METHOD FOR OPTIMIZED OPERATION OF A FAN OR FAN ARRANGEMENT

Description

1. Digital twin and twin algorithm

[0001] The underlying principle here is to ensure the best possible efficiency and running performance at every operating point of the fan. This is difficult due to conflicting operating parameters.

[0002] Experience with fans has shown that the ball bearing and bearing grease are critical parameters for the fan's service life. The service life of the ball bearing and bearing grease depends significantly on the operating temperature in or around the motor and the mechanical forces acting on the ball bearing. Since neither temperature nor force sensors can be positioned in the immediate vicinity of the bearing, neither the bearing temperature nor the bearing forces acting on it can be measured. Consequently, it is necessary to either measure these parameters indirectly or determine them computationally.

[0003] From DE 10 2010 002 294 A1, a system or method for determining the condition of the bearing of an electric machine is known. Real sensor units determine a measured value which is transmitted to a simulation unit. Using the simulation unit, a result value is determined, which is either a bearing current value or a value dependent on the bearing current. The result value is transmitted to another unit for further calculation. Due to the required sensors, the known system / method is complex and difficult to apply to fans due to the lack of sufficient installation space.

[0004] Document EP 1 669 226 A1 discloses a model-based method for controlling the air conditioning system of a motor vehicle. Furthermore, document US 2017 / 323274 A1 discloses a method for controlling an aircraft and its propulsion components. Document US 2016 / 333854 A1 discloses a user interface for operating a wind farm. A method for operating a gas turbine is previously known from US 2012 / 060505 A1.

[0005] The digital twin and twin algorithm are based on the creation of a digital representation of a real fan, namely by mapping its properties using mathematical calculation models and, if necessary, incorporating known data, possibly including real measurement data. The real measurement data can be current measurement data from the ongoing operation of each individual motor (and possibly its history). Furthermore, at least one operating parameter-specific algorithm is created, taking into account known relationships, characteristic curves, etc., and used for further calculations.

[0006] Component states of the fan are determined or calculated via a digital model using virtual sensors. These component states are fed into an operating- or operating-parameter-specific or product-specific algorithm, which determines or calculates specific operating parameters of the fan from the component states and, if necessary, derives predictions regarding the fan's operation, such as predictions about its service life. Crucially, the combined use of determined component states and real-world measurement data is possible.

[0007] This involves the use of two different software components: a first software component relating to the digital twin and a second software component relating to the operating parameter-specific algorithm, which can be described as an "intelligent" algorithm.

[0008] The digital twin is a digital representation of a real, individual object; in the case of the invention, a fan or fan system. The digital twin models the fan's properties using a computational model and, optionally, by incorporating known fan data. The task of the digital twin is to calculate the component states of the fan's components as a function of the respective operating state using virtual sensors. The component states determined based on such a calculation are transmitted to the operating parameter-specific algorithm, which uses the digital twin's operating data to determine / calculate operating parameters or operating states of the fan, for example, the bearing life and / or the bearing grease life. Based on the result, situation-appropriate control adjustment is possible.Operating parameters and operating states are equally relevant insofar as they are calculable quantities.

[0009] The previously discussed combination of digital twin and operating parameter-specific algorithm can be implemented as a digital twin algorithm on a microprocessor assigned to the motor of the fan and thus assigned to the fan as a fixed component.

[0010] The digital twin algorithm is the combination of a digital twin describing the fan with a type of intelligent algorithm that is designed to be specific to operating parameters.

[0011] With a suitably designed fan, predictive maintenance can be implemented, aiming to prevent fan failure due to, for example, a defective bearing or bearing grease. The goal is to adjust the system parameters to the specific situation in order to achieve virtually the maximum possible service life of the fan.

[0012] Using a digital model of the fan and operating parameter-specific algorithms, predictive maintenance aims to maximize the service life of the fan components while simultaneously preventing any fan failure. The fan's service life is calculated based on the calculated component conditions and the resulting operating parameters.

[0013] The digital twin uses physical, mathematical, statistical, empirical, and / or combined models to calculate thermal and mechanical component states. This includes mathematical, physical, and non-physical models. The operating parameter-specific algorithm (intelligent algorithm) requires the component states determined by the digital twin to calculate arbitrary operating parameters, such as predicting fan failure. Since a fan's lifespan depends primarily on the ball bearings and the ball bearing grease, the operating parameter calculation focused on the ball bearing grease and the ball bearings plays a particularly important role.

[0014] It is well known from experience that the service life of bearing grease depends significantly on the operating temperature. The higher the operating temperature over the entire service life, the faster the bearing grease is consumed. Consequently, it is necessary to determine the bearing temperature in order to ascertain the service life of the bearing grease.

[0015] To determine the bearing temperature, a temperature sensor would have to be positioned in the immediate vicinity of the bearing. This is not possible due to the geometric and functional characteristics of the fan / motor. Therefore, according to the invention, component states such as the bearing temperature are calculated using the digital twin and an operating parameter-specific algorithm.

