Industrial-scale method and apparatus for producing suspensions for batteries

Abstract

To provide a method and apparatus for industrial scale production of a suspension for a battery, especially a suspension for an electrode of a lithium ion storage battery.SOLUTION: At least one input material EM is processed via ball milling in a rotating chamber of a ball milling device KM, the ball milling is executed as a continuous process with a continuously controlled addition of the input material and with a continuously controlled delivery of the processed output material ZP, the input material and many process parameters are detected as first parameters during production of the suspension SP, results of laboratory analyses about the state or quality of the manufactured suspension are detected as second parameters in a learning phase, the first and the second parameters are used in the learning phase for training a model for predicting the state or quality via machine learning, and the ball milling device is controlled or adjusted outside the learning phase via the first parameters and the trained model.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The invention relates to a method for the industrial-scale production of suspensions for batteries according to the preamble of claim 1 and to an apparatus for the industrial-scale production of suspensions for batteries according to the preamble of claim 13. [Background technology]

[0002] For example, the growing global demand for lithium-ion batteries has led battery cell manufacturers to seek ways to expand their production. In particular, the batteries for electric vehicles require further expansion of production capacity, not only in terms of the number of units produced, but also in terms of the absolute amount of materials. In this case, manufacturers must address two issues: Increasing production capacity can be achieved either by quantitative scaling of current manufacturing machinery or by qualitative changes to processes. The latter is more attractive to manufacturers because a more efficient process increases both production capacity and cost-effectiveness. Such process adaptations to minimize quality variations within production are particularly important due to high reject rates and partially unknown influence relationships within production; are facing.

[0003] Industrially, ball milling and mixing of active substances, binders / additives at the start of electrode production is established as a batch process, i.e. always a defined amount of input material is processed in one step and then output. The throughput is quantitatively scalable mainly by the number of mixing lines. Also so-called batch processes differ in terms of their quality. In contrast, continuous ball milling (where the input material is continuously fed and the processed output material is output) offers a higher throughput and (theoretically) a continuous material flow, resulting in a reduced variability in the quality of the electrode paste produced thereby (also called suspension or "slurry" in the following). The suspension here is the active paste that is mixed and then spread or extruded onto the electrode film. However, continuous ball milling has not been applied in industrial-scale practice, since the control of the process is difficult for long-term quality measurement of the finished product.

[0004] To be able to practically exploit the advantages of continuous ball milling and mixing, i.e. to continuously produce a slurry of constant quality, a precise adjustment of process characteristic values ​​is required. This adjustment is difficult since some of the numerous influencing parameters cannot be measured online (i.e. during a productive industrial-scale production process) and parameters such as the wear progression on the grinding balls, their thermal stability and the residual grinding forces can only be simulated inadequately. A further consideration is that the quality of a slurry is a multidimensional characteristic, some of whose elements, such as its viscosity, its particle size and solid content distribution, cannot be measured online. The "online" measurement of these values ​​fails due to the unavailability of suitable measuring means and methods that can achieve this in a very short time and / or at moderate cost.

[0005] Patent document 1 (Dräger "Method for the production of batteries") discloses a method for the production of suspensions (electrode pastes) for batteries, in particular for batteries of electric vehicles, in which an extruder is controlled with the suspension, linked to the extruder's loading, on the basis of the specific energy acting on the suspension during the extrusion process. However, this method cannot be applied to ball milling, especially for the reasons mentioned above, so that on an industrial scale, ball milling is still carried out in "batch mode", i.e. batchwise.

[0006] Continuous ball mill grinding and mixing has so far mainly been attempted in a laboratory / pilot framework. Industrially, batch-oriented methods are carried out as mentioned above. The grinding of the active materials and the mixing of the raw materials are controlled here by a so-called recipe. This recipe prescribes the ingredients and the method sequence (e.g. duration, temperature, rotation speed of the mixing drum) for a specific cell chemistry. The machine operator can make control interventions in the process based on empirical knowledge (and with the help of laboratory analytical tests of the input or raw materials).

[0007] Industrial mixing lines are equipped with only rudimentary sensor systems and do not allow online measurement of the slurry quality: recipes are typically validated in pilot facilities and monitored by laboratory samples taken randomly during the process.

[0008] Continuous ball milling and mixing is an active area of ​​research in chemical processing. Recent literature in this regard includes laboratory studies testing continuous mixing for specific pharmaceutical applications and exploring suitable sensor systems (e.g., near-infrared spectroscopy (NIRS)).

