Method and apparatus for manufacturing a suspension for batteries on an industrial scale

Abstract

The invention relates to a method and a device for manufacturing a suspension for batteries on an industrial scale, at least one input material being processed by means of ball milling in at least one rotating chamber provided with milling balls of a device for ball milling. The ball milling is implemented as a continuous process with a continuous controlled addition of at least one input material and with a continuous controlled delivery of the processed output material, state parameters of the input material and a plurality of process parameters of the production facility are acquired as first parameters during the manufacturing of the suspension, results of a laboratory analysis regarding at least the quality or state of the manufactured suspension are acquired as second parameters during the manufacturing at least in a learning phase, the first parameters and the second parameters are used for training a model for predicting the state or the quality by means of machine learning in the learning phase, the device for ball milling is open-loop controlled or closed-loop controlled by means of the first parameters and the trained model at least outside the learning phase.

Description

Technical Field

[0001] The present invention relates to a method for mass production of suspensions for batteries, and an apparatus for industrial-scale production of suspensions for batteries. Background Technology

[0002] Due to the ever-increasing global demand for batteries such as lithium-ion batteries, battery manufacturers are seeking ways to expand their production capacity. This is particularly true for batteries for electric vehicles, which are placing increasing demands on production capacity not only in terms of the number of units but also in the absolute quantity of materials. Here, they face two challenges:

[0003] • Increased capacity can be achieved by expanding the number of existing production machines or by changing the quality of the process. The latter variant is more attractive to manufacturers because the more efficient methodological steps improve both capacity and cost efficiency.

[0004] Due to high scrap rates and unknown internal factors in production, process adaptation is particularly important for minimizing quality fluctuations within production.

[0005] In industry, at the start of electrode production, the ball milling and ball mixing of active materials and binders / additives are defined as batch processes. This means that a limited amount of input material is always processed and then output in a single step. Production volume can be quantitatively scaled by the number of mixing lines. Additionally, so-called batches differ in their quality. Continuous ball milling, on the other hand (where input material is continuously fed and processed output material is output), provides higher production volumes and (theoretically) reduced quality fluctuations in the electrode paste (hereinafter also referred to as suspension or "slurry") produced by a continuous material flow. Here, the suspension is an active paste that is coated or extruded onto the electrode film after mixing. However, continuous ball milling has not been used in industrial-scale practice due to the difficulty in process regulation caused by the long duration of finished product quality measurement.

[0006] To leverage the advantages of continuous ball milling and mixing in practice, precise tuning of process characteristic variables is necessary to consistently output slurry of consistent quality. This tuning is challenging due to a large number of influencing parameters, some of which cannot be measured online (i.e., during industrial-scale manufacturing processes) and can only be adequately simulated: for example, the wear process of the grinding balls, their temperature resistance, and residual grinding force. Furthermore, slurry quality is multidimensional, and some factors cannot be measured online: for example, its viscosity, particle size, and solids content distribution. “Online” measurement of these variables is unsustainable due to the lack of availability of suitable measuring instruments and methods that can be performed quickly and / or at a reasonable cost.

[0007] The document WO 2020 / 216491 A1 Dreger's "Method for Manufacturing Batteries" proposes a method for manufacturing suspensions (electrode pastes) for batteries, particularly those for electric vehicles, in which an extruder with a suspension is controlled based on the relationship between a specific energy acting on the suspension during the extrusion process and the filler load of the extruder. However, for the reasons mentioned above, this method cannot be used for ball milling, so in industrial-scale applications, ball milling is typically performed in a "batch run" manner, i.e., batch by batch, as before.

[0008] To date, continuous ball milling and ball mixing have been primarily attempted at the laboratory / laboratory scale. As previously mentioned, batch-oriented process steps dominate in industry. Here, the milling of active materials and the mixing of raw materials are controlled via so-called formulations. For a specific single-pool chemistry, the formulation stores the contents and process flow (e.g., duration, temperature, rotational speed of the mixing drum). The machine operator is able to intervene in the process in a guided manner based on their experience (and with the aid of research on laboratory analysis of the input materials or raw materials).

[0009] Industrial mixing lines are equipped with only rudimentary sensors that do not allow for online measurement of slurry quality. Formulations are typically validated in pilot facilities and monitored continuously using randomized laboratory samples.

[0010] Continuous ball milling and ball mixing are active areas of research in chemical process procedures. To this end, experimental laboratory studies have been found in recent literature that test continuous mixing for specific pharmaceutical applications and explore suitable sensing devices (e.g., near-infrared spectroscopy NIRS).