[0016] The calculation is based on a mathematical model, which in turn is based on a reduced coupled thermomagnetic calculation model. The combination of a digital twin and an operating parameter-specific algorithm calculates heat sources, heat sinks, and the thermal state of the entire system pertaining to the fan motor. This allows the bearing grease temperature to be determined as a function of the fan / motor's operating state via the virtual sensors of the digital twin and fed into the operating parameter-specific algorithm as the operating state.

[0017] Both the digital twin, including its virtual sensors, and the operating parameter-specific algorithm can be implemented in machine code (C code) on the existing microprocessor, thereby giving the fan a certain degree of machine intelligence.

[0018] The preceding explanations describe a method for determining the operating states of a fan using a digital representation (digital twin) of the fan, incorporating at least one operating parameter-specific algorithm. This forms the basis for the subsequent process steps, which calculate operating states determined by virtual sensors using a digital twin algorithm. A workflow is defined for implementing the digital twin algorithm with respect to the fan. In particular, the aim is to eliminate the need for physical sensors to determine operating states. 2. intelligent fan

[0019] A quasi-"intelligent" fan is created through a process for optimizing the efficiency and / or running performance of a fan. This involves starting with component- or function-specific numerical detail models and applying at least one algorithm to reduce the model and thus the data (data refinement) to component- or function-specific behavioral models. The reduced data of the behavioral models are then coupled or combined in a system simulation to form a system behavioral model with input and output variables. The input variables and associated output variables of the fan from the system behavioral model are then made available to an optimizer for selection in order to achieve optimized control of the system depending on the framework conditions.

[0020] This concerns the digital twin algorithm in the context of an "intelligent" fan in the sense of a further development of the twin algorithm according to point 1.

[0021] The further development of the digital twin algorithm can be understood as an independent, situation-appropriate adjustment of the system parameters of the fan or fan system in order to guarantee the best possible efficiency and the best possible running performance at every operating point.

[0022] First, detailed numerical models are created, for example, a thermal model, a magnetic circuit model, or a model relating to blade position and flow or flow conditions. The detailed model can also be a digital twin, as discussed in the introduction, of a fan environment, such as a data center as a complete system. The detailed model can also be the digital twin of a fan assembly. Other detailed models are conceivable.

[0023] In the next step, the detailed models are reduced to so-called behavioral models. This is accompanied by a significant reduction in the amount of data generated.

[0024] Subsequently, in the system simulation, the behavioral models with reduced data are coupled, resulting in a behavioral study with a combined behavioral model.

[0025] The simulation of the entire system is performed in system space with a homogeneously distributed combination of input variables. The result is a table containing the input combinations and their corresponding system output variables. This table reflects a system behavior model, specifically the input variables and their corresponding output variables of the fan. Optimization can then be performed based on these variables.

[0026] During operation, an optimizer, depending on environmental conditions, selects the best possible system output variables, such as system efficiency, from the behavioral model's table, preferably in real time. Once the best possible system output variable has been found, the corresponding input variables can be read from the table. The system is then controlled using these input variables, preferably in real time.

[0027] In light of the foregoing, it is essential that the optimizer selects the optimal system efficiency from the system behavior table and provides the control system with the necessary input variables. This enables continuous optimization. 3. Optimization according to the invention

[0028] The present invention, based on the digital twin including twin algorithm and optimization of the efficiency and / or running performance of a fan, using an "intelligent" fan, aims to generate, on the one hand, a continuous improvement of the digital twin algorithm and, on the other hand, improved products (fans) through product innovation, namely by using calculated operating states.

[0029] The above problem is solved by the features of claim 1.

[0030] The basis for the method according to the invention is the use of a digital twin and the twin algorithm used therein, in accordance with the introductory discussion to paragraphs 1 and 2.

[0031] Through innovation analysis on the one hand and algorithm analysis on the other, the digital twin algorithm is continuously improved and new product innovations are generated.

[0032] The starting point for the inventive method is the digital twin algorithm, which calculates the operating states of the fan during operation, i.e., on-site at the customer's premises. These operating states are sent to the cloud for analysis and subsequent processing. The analysis advantageously utilizes a special tool, namely a machine learning program unit in the cloud. "Machine learning" refers to the artificial generation of knowledge from experience.

[0033] Such an artificial system learns from examples of calculated operating states according to a digital twin algorithm and can generalize these after the actual learning phase is complete. This type of system recognizes patterns and regularities in the training data based on the calculated operating states.

[0034] In light of the foregoing, product innovations and improvements to algorithms can be made.

[0035] Furthermore, it is essential that the calculation and recording of component states serves to gain new insights, particularly in the design of new products. There is a continuous increase in digital know-how.

[0036] Innovation analysis also makes it possible to determine what the customer actually needs in terms of the fan, based on a specific requirements profile. This leads to new innovations through fan products individually tailored to the customer.