[0009] Non-Patent Document 1 discloses one such approach.

[0010] These studies herein are directed to uniform mixing of chemical substances and are essentially different from the present invention in the following three respects. a) The concentration of the included component is evaluated during the course of the mixing process. Here, no feedback (in the sense of control) is provided to the characteristic variables of the process itself. b) A single objectively measurable parameter is predicted against slurry quality as a multidimensional and latent criterion (active ingredient concentration). c) The chemical starting materials are already milled, eliminating the complications of gradual wear of the milling balls.

[0011] The mixing process in ball mills is a widely used process in the process industry, e.g. in cement production. However, in that case, online sensor systems with feedback loops are not used, since many production parameters (grinding degree, homogeneity of the ground material, particle size distribution, ...) as well as some process parameters during processing (degree of wear of the balls) cannot be measured. If quality variations are not important in such applications, ball milling / ball mixing is also used in a continuous manner. Due to the quality requirements in battery production, ball milling / ball mixing is used on an industrial scale only in batch mode with constant initial conditions (e.g. each new mixing ball, fixed amounts and conditions of starting materials, constant ambient conditions, etc.). [Prior art documents] [Patent documents]

[0012] [Patent Document 1] International Publication No. 2020 / 216491A1 [Non-patent literature]

[0013] [Non-Patent Document 1] Aditya U. Vanaraze, Manel Alcala, Jackeline I. Jerez Rozo, Fernando J. Muzzio and Rodolfo J. Romanach, "Real-time monitoring of drug concentrations in a continuous powder mixing process using NIR spectroscopy," Chemical Engineering Science, 65(21):5728-5733, 2010 Summary of the Invention [Problem to be solved by the invention]

[0014] It is therefore an object of the present invention to improve existing methods and devices for ball milling in such a way that suspensions for battery electrodes, in particular for BEVs (battery electric vehicles), can be produced with consistently high quality in a continuous production process. [Means for solving the problem]

[0015] These problems are solved by a method according to claim 1 and by a device according to claim 13.

[0016] In this case, a method is proposed for the industrial-scale production of suspensions for batteries, in particular for electrodes of lithium-ion accumulators, in a production facility, in which at least one input material is processed (ground and possibly mixed) by ball milling in at least one rotating chamber with grinding balls and the resulting output material (intermediate product) is mixed with a number of other materials and output by a downstream processing unit, in which the ball milling is carried out as a continuous process with a continuously controlled supply of at least one input material and a continuously controlled output of the processed output material or intermediate product to a downstream processing unit, in which state parameters of the input material and a number of process parameters of the production facility are detected during the production of the suspension, in which at least in a learning phase during the production, results of laboratory analyses on the state or quality of at least the produced suspension or intermediate product are detected as second parameters, in which the first and second parameters are used in the learning phase for training a model for predicting the state or quality by machine learning, and in which at least outside the learning phase the ball milling device is controlled or regulated by the first parameters and the trained model. The use of "soft sensors" thus formed for the achievable quality in the regulating system allows closed-loop control or regulation of the grinding and mixing process in an apparatus for ball milling and / or mixing as a continuous process, with the process characteristic values ​​being adapted such that the quality characteristics of the output slurry remain constant.

[0017] The problem of the production installation furthermore concerns an apparatus for the industrial-scale production of suspensions for electrodes of batteries, in particular lithium-ion accumulators, in which a ball milling device is provided for processing at least an input material in at least one rotating chamber with grinding balls, the ball milling device being configured to carry out a continuous process with a continuously controlled supply of at least the input material and a continuously controlled output of the processed output material to a downstream processing unit, a first sensor is provided for detecting a state parameter of the input material and a number of process parameters of the production installation as a first parameter during the production of the suspension, a second sensor and / or at least one analysis device is provided for detecting a state or quality of at least the produced suspension, a calculation device for training a prediction of the state or quality by machine learning based on the first and second parameters is provided in the learning phase, and a control device is provided that is configured for controlling or regulating the ball milling device by the first parameters and the trained model at least outside the learning phase. The advantages obtained with the above-mentioned method can thereby be achieved.

[0018] Advantageous configurations of the method are disclosed in the dependent claims. These features and advantages can be realized both individually and in any obvious combination. The features and advantages disclosed in connection with the method are equally applicable to the device according to the invention.

[0019] In an alternative embodiment, the computing device and the control device may be the same.