[0011] Publications Aditya U. Vanarase, Manel Alcalà, Jackeline I. Jerez Rozo, Fernando J. Muzzio and Rodolfo J. The paper “Real-time monitoring of drug concentration in a continuous powder mixing process using NIR spectroscopy” in Chemical Engineering Science, 65(21):5728-5733, 2010 demonstrates this approach.

[0012] Here, this study considers the uniform mixing of chemically active materials and differs from the present invention in three main points:

[0013] a) Estimate the concentration of inclusions during the progress of the mixing process. No feedback (in a control sense) is provided on characteristic variables of the process itself.

[0014] b) In contrast to slurry quality, which is a multidimensional and potential standard, predict a single, objectively measurable parameter (active material concentration).

[0015] c) The initial chemical materials have been ground; eliminating the complexity caused by the gradual wear of the grinding balls.

[0016] The mixing process in ball milling is common in processing industries, such as cement manufacturing. However, online sensing devices with feedback loops are not used here because many product parameters (grind, uniformity of the ground material, particle size distribution, etc.) cannot be measured, but some process parameters (ball wear) cannot be measured. If quality fluctuations are not significant in such applications, ball milling / ball mixing is also used in continuous methods there. Due to the quality requirements in battery production, ball milling / ball mixing is used there on an industrial scale only in batch operations with defined initial conditions (e.g., new mixed balls per batch, specified amounts and conditions of initial materials, constant environmental conditions, etc.). Summary of the Invention

[0017] Therefore, the object of the present invention is to adapt existing methods and equipment for ball milling so that suspensions, particularly for electrodes of batteries for BEVs (electric vehicles), can be manufactured with consistently high quality in a continuous production process.

[0018] This objective is achieved by the method and apparatus according to the invention.

[0019] Here, a method is proposed for industrial-scale manufacturing of suspensions for electrodes used in batteries, particularly lithium-ion batteries, in a production facility. At least one input material is processed (grinding and, if necessary, mixing) by ball milling in at least one rotating chamber equipped with grinding balls, and the resulting output material (intermediate product) is mixed with various other materials and discharged by means of a subsequent processing unit. Ball milling is implemented as a continuous process, utilizing the continuous controlled addition of at least one input material and the continuous controlled delivery of the processed output material or intermediate product to the subsequent processing unit. During suspension manufacturing, state parameters of the input material and multiple process parameters of the production facility are acquired as first parameters. At least during a learning phase, the results of laboratory analysis of the quality or state of at least the manufactured suspension and possible intermediate products are acquired as second parameters. During the learning phase, the first and second parameters are used to train a model that uses machine learning to predict state or quality. At least outside the learning phase, the ball milling apparatus is subjected to open-loop or closed-loop control using the first parameters and the trained model. The use of “soft sensors” formed in this way to achieve achievable quality within a closed-loop system allows for closed-loop control of the grinding and mixing processes as continuous processes in a device for ball grinding or ball mixing, wherein process characteristic variables are adapted so that the quality characteristics of the leaving slurry remain constant.

[0020] This objective is also achieved by an apparatus for industrially manufacturing suspensions of electrodes for batteries, particularly lithium-ion batteries, in a production facility, wherein an apparatus for ball milling is provided to process at least one input material in at least one rotating chamber equipped with grinding balls. Here, the apparatus for ball milling is configured to perform ball milling as a continuous process by continuously and controlledly adding at least one input material and by continuously and controlledly conveying the processed output material to subsequent processing units. A first sensor is provided to acquire state parameters of the input material and multiple process parameters of the production facility as first parameters during suspension manufacturing. A second sensor and / or at least one analytical device are provided at least during a learning phase to acquire the quality or state of at least the produced suspension. A computational device is provided for training a model for predicting the state or quality during the learning phase using the first and second parameters via machine learning. A control device is provided, configured to perform open-loop or closed-loop control of the apparatus for ball milling using the first parameters and the trained model at least outside the learning phase. This enables the realization of the advantages discussed according to the method.

[0021] The features and advantages of this method can be achieved individually or through meaningful combinations. The features and advantages disclosed in this method are also applicable to the device according to the invention.

[0022] In one implementation variant, the computing device and the control device can also be the same.

[0023] Supervised learning or reinforcement learning is advantageously used in machine learning. Here, after the learning process ends, a good estimate of the characteristics (quality) of the final product can be made, and depending on the implementation, the characteristics (quality) of intermediate products can also be estimated. Furthermore, the resulting model can be further improved during operation, for example, by retraining using recorded process parameters and random analysis of the products manufactured therein. If reinforcement learning is used, a reward function is advantageously employed, which aims to reward the smallest possible deviation between the predicted state or quality of at least the produced suspension and the actual state or quality.