[0037] The innovation analysis provides "feedback to design" through the analysis of smart data, as generated by model reduction according to point 2 of the introductory discussion.

[0038] Existing or improved algorithms can be used to calculate the lifespan or optimize the performance of the fan and can be further refined during the analysis. These improved algorithms make the fan "intelligent" and enable the best possible maintenance prediction. This is of particular importance.

[0039] Ultimately, the improved algorithms and new product innovations are incorporated into the development of new fan products and also lead to a continuous improvement of the digital twin algorithms.

[0040] There are various ways to advantageously develop and refine the teaching of the present invention. Reference is made, on the one hand, to the claims subordinate to claim 1 and, on the other hand, to the following explanation of a preferred embodiment of the invention with reference to the drawing. In conjunction with the explanation of the preferred embodiment of the invention with reference to the drawing, generally preferred embodiments and further developments of the teaching are also explained. The drawing shows Figs. 1 to 16 show process steps for realizing the teaching according to the invention with special characteristics, wherein the teaching according to the invention is exemplified by means of Figure 6will be explained.

[0041] The Figures 1 to 5 They serve to explain the teaching according to the invention and refer to the digital twin and the digital twin algorithm as the basis for the intelligent fan.

[0042] Specifically, it shows Figure 1 The combination of the digital twin with at least one operating parameter-specific algorithm, which is hereinafter referred to as the digital twin algorithm. This can be discussed using the example of the service life of the bearing grease and / or the bearing itself.

[0043] As previously explained, the service life of bearing grease and bearings depends on the operating temperature and the speed of the motor. Since no temperature sensor can be positioned in the immediate vicinity of the bearing, the bearing temperature must be calculated using a model, according to the invention with the digital twin algorithm, which results from a combination of a digital twin and an operating parameter-specific algorithm (intelligent algorithm).

[0044] The digital twin is simply a mathematical model based on a simplified coupled thermomagnetic and mechanical computational model. The digital twin calculates the thermal and mechanical state of the entire system pertaining to the engine. Through the virtual sensors associated with the digital twin, it can determine the bearing grease temperature as a function of the engine's operating state.

[0045] The intelligent algorithm requires component states for further data processing, for example, to predict fan failure. Failure characteristic curves allow the motor's failure to be calculated, or at least estimated. All software related to the digital twin algorithm is implemented in machine code (C code) on the motor microprocessor, eliminating the need for additional electronics.

[0046] Figure 2 This illustrates the process of calculating the bearing lifespan of the grease in a fan motor bearing. Creating a digital representation of the real fan requires detailed numerical models, specifically thermal models, magnetic circuit models, etc. Furthermore, algorithms for calculating the grease lifespan are developed.

[0047] The detailed models are then reduced to behavioral models to make the data volume manageable.

[0048] Subsequently, the behavioral models and the algorithm calculating bearing grease life are coupled in a system simulation, specifically through a combination of the digital twin with the operating parameter-specific algorithm, which in this case calculates the bearing grease life. The C code is then generated from the system simulation and directly implemented on the engine microprocessor.

[0049] As previously explained, reducing the detailed model to a behavioral model is necessary to decrease computation time. This allows the digital twin algorithm to be implemented on the motor's microprocessor. Various methods can be used for thermal model reduction, such as Krylov's method. This method reduces the data of the detailed model by decreasing the model order.

[0050] The detailed magnetic model can be reduced using an algorithm or a table. The table defines pre-calculated results for specific configurations, allowing complex calculations to be replaced by a quick value search. Using these reduced models, the bearing grease temperature and the bearing temperature can be calculated. The calculated values ​​are then used with the operating parameter-specific algorithm—in this case, the algorithm for calculating bearing grease life—to determine the service life of both the bearing grease and the bearing itself.

[0051] Furthermore, it is possible to weight the consumed service life of the bearing / bearing grease preferably exponentially depending on the operating temperature.

[0052] Figure 3This shows the progression of such a weighting factor over the temperature range, using exemplary parameters such as continuous operation, bearing type, viscosity, rotational speed, grease temperature, and operating time / lifespan for calculating bearing grease lifespan. The calculation example shows a consumed lifespan of 15 minutes for an operating time of four minutes.

[0053] The reduced models based on the digital twin and the operating parameter-specific algorithm concerning bearing grease life are integrated into a system simulation and linked together. The system simulation can be created, for example, in the MATLAB program. Using MATLAB's code generator, it is possible to translate the system simulation into C code and implement it on the engine microprocessor.

[0054] The Figures 4 and 5The individual steps of the process for creating the "intelligent" fan are shown, whereby Figure 4 on the position of the wing angle and Figure 5 This relates to the load distribution of fans in a data center. The respective process steps, from creating or providing a detailed model to a reduced model, a system simulation, and finally a behavioral model, are identical in both cases. Based on the behavioral model, the optimizer selects the optimal system efficiency from the system behavior table and passes the corresponding input variables to the controller, with control occurring in real time. The procedure and data are generated in C code, allowing the optimization to run on a standard processor.