[0020] Advantageously, supervised or reinforcement learning is used for the machine learning. Thus, after the end of the learning phase, a good assessment of the properties (quality) of the final product and, depending on the embodiment, of the intermediate product can be estimated. Furthermore, the models thus generated can also be further improved, for example by retraining with specific process parameters and spot-check analysis. If reinforcement learning is used, a reward function is advantageously used that aims at rewarding at least the smallest possible deviation between the predicted state and the actual state or quality of the produced suspension.

[0021] An important factor affecting quality is considered to be the wear rate of the grinding balls used as one of the process parameters. If the wear rate is calculated from the usage history of the grinding balls, this eliminates the need for repeated measurement analysis of the balls during operation.

[0022] Advantageously, the rotation speed of the ball mill grinding apparatus is used as one of the process parameters, since this influences both the quality or properties of the product and the wear of the grinding balls.

[0023] The properties of the ground material can be determined indirectly by using the energy consumption or the required drive torque of the ball mill grinding equipment as process parameters, whereby these values ​​can also be evaluated with regard to fluctuations.

[0024] The prediction and therefore the accuracy of the control of the production device can be increased by using context parameters from the ball mill grinding device as one of the process parameters. As context parameters, preferably sound emissions from the mixing chamber of the ball mill grinding device and possibly also video images are evaluated and used.

[0025] Advantageously, the suspension is shaped by downstream processing units, in particular by an extruder, to form a blank for an electrode, at the outlet side of which it is possible both to obtain material samples for laboratory analysis checks during the learning phase and to obtain easily accessible parameters in the operating environment, which serve as control "feedback" and thus as input values ​​for the trained model (artificial intelligence).

[0026] An embodiment of the method according to the invention will now be described with reference to the drawing, which also serves to explain the device according to the invention. [Brief description of the drawings]

[0027] [Figure 1] 1 shows a schematic diagram of an apparatus arrangement with one ball milling apparatus and two extruders. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] Figure 1 shows an installation consisting of a ball mill KM and two extruders EX1, EX2. The input material EM is fed to the ball mill KM ("ball milling installation"). Its output stream consists of the intermediate product ZP, which in this example is fed to the extruder EX1 together with the output product of the extruder EX2. Its output stream consists of the finished suspension SP (abbreviated as slurry), which is extruded in the form of electrode blanks (extrusion).

[0029] The manufacturing equipment, in particular the ball mill KM, is equipped with a first sensor device S1 for monitoring and controlling or regulating the working operating environment (normal operation), which is shown in the figure in the form of a microphone in the grinding drum. This first sensor device S1 detects first parameters of the state of the input material EM (temperature, particle size, feed mass flow, etc.). This first parameter includes process parameters of the manufacturing equipment, in particular the ball mill KM (e.g. rotation speed; life, amount and size of the grinding balls MK; temperature; drum filling; required power or rotation torque / torque profile of the drive A) as well as context parameters (e.g. sound recordings from the grinding drum; fluctuations in the drive torque ("ripple"); vibrations; video images). Similarly, non-measurable characteristics of the input material EM, such as origin / supplier, time in storage, etc., are also part of the first parameters. Also, the several sensors of the first sensor arrangement S1 can be installed in various ways, for example to continuously detect the properties of the extrusion material (suspension SP "slurry"), such as parameters of the flow behavior (pressure, dispersion), at the output side of the extruder EX1. It should be noted here that the sensor arrangement S1 operates continuously in real time ("online") during ongoing normal operation.

[0030] Furthermore, the production facility is at least temporarily equipped with a second sensor device S2 or this device is added or made available from the outside, especially in the learning phase. This includes, for example, a laboratory analysis (device for) of the final product, i.e. the suspension SP. In the operating environment, this second sensor device S2 is not available at all or is only available for random samples or at least its values ​​are too slow / slow to be used for closed-loop control. The second parameters thus obtained are primarily related to the quality of the final product SP. By way of example, hyperspectral analysis or near-infrared spectroscopy can be used for this purpose. The second parameters also include other mechanical properties (e.g. toughness, fracture strength) or chemical properties or electrical properties.

[0031] In one embodiment, at least some of the first and second parameters may also be obtained in the intermediate product ZP or exclusively therein, in particular when the second extruder EX2 with further materials is not used and the first extruder EX1 is only used for the extrusion of the suspension SP and does not serve for the incorporation of further materials and therefore has little effect on the essential product properties.

