value of the feed flow of the polymer, reducing the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or increasing the optimized value of the feed flow of the sludge; If the filter cake thickness quality value is low, increase the optimized value of the feed flow of the polymer, increase at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or reduce the optimized value of the feed flow of the sludge.

7. The computer-implemented method of claims 5-6, wherein: Determining the optimization value further includes: if the centrate purity quality value is high, reducing at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or increasing the optimized value of the feed flow of the sludge; If the centrate purity quality value is low, at least one of the optimized values ​​of the differential speed of the vortex of the decanter and the optimized values ​​of the speed of the decanter bowl is increased and / or the optimized value of the feed flow of the sludge is decreased.

8. The computer-implemented method of claims 5-7, wherein: Determining the optimization value further includes: If the polymer dosage quality value is high, increasing the optimized value of the feed flow of the polymer, and / or increasing at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl, and / or decreasing the optimized value of the feed flow of the sludge; If the polymer dosage quality value is low, the optimized value of the feed flow of the polymer is reduced, and / or at least one of the optimized value of the differential speed of the vortex of the decanter and the optimized value of the speed of the decanter bowl is reduced, and / or the optimized value of the optimized value of the feed flow of the sludge is increased.

9. The computer-implemented method of any one of claims 5 to 8, wherein: Predicting, by the enhanced AI engine, a plurality of adjusted operating parameters further comprises: The plurality of adjusted operating parameters are derived based on the optimized values ​​for each of the differential speed of the scroll of the decanter, the speed of the decanter bowl, the feed flow of the sludge, and the feed flow of the polymer.

10. The computer-implemented method of any preceding claim, further comprising: The enhanced AI engine is adjusted by the intelligent agent based on the plurality of material parameters and the plurality of operating parameters.

11. The computer-implemented method of any preceding claim, further comprising: Perform steps b through g more than once.

12. A computer-implemented method for training an enhanced artificial intelligence (AI) engine that can be used to operate a decanter centrifuge, the method comprising the steps of: a. Simulating the operation of a decanter centrifuge to generate a plurality of operating parameters; b. Determine a plurality of material parameters of the simulated decanter centrifuge; c. passing the plurality of operating parameters and the plurality of material parameters to the enhanced AI engine; d. The intelligent agent adjusts the enhanced AI engine based on the multiple material parameters and the multiple operating parameters.

13. An enhanced artificial intelligence (AI) engine, which is trained according to the method of claim 12 and can be used in the methods of claims 1-11.

14. An apparatus comprising means for performing the method according to any one of claims 1-11 or 12-13.

15. A computer program comprising instructions which, when executed by a processing system, cause the processing system to perform the method according to any one of claims 1-11 or 12-13.

Images