[0055] According to the presentation in Figure 4The angle of the fan blades must be controlled to achieve optimal system efficiency at each operating point. A reduced model is derived from the detailed model using a suitable algorithm. From this reduced model, a behavioral study and a resulting behavioral model are generated through system simulation using multiple detailed models. The optimizer selects the optimal system efficiency from the system behavior table and passes the corresponding input variables, which enable optimization, to the control mechanism. The entire system is controlled in real time on a microprocessor, based on the behavioral model and the optimization algorithm. The data and the programming algorithm are written in C code.

[0056] According to the presentation in Figure 5This concerns the load distribution of an arrangement of multiple fans; in the chosen embodiment, this involves the load distribution of fans in a data center. The flow velocity and the resulting necessary load distribution of individual fans must be controlled to achieve optimal system efficiency, depending on the prevailing temperature in the data center. Here, too, the optimizer selects the optimal system efficiency from the system behavior table and provides the corresponding input variables to the control mechanism, enabling real-time control of the entire system on a microprocessor. Again, the data from the behavioral model are fed into the optimization algorithm, with the program running in C code on standard processors.

[0057] Based on the previously discussed data reduction, this method enables the creation of a compact C code that can run on standard microprocessors. A type of data refinement (Big Data → Smart Data) takes place on the microprocessor, and this is the result of the calculation. Only the compressed, refined data is further processed or, for example, sent to a cloud. It goes without saying that this significantly reduces the streaming volume of the connection to the cloud.

[0058] Furthermore, the operating parameters determined using a digital twin and an operating parameter-specific algorithm can be used for predictive maintenance and the maintenance of a fan on the one hand, and for optimizing the design and operation of a fan on the other, whereby the digital twin algorithm is further developed to independently adapt the system parameters to the situation in order to guarantee the best possible efficiency and the best possible running performance at every operating point.

[0059] Figure 6 Figure 1 schematically illustrates the process of the inventive method, wherein the digital twin algorithm runs in the motor microprocessor during fan operation. Operating states are calculated using virtual sensors. Based on these operating states, control adjustment and the calculation of the fan's service life are possible.

[0060] The data is transferred from the motor microprocessor to the cloud, on the one hand for innovation analysis and on the other hand for algorithm analysis.

[0061] In the field of innovation analysis, machine learning is used, resulting in continuous product innovations.

[0062] In the field of algorithm analysis, machine learning also takes place, resulting in increasingly intelligent algorithms.

[0063] The results of both analyses – innovation analysis and algorithm analysis – are fed back to improve the digital twin and the digital twin algorithm, with the aim of creating "intelligent" algorithms and "intelligent" fans.

[0064] For the invention, in a particular embodiment according to the features of claim 5, it is of particular importance that patterns and regularities are recognized. In addition to the foregoing, the following can be stated: "Artificial Intelligence" is the umbrella term for describing all research areas concerned with the performance of human intelligence by machines. A subfield of artificial intelligence is "machine learning," according to which machines are given the ability to generate "knowledge" from experience, in other words, to learn. "Deep learning" with artificial neural networks is a particularly efficient method of continuous machine learning based on statistical analysis of large datasets and is therefore the most important future technology within artificial intelligence.

[0065] Deep learning is a learning method within the field of machine learning. Using neural networks, the machine is able to recognize structures, evaluate this recognition, and independently improve or optimize itself through multiple forward and backward iterations.

[0066] For this purpose, artificial neural networks are divided into multiple layers. These act like a weighted filter, working from coarse to fine, thus increasing the probability of pattern recognition and outputting a correct result. This is modeled on the human brain, which functions similarly.

[0067] Artificial neural networks can be represented as matrices. This has the advantage that the necessary calculations can be performed quite easily. Computation on a conventional microprocessor, such as that of a fan, is possible. This enables the intelligent fan without the need for an internet connection.

[0068] Artificial neural networks can be used to adapt the digital twin algorithm to a manufacturer's diverse product range and to different customer applications during operation. Furthermore, artificial neural networks can be used to predict fan failure based on fault patterns. This involves considering several interrelated factors, such as increased current and elevated electronics temperature.

[0069] It is important that the development of modern fans is based on a digital twin of the fan and the application of a digital twin algorithm. Operating states are determined from the fan's operation, and control adjustments and influences on its lifespan are made. The innovation analysis and algorithm analysis are fed back into the digital twin algorithm, leading to improvements in the algorithm's execution, with the ultimate goal of creating an "intelligent" fan.

[0070] The presentation in the Figures 7 and 8 It serves to further explain the claimed doctrine.