[0032] The second parameters are important adjustments for the control of the production equipment, in particular the ball mill KM, and for the inflow and outflow of the ball mill KM. However, they are often not available at all times and not in real time. It is therefore necessary to provide these second parameters by means of a model, i.e. for the operating environment, the second sensor device S2 is intended to be replaced by a parameterized model in a training phase, which model supplies the required second parameters in real time. This is also called a "soft sensor" or "virtual sensor".

[0033] For this purpose, the production facility is provided with a calculation device, for example an industrial control device (not shown) with a neural processor, which stores or processes, for example, the model in the form of a trained neural network, which is controlled by the sensor values ​​from the first sensor device and other first parameters obtained during ongoing operation, according to whose parameterization a second parameter, i.e. a given quality of the finished suspension SP, is obtained.

[0034] For training (also called parameterization) of the model, "supervised learning" or "reinforcement learning", i.e. well-established methods in artificial intelligence, are used. Training can take place in the computing device mentioned above (mainly industrial control devices). However, the model can also be trained externally. From time to time, the model can be controlled or refined (retrained) by laboratory analysis of random samples, i.e. by the second sensor device S2. For (re)training, the recorded ("logged") data of the first parameter are associated with the second parameter, which is obtained later, in this case on the basis of the produced material, and the model is trained or refined by it. That is to say, for the purpose of training the model, the results do not have to be already available during operation, but rather are associated with the recorded ("logged") production data, so that a long-term analysis of the suspension is possible. In contrast, in the operating environment, the model trained here is "online", here with practically no delay at all, and can provide a prediction of the quality or properties of the currently produced material, which allows for a closed adjustment interval (closed loop) and thus a continuous production process instead of a "batch-based" (charge-based) production process. In this case further parameters can be evaluated and used in the model, in particular the degree of wear of the grinding balls MK can be determined.

[0035] The method described above uses a "supervised learning" approach to derive a predictive model for the online assessment of slurry quality in the grinding-mixing process of electrode manufacturing, influenced among other things by process characteristic values ​​as time series (e.g., rotation speed, number and lifetime of grinding balls), context parameters (e.g., audio recording and vibration of the mixing drum, ambient temperature) as well as characteristic values ​​of the raw material EM (e.g., laboratory analysis, origin, lifetime, temperature at filling).

[0036] The quality of the slurry is determined by laboratory analysis during model training and possibly at later points (quality control, retraining). The analysis can include target values ​​for a "good" slurry with online sensors that can be used to detect the quality during production. In addition, more expensive laboratory analyses that are not performed online in production can be used to determine the quality of the slurry. For example, visual information (e.g., spectrum) or flow properties (e.g., dispersion, viscosity) for a good quality slurry are detected.

[0037] In typical industrial scale production, it is possible to detect exclusively online measurable parameters (e.g. the spectrum of the slurry exiting the mixing process, but often not including e.g. dispersion, viscosity, etc.). These can be directly compared with comparative values ​​of good slurries from a laboratory setting or correlated with measurements of expensive and complex laboratory analyses by the soft sensors mentioned above.

[0038] This online quality model is then used for continuous diagnosis of the slurry quality during ball milling and mixing (e.g. directly in the ball mill or extruder EX1) in the next step, for example by using methods related to "reinforcement learning" to learn the control policy (control specification) of the production parameters (e.g. rotation speed, addition of new grinding balls, amount of solvent, additives) and to adapt the slurry quality predicted by the predictive model to the target values ​​of the quality dimensions.

[0039] In this case, the regulation can either serve as an assistant to the machine operator, for example by means of a screen output, or can intervene autonomously in the process as a closed-loop control. Mixed forms are also possible.

[0040] Ball mill grinding and mixing can be temporarily scaled as a batch process by procurement of new lines / machines. Continuous grinding and mixing offers more efficient scaling of production capacity by adapting production characteristic parameters (e.g. rotation speed). Furthermore, throughput can be flexibly controlled thereby.

[0041] Continuous discharge of material eliminates batch-to-batch quality variations, resulting in a more consistent production process and reduced reject rates.

[0042] The use of soft sensor concepts allows the use of readily available online measurements in the production process for a live quality assessment of the output slurry, which saves the costs of expensive measuring instruments on each line. Such quality assessments, which could previously only be obtained by laboratory analytical measurements and could not be done online (even with expensive sensor systems), can now be done online (in real time) with soft sensors.

[0043] The use of soft sensors within the control system allows for closed loop control of the grinding and mixing process, which adapts process characteristic variables so that