[0024] By using the wear degree of the grinding balls as one of the process parameters, a fundamental factor affecting quality is considered. If this wear degree is calculated from the usage history of the grinding balls, there is no need for repeated measurement and technical analysis of the balls during operation.

[0025] Advantageously, the rotational speed of the ball grinding device is used as one of the process parameters, because it affects the quality or characteristics of the product, as well as the wear of the grinding balls.

[0026] By using the energy consumption or required drive torque of the apparatus used for ball milling as one of the process parameters, the material properties of the material being milled can be indirectly detected. Furthermore, the fluctuations in these values ​​can also be evaluated.

[0027] By using contextual parameters from the apparatus used for ball milling as one of the process parameters, the predictability and accuracy of the control of the production apparatus can be improved. Preferably, acoustic emissions and possible video images from the mixing chamber of the apparatus used for ball milling are evaluated and used as contextual parameters.

[0028] Advantageously, the suspension is then processed in subsequent units, particularly by an extruder, into a preform for electrodes. At the extruder outlet, during the learning phase, material samples for laboratory analysis and parameters easily detectable during operation are obtained, serving as "feedback" for control and further as input values ​​for the learned model (artificial intelligence). Attached Figure Description

[0029] An embodiment of the method according to the present invention will now be explained with reference to the accompanying drawings. These drawings also serve to explain the apparatus according to the present invention.

[0030] Here, Figure 1 The diagram schematically shows an apparatus with a ball milling device and two extruders. Detailed Implementation

[0031] Figure 1 The diagram illustrates an apparatus consisting of a ball milling device KM and two extruders EX1 and EX2. Input material EM is fed to the ball mill KM (“the ball milling device”). Its output stream consists of an intermediate product ZP, which, in this example, is fed to the extruder EX1 along with the output product of extruder EX2. The output stream also consists of a prepared suspension SP (abbreviated as: slurry), which is extruded (extruded) in the form of electrode preforms.

[0032] The production facility, particularly the ball mill KM, is equipped with a first sensing device S1 for monitoring and open-loop or closed-loop control operation (normally); in the accompanying drawings, a microphone in the grinding cylinder is shown as an example for the sensing device S1. The first sensing device S1 detects first parameters (temperature, particle size, inflow mass flow, etc.) of the state of the input material EM. However, process parameters of the production facility, particularly those of the ball mill KM (e.g., rotational speed; lifespan, quantity, and size of the grinding balls MK; temperature; cylinder fill level; required power or torque / torque curve of the driver A), and contextual parameters (e.g., recordings from the grinding cylinder; fluctuations in drive torque (“ripples”); vibration; video images) are also part of the first parameters. Similarly, non-measurable characteristics of the input material EM, such as source / supplier, shelf life, etc., are also part of the first parameters. Some sensors of the first sensing device S1 can also be placed differently; for example, the sensor can continuously detect characteristics of the extruded material (suspension SP – “slurry”) at the output of the extruder EX1, such as parameters of flow behavior (pressure, dispersion). It should be noted that the sensing device S1 operates continuously and rapidly (“online”) during normal operation.

[0033] The production facility is also at least temporarily equipped with a second sensing device S2, or, especially during the learning phase, a second sensing device is provided externally or added. Here, for example, it is a device for laboratory analysis, particularly for the manufactured product, i.e., the suspension SP. The second sensing device S2 is not available at all during operation, or is only available randomly, or is at least slow / sluggish, making its values ​​unusable for closed-loop control. Therefore, the second parameter detected in this way primarily relates to the quality of the manufactured product SP in the first line. For this purpose, for example, hyperspectral analysis or near-infrared spectroscopy can be used. Other mechanical properties (e.g., toughness, fracture strength) or chemical or electrical properties also fall under the category of second parameters.

[0034] In one implementation, at least some of the first and second parameters can also be detected, or only at the intermediate product ZP. This is particularly applicable when a second extruder EX2 with other materials is not used and the first extruder EX1 is used only to extrude the suspension SP without mixing other materials, thus having little impact on the main product characteristics.

[0035] The second parameter is an important control variable used to control the inflow and outflow of the production facility, particularly the ball mill KM and ball mill KM. However, they are not always available and are usually not immediately usable. Therefore, it is necessary to provide the second parameter by means of a model, that is, the second sensing device S2 should be replaced for operation by means of a parameterized model during the training phase, which quickly provides the required second parameter. Here, this is also referred to as a "soft sensor" or "virtual sensor".