[0071] The use of a digital representation, namely a digital twin of the fan, is essential. The digital twin is created through data processing. Specifically, it is generated from a combination of known input variables or sensor readings with calculated values ​​and models. Using the digital twin, component temperatures, currents, losses, etc., are determined at specific, predetermined points on the fan. Real values, such as specific component temperatures, are virtually determined using the digital twin, particularly when there is no economically or structurally feasible way to measure them using sensors at a given point on the fan.

[0072] Of further importance for the claimed teaching is the operating parameter-specific algorithm. Based on the results or data provided by the digital twin, such as the bearing temperature, key performance indicators (KPIs) such as the probability of failure or the remaining service life of the fan or the fan bearing are determined. These KPIs depend on the current operating parameters of the fan and their history, i.e., the operating points and environments in which the fan is / was operated.

[0073] The Figures 7 and 8 The inventive method for determining the operating states of a fan using a digital image of the fan, taking into account the preceding explanations, is illustrated by means of a concrete example.

[0074] In the left column of Figure 7The measured or calculated input variables, including the units assigned to the arrows, are shown. These input variables are measured using existing standard sensors or are known.

[0075] Heat sources and heat sinks are calculated from these input variables. This calculation is based on simulation-based models that consider heat sources such as copper, iron, and electronic losses, as well as heat sinks such as engine cooling (cooling wheel, airflow, and ambient temperature). This results in input variables for a reduced thermal model with virtual sensors. All of this corresponds to the digital twin in the sense of a thermal model.

[0076] Component temperatures are calculated from the reduced thermal model with virtual sensors. This thermal model simulates the fan physics and, based on virtual sensors, calculates the temperature in the bearing, winding, magnet, and various electronic components as needed.

[0077] Figure 8 shows as a continuation of Figure 7 It is clear that output variables from the reduced thermal model, possibly with additional parameters, are used as input variables for calculating the aging process. Underlying aging models are based on historical data and can be stored as characteristic curves. This allows the remaining service life, limited by aging, to be individually calculated or corrected on-site based on the actual fan history and current operating condition.

[0078] The respective aging models result in a calculated service life in days or hours, which, in itself, can serve as mere information. This information can then be used for further forecasting, namely predicting the remaining service life of individual components or the entire fan. This forecast can then be used for intelligent remaining service life optimization. To extend the remaining service life, measures can be implemented, such as reducing the rotational speed or intelligently distributing the load across multiple fans. These measures can be communicated via a control variable.

[0079] Figure 9This section again shows the digital twin, down to the reduced thermal model with virtual sensors, which represents the fan and motor. As previously explained, the thermal model simulates the fan physics and calculates various temperatures based on virtual sensors. These temperatures are used for different purposes, such as: for monitoring: Determining operating parameters using virtual sensors and using them for monitoring. This can include: warning messages, status LEDs, comments in a readable error code, images in the cloud or app application, and display in user interfaces. Regarding Predictive Maintenance:A method for determining the aging of a fan, consisting of numerous subsystems such as ball bearings, windings, electronic components, magnets, and predicting the remaining service life. Applications include, for example, planning maintenance intervals, achieving the longest possible service life before the next maintenance interval (i.e., avoiding premature maintenance), automatic scheduling of maintenance appointments, notification of maintenance needs, and automatic ordering of spare parts. for optimization: Methods for determining operating conditions relating to product performance, i.e. efficiency, component temperatures, rotational speed, output power, volume flow, noise level, vibrations, etc. to create an intelligent fan:Responding to specific operating conditions to improve performance or achieve specific goals. Changing the operating point / control parameters for optimal efficiency. Changing the operating point to achieve the longest possible service life. Reducing the rotational speed when the probability of failure is very high: Changing the operating point during a day-night cycle for the quietest possible night operation. Outputting a control variable for auxiliary or customer devices, e.g., temperature output for use in controlling a heat pump or for additional cooling. Targeted avoidance of critical system states (e.g., resonances, overheating, etc.).

[0080] For a better understanding of the teaching according to the invention, both the sequence of the process steps and their content are important. The sequence of the respective process steps can be derived from the development workflow of the underlying algorithm. This is described in Figure 10 shown, whereby in a final step the procedure can be further developed.

[0081] The following principles are fundamental to creating a detailed model: A model is a representation or approximation of reality, meaning, by definition, an approximation. A model is always limited to a section that is relevant for the intended representation. Furthermore, a model is inherently incomplete, as it is either reduced for ease of use with regard to its necessary input parameters, or certain physical behavioral elements are unknown during model creation. Depending on the intended use and objective, a different type of model building is necessary, i.e., for example, a different area under consideration, a different required accuracy in the results, or a different speed of computation.There are many types of models, although in the technical field a model is usually combined with a mathematical representation, for example with algebraic equations or inequalities, systems of ordinary or partial differential equations, state space representations, tables, graphs.

[0082] Virtual product development using finite element simulation (FE simulation) is an integral part of modern product development. Typically, a physical domain (e.g., strength, thermal properties, or magnetic circuitry) is represented in a very large (on the order of 100 gigabytes) and computationally intensive model, and the results are determined at millions of locations (nodes) within the model. This is one type of detailed model. The general process for creating these detailed models can be outlined as follows: 1. Import of a 3D geometry, for example from CAD software, 2. Assignment of boundary conditions, i.e., fixed constraints, material definitions, contact conditions (glued joints, sliding connections, thermal insulation), 3. Meshing (splitting the geometry into millions of small, linked elements), 4. Application of loads, i.e., forces, heat sources / sinks, magnetic fields, 5. Automatic solution of the resulting differential equations for each individual element and combination into a single result for the overall model, 6. Evaluation of the results.

[0083] The creation of detailed models with virtual sensors for fans / complete systems with fans proceeds as follows: Detailed models are created to represent the physics of the fan and / or the overall system. So-called virtual sensors are defined computation points within these detailed models. These virtual sensors calculate component states, such as the winding temperature in the thermal detailed model of the fan. Detailed models are simulation models that are computationally intensive in terms of computation time, processing power, and memory requirements. Such detailed models, for example, thermal models, magnetic circuit models, electronic models, control models, force models, or vibration models, are used to calculate non-linear operating conditions.The physical effects of the system involve interactions between the domains, which is why the individual models must be considered coupled within the overall system. Computations with detailed models within the overall system are not practical in terms of computation time, as interactions cannot be evaluated in real time. Model reduction is therefore necessary.

[0084] The creation of reduced models can proceed as follows: Model reduction generally describes taking an existing model and reducing it further to optimize it, for example, in terms of memory requirements or processing speed. Depending on the specific use case, there are many variations, such as: ▪ Approximation of simple mathematical functions, such as polynomial functions, by simply storing the coefficients. ▪ Creating tables for various input variables and then either using these discrete values ​​or interpolating between them. ▪ Approximation of statistical models that provide predictions based on past values. ▪ Graphs / logic gates, for example: If T>200°C, then the fan is defective. Generation of the reduced models - Example a)

[0085] The starting point for the reduced model is a finite element (FE) model of the thermals, which maps the temperatures at each point of the model depending on the heat input and heat output. In the following example, the model reduction is simplified to only one heat input and one heat output, only one temperature to be determined at point A, and only the values ​​"high" and "low." A parameter study is performed for this purpose, resulting in the following so-called "look-up table": Temperature at point A Heat input = Low = 1W Heat input = High = 11W Heat output = Low = 1W 40°C 80°C Heat output = High = 5W 20°C 60°C

[0086] There are then several ways to use the results. ▪ Using the table directly and discretely. Example: If a temperature at point A is to be predicted for a heat input of 4W and a heat output of 1W, the value of 40°C is used directly. ▪ Using the table and linear interpolation between the values. Example: If a temperature at point A is to be predicted for a heat input of 5W and a heat output of 1W, the value of 60°C is determined via linear interpolation. ▪ Using the table to determine a temperature prediction function using regression. Examples of objective functions are polynomial functions, linear functions, exponential functions, statistical functions, differential equations, etc. The temperature is then determined using this function. Generation of the reduced models - Example b)

[0087] The starting point for the reduced model is a finite element (FE) model of the thermal system, which maps the temperatures at each point in the model based on heat input and output. Subsequently, compact state-space models can be approximated using mathematical assumptions, calculations, and transformations (for example, the LTI system or Krylov subspace method). These consist of two essential differential or integral equations and four matrices describing the entire system (for example, 200x200 matrices filled with scalar values). However, these matrices no longer represent the temperature at millions of nodes, but only at a few selected locations. Furthermore, the approximation leads to a deviation in the results depending on the size of the state-space model. Generally, the larger the model and its matrices, the smaller the deviation.

[0088] State-space models are available as procedures, modules, or objects in many computer algebra programs, such as Matlab, and also in programming languages, allowing such models to be calculated simply by importing the matrices. Input variables include, for example, heat input into the system and heat sinks due to convection; output variables include, for example, specific component temperatures (e.g., three different component temperatures). Generation of the reduced models - Example c)

[0089] In this example, the starting point for model reduction is experimental results. As in example a), a table of measurement results would be created, and then the process would proceed accordingly (discrete use, linear interpolation, or regression using mathematical functions).

[0090] Of further importance may be the coupling of physical domains or different models.

[0091] Traditionally, virtual product development considers domains individually, as a combined approach is computationally and memory-intensive and hardly practical. Model reduction makes it possible to couple models from different domains. For example, coupling a detailed magnetic circuit model, which takes several days or even weeks to compute on a high-performance computing cluster, with a thermal model is not advantageous. Such coupling is often necessary to achieve the most accurate possible representation of real-world behavior.