[0036] To this end, the production facility is equipped with computing devices, such as industrial control units with neural processors (not shown in the figure), in which models are stored or processed, for example, in the form of trained neural networks. This model is driven by sensor values ​​available during continuous operation of the first sensing device and other first parameters, and returns a second parameter based on its parameterization, namely the predicted quality of the produced suspension SP.

[0037] To train (also known as parametric) the model, "supervised learning" or "reinforcement learning," established methods of artificial intelligence, are used. Training can be performed on the described computing device (primarily an industrial control unit). However, the model can also be trained externally. Sometimes, the model can be examined or simplified (retrained) by means of laboratory analysis of random samples, i.e., by means of a second sensing device S2. For (re)training, the possible recorded ("registered") data of the first parameter are correlated with the second parameter to train or simplify the model, which is later obtained based on the material produced therein. That is to say, long-term analysis of the suspension is also possible, since the results used for model training do not need to be available during operation, but are correlated with the recorded ("registered") production data. The model trained therein can provide predictions of the quality or properties of the material being produced "online" and with almost no delay during operation, which realizes a closed-loop control route, enabling continuous production instead of a "batch-based" production process. Here, other parameters can also be estimated and incorporated into the model, particularly the wear of the grinding balls MK.

[0038] The described method employs a "supervised learning" approach to derive a predictive model for estimating slurry quality during the grinding and mixing processes in online electrode production. Here, process characteristic variables are used as time series (e.g., rotational speed, number of grinding balls, and lifespan), contextual parameters (e.g., recordings and vibrations of the mixing drum, ambient temperature), and characteristics of the raw material EM (e.g., laboratory analysis, origin, storage period, and temperature at filling).

[0039] The quality of the slurry is determined through laboratory analysis during model training and, where possible, at subsequent time points (quality control, retraining). This analysis can include target values ​​for "good" slurry on online sensors available for quality inspection during manufacturing. Furthermore, more costly laboratory analyses not performed online within manufacturing can be used to determine slurry quality. Examples include visual information (e.g., spectra) or flow properties (e.g., dispersion, viscosity) used to detect good-quality slurry.

[0040] During routine, industrial-scale manufacturing, only parameters that can be measured online (such as the spectrum of the slurry flowing out from the mixing process, but typically without dispersion or viscosity, etc.) can be detected. These parameters are either compared directly to comparative values ​​from a good slurry in a laboratory setting, or correlated with more costly laboratory analytical measurements using the aid of the described soft sensors.

[0041] In the next step, an online quality model is used to continuously predict slurry quality during ball milling and mixing (e.g., directly in ball milling or in extruder EX1). Using methods such as reinforcement learning, a regulation strategy (regulation rule) is learned for production parameters (e.g., rotational speed, addition of new grinding balls, solvent quantity, additives). This regulation strategy teaches target values ​​for the quality dimensions based on the slurry quality predicted by the prediction model.

[0042] Here, adjustment can serve as an assistant to one or more machine operators via screen output or similar means, or as an autonomous intervention process by a closed-loop control device. A hybrid approach is feasible.

[0043] As a batch process, ball milling and mixing can be scaled up primarily by acquiring new lines / machines. Continuous milling and mixing offers a more efficient scaling up of capacity by matching manufacturing parameters (such as rotational speed). This also allows for more flexible control of production volume.

[0044] Continuous material outflow eliminates quality fluctuations between batches, which stabilizes the manufacturing process and reduces scrap rates.

[0045] By using the soft sensor concept, on-site quality assessment of the leaving slurry can be performed during production using simple and readily available online measurements. This saves on the cost of expensive measuring instruments on each line. Quality assessments that have previously been obtained only through laboratory analysis and measurements and could not be performed online (even with expensive sensing devices) are now provided online (rapidly) via soft sensors.

[0046] The use of soft sensors within the control system allows for closed-loop control of the grinding and mixing processes, which are adapted to maintain the quality characteristics of the exiting slurry at a constant and desired level.

[0047] Soft sensors and conditioning systems expand the automation solutions available for electrode production. They enable the integration of soft sensors into quality monitoring devices throughout the manufacturing process to predict slurry quality. Autonomous conditioning technology enhances process control and allows for the incorporation of data-driven optimization into existing automation solutions.