[0092] Coupling of physical domains or different models - examples ▪ The winding resistance is approximately linearly dependent on the copper temperature. The power loss in the winding changes approximately linearly depending on the winding resistance. Depending on the power loss, the thermal behavior changes, for example, winding and bearing temperatures, in a strongly non-linear fashion, which in turn affects the winding resistance. Therefore, a coupling is necessary here, depending on the requirements of the model results. ▪ The required torque and speed of a fan are strongly dependent on the system resistance and, for example, the pressure differential and temperature of the conveyed medium. Depending on the load torque, the behavior of the magnetic circuit changes, i.e., currents through the winding, magnetic field, speed, etc. Depending on this, power consumption, losses, and achievable speed also change.Here too, in the case of a customer application, it is conceivable to link the fan behavior to the installation situation, depending on the application.

[0093] Specification of a technical implementation - example a) ▪ Creation of a thermal FE model of a fan → Computationally intensive and memory-intensive FE model with 1,000,000 elements in addition to the polynomial function. Heat sources and sinks are represented as polynomial functions depending on input current and rotational speed. ▪ Creation of a reduced thermal model using statistical methods, which represents the electronic component temperature as a function of input current and rotational speed. → Polynomial function describing the temperature as a function of input current and rotational speed = virtual temperature sensor. ▪ Characteristic curve from the datasheet for the lifetime of the electronic component as a function of its temperature → Operating parameter-specific algorithm that calculates the probability of failure from the virtual temperature sensor. ▪ Use for predictive maintenance, monitoring, or optimization of the operating point → Intelligent algorithm.

[0094] Specification of a technical implementation - example b) • Current pointer and motor speed are measured by integrated electronics / control. The electromagnetic operating point is derived from this data. • Based on this operating point, the losses of the motor and power electronics are calculated using lookup tables or polynomial functions. • A thermal model processes the loss values ​​and determines the temperatures of critical system components such as ball bearings or semiconductor components. • Simultaneously, component vibrations are recorded via a physical sensor. The local vibrations are virtually projected onto the overall system using behavioral models, allowing, for example, the estimation of bearing loads due to vibrations. • Using operating parameter-specific algorithms, the determined temperatures and vibration values ​​are converted into an estimate of component and fan lifetime. • This enables more comprehensive measures such as predictive maintenance.▪ At the same time, knowledge of the losses allows the operating point and system efficiency to be optimized through control engineering adjustments, such as varying the feedforward angle.

[0095] In the context of applications involving fans and / or fan systems, the preceding explanations regarding model reduction of the detailed models apply, whereby order reduction can be achieved using the Krylov subspace method. The goal is to minimize computation time, required computing power, and memory requirements, thus enabling real-time calculations. The virtual sensors are retained and continue to provide output values.

[0096] According to Figure 10In the next step, the reduced models are linked to form the fan system model. Specifically, the reduced models, such as the thermal model, the magnetic circuit model, the software model, the electronics model, etc., are combined into a fan system model. This fan system model represents the physics of an individual fan, a group of fans, or a fan system and calculates the efficiency, operating performance, and any interactions between the individual models, depending on the environmental conditions and operating states.

[0097] In the next step, the fan system models are linked to the plant model, creating a comprehensive system model. This comprehensive system model consists of several fans and a plant unit, for example, including a compressor and / or condenser. The plant model can be implemented using the same workflow as the fan system model. Subsequently, the fan system models and the plant model can be linked to form the comprehensive system model.

[0098] The next step involves a behavioral study, namely the calculation of response sizes with input parameter combinations.

[0099] The aim of the study is to determine the behavior of the overall system model and to use this knowledge to control the system in real time.

[0100] According to the overall system behavior, effects and influences of model input variables on the model response variables are transferred or mapped in the design space.

[0101] The design space is a multidimensional space defined by the possible input variables. The number of input variables corresponds to the dimension of the design space. This means that with ten input variables, there are ten dimensions.

[0102] The model input variables are varied within defined limits. This results in parameter combinations that evenly cover and thus describe the multidimensional space. The model response variables, such as efficiency and operating performance, are calculated based on these parameter combinations. The behavioral study provides a design space filled with response variables that depend on the input variables. This space represents the overall system behavior.

[0103] Figure 11This shows how various input variables are incorporated into fan system models and cooling circuit models, resulting in the overall cooling system model. A corresponding output variable is understood as the result of the overall cooling system model. The resulting knowledge can be used according to... Figure 12 The data can be transferred into a behavior table. Once the overall system behavior is known, the input variables can be adjusted to achieve the best possible response.

[0104] The adjustment or selection of the response variable and the corresponding input variable combination from the behavior table is implemented by an optimizer, namely according to the one described in Figure 10 further process step shown, according to which the optimizer selects the best response size and thus chooses the best possible input parameter combination for the current operating state.

[0105] According to the presentation in Figure 13The optimizer selects the best response parameter and determines the optimal combination of input parameters for the current operating state. In other words, the optimizer selects the best possible model response based on the environmental conditions / operating state. The corresponding parameter combinations of the input variables are then set. This allows for optimal system control. The overall cooling system behavior table can run on any processor, preferably on the fan's microprocessor, which is already present. This enables precise control.

[0106] Figure 14 refers to a possible further development, according to which the overall cooling system behavior table would be as shown in Figure 13 The system is enhanced by a cloud-based system simulation of the cooling circuit. This system includes digital twins of the fan, digital twins of the cooling circuit, a validation unit, and a virtual controller / optimizer.

[0107] The digital twins of the fan and the cooling circuit physically represent the system. The virtual controller has access to the knowledge of the overall cooling system behavior table as shown in [reference]. Figure 13 Furthermore, the virtual controller can learn through machine learning, for example, in relation to specific customer applications. The validation unit improves the digital twins by comparing target and actual values. This gives the system the ability to simulate specific customer patterns and improve itself based on the resulting insights.

[0108] According to the combination of features in claim 1, innovation analysis is of particular importance. The goal of innovation analysis is to improve efficiency, operational performance, and reduce material costs by gaining insights from customer applications. Digital representations of the customer's actual operations are collected. Through pattern recognition, optionally using AI (artificial intelligence, machine learning), the characteristic properties of the customer's operations are identified and recognized. A customer model is created based on these characteristic properties. Artificial intelligence is then used to analyze the model for weaknesses.

[0109] An example of such a weakness is the operation of the fan in a range where heat dissipation is suboptimal. Compared to catalog measurements, the efficiency has decreased. Once such weaknesses are identified, the optimization goal is to increase efficiency at the customer's operating point. The optimizer uses the detailed model of the fan and the customer's sample model for optimization. The result of the optimization could be, for example, a winding adjustment and / or a geometry modification to improve the fan's heat dissipation.

[0110] Figure 15 shows a workflow that takes place autonomously in / in a cloud.

[0111] Figure 16 refers to algorithm analysis in the cloud.

[0112] The goal of these intelligent algorithms is to improve and develop new algorithms for classifying and predicting fan failures. Digital twins are collected during real-world customer operations. Artificial intelligence creates a pattern from the collected data and recognizes specific changes in this pattern over time, up to the point of a failure. Based on these pattern changes, a failure can be classified and a prediction of its occurrence can be generated. This prediction is then implemented in an algorithm, typically based on neural networks or self-learning systems. After improvement or development, the algorithm is deployed to the customer's operations, with improvements being implemented cyclically or periodically. The workflow is automated in the cloud, as described in the diagram. Figure 16 instead of.

[0113] Finally, it should be expressly noted that the exemplary embodiments described above serve only to illustrate the claimed teaching, but do not limit it to these exemplary embodiments.

Claims

1. Method for optimised operation of a fan or a fan arrangement, wherein using a digital image of the fan or the fan arrangement and at least one operating-parameter-specific algorithm, whose combination forms a digital twin algorithm, the data and knowledge obtained during operation are supplied to an innovation and algorithm analysis so that, on the one hand, production innovations, that is to say, an improved fan or an improved fan arrangement, and, on the other hand, improved intelligent algorithms are generated, wherein with the innovation analysis digital images are collected during real client operation and by means of pattern recognition the characteristic properties of the client operation are filtered out and identified, wherein with reference to the characteristic properties from the client operation a client model is produced and wherein by means of artificial intelligence an analysis of the model with respect to weaknesses is carried out, wherein the algorithm analysis uses a program for machine learning for artificial generation of knowledge from experience during operation of the fan or the fan arrangement, wherein thermal and mechanical component states of the fan or the fan arrangement are calculated via the digital twin algorithm during operation of the fan by means of virtual sensors, wherein the operating-parameter-specific algorithm establishes from the component states specific operating parameters of the fan or the fan arrangement, wherein characteristic variables of the fan are calculated taking as a basis calculated component states and resulting operating parameters, and wherein the improvement of the algorithms is used for optimum calculation of the servicelife and / or in order to optimise the operation and / or best possible prognosis for the maintenance of the fans or the fan arrangements.

2. Method according to claim 1, characterised in that component or function-specific digital detailed models, which by way of a model and data reduction, where applicable a data enhancement, are converted into component or function-specific behaviour models and in a system simulation are coupled or combined with input and output variables to form a system behaviour model, are used to make available the input variables and associated output variables of the fan or the fan arrangement from the system behaviour model to an optimiser for selection.

3. Method according to claim 1 or 2, characterised in that the calculated component states of the respective algorithm analysis are preferably supplied via the / in the cloud.

4. Method according to any one of claims 1 to 3, characterised in that, from learning data which are obtained by means of examples of the states of the digital twin algorithm, patterns and laws are derived.

5. Method according to any one of claims 1 to 4, characterised in that the innovation analysis provides information relating to the requirement of the operator of the fan or the fan arrangement.

6. Method according to any one of claims 1 to 5, characterised in that the innovation analysis provides innovations relating to fans or fan arrangements which are individually adapted to the client.

7. Method according to any one of claims 1 to 6, characterised in that the innovation and algorithm analysis is carried out continuously.