[0048] Information technology methods used to learn adjustment strategies based on soft sensors or sensor data are known machine learning methods, i.e., artificial intelligence. However, their application to continuous ball milling and mixing is novel. In particular, the fundamental core aspects of this solution are to consider ball wear as part of the process characteristic variables, to provide a multidimensional description of slurry quality (viscosity, solids distribution, moisture content, etc.) in a single soft sensor, and to omit dedicated sensors during operation.

Claims

1. A method for industrial-scale production of a suspension (SP) in a production facility, said suspension being used in batteries. in, At least one input material (EM) is processed by ball milling in at least one rotating chamber equipped with grinding balls (MK) in an apparatus (KM), and the output material (ZP) is mixed with a variety of other materials and discharged by means of a subsequent processing unit (EX1). Its features are, The ball milling is performed as a continuous process by continuously and controlledly adding at least one of the input materials (EM) and by continuously and controlledly conveying the processed output material (ZP) to a subsequent processing unit (EX1). During the manufacture of the suspension (SP), the state parameters of the input material (EM) and multiple process parameters of the production facility are obtained as first parameters, wherein the wear degree of the grinding balls (MK) is used as one of the process parameters. During the manufacturing process, at least during the learning phase, the results of laboratory analyses regarding the state or quality of the at least produced suspension (SP) are acquired as a second parameter, which cannot be used for closed-loop control. During the learning phase, the first parameter and the second parameter are used to train a model, which is then used to predict the state or quality using machine learning. At least outside the learning phase, the device (KM) for ball grinding is controlled in an open-loop or closed-loop manner by means of the first parameter and the trained model.

2. The method according to claim 1, Its features are, Supervised learning or reinforcement learning is used in the machine learning process.

3. The method according to claim 2, Its features are, In the case of the reinforcement learning, a reward function is used, which is designed to reward the smallest possible deviation between the predicted state or quality and the actual state or quality of at least the prepared suspension (SP).

4. The method according to claim 1, Its features are, The degree of wear is calculated from the usage history of the abrasive ball (MK).

5. The method according to any one of claims 1 to 3, Its features are, The rotational speed of the apparatus (KM) used for ball milling is used as one of the process parameters.

6. The method according to any one of claims 1 to 3, Its features are, The energy consumption or required drive torque of the device (KM) used for ball grinding is used as one of the process parameters.

7. The method according to any one of claims 1 to 3, Its features are, The context parameters from the apparatus (KM) used for ball milling are used as one of the process parameters.

8. The method according to claim 7, Its features are, The acoustic emission from the mixing chamber of the device (KM) used for ball milling is used as the context parameter.

9. The method according to any one of claims 1 to 3, Its features are, The results of laboratory analysis of the intermediate product (ZP) at the outlet of the apparatus (EM) used for ball milling are also used as the second parameter.

10. The method according to any one of claims 1 to 3, Its features are, The value of the state sensor of the intermediate product (ZP) at the outlet of the apparatus (EM) used for ball milling is also used as the first parameter.

11. The method according to any one of claims 1 to 3, Its features are, The suspension (SP) is then shaped into a blank for use as an electrode by the subsequent processing unit (EX1).

12. The method according to claim 1, Its features are, The suspension is used as an electrode in lithium-ion batteries.

13. The method according to claim 11, Its features are, The processing unit is an extruder.

14. An apparatus for industrial-scale production of a suspension (SP) in a production facility, said suspension being used in batteries. in, An apparatus (KM) for ball grinding is provided, the apparatus being used to process at least one input material (EM) in at least one rotating chamber equipped with grinding balls (MK). Its features are, The apparatus (KM) for ball milling is configured to perform the ball milling as a continuous process by continuously and controlledly adding at least one input material (EM) and continuously and controlledly conveying the processed output material (ZP) to a subsequent processing unit (EX1). A first sensor (S1) is provided to acquire, during the manufacturing of the suspension (SP), the state parameters of the input material (EM) and multiple process parameters of the production facility as first parameters, wherein the wear degree of the grinding balls (MK) is used as one of the process parameters. At least during the learning phase, a second sensor (S2) and / or at least one analytical device are provided to obtain the state or mass of at least the prepared suspension (SP). The values ​​of the second sensor (S2) and / or at least one analytical device cannot be used for closed-loop control. A computing device is provided for training a model, which is used in the learning phase to predict the state or quality using machine learning based on the first and second parameters. A control device is provided, which is configured to, at least outside the learning phase, perform open-loop or closed-loop control of the device (KM) for ball grinding using the first parameter and the trained model.

15. The device according to claim 14, Its features are, The suspension is used as an electrode in lithium-ion batteries.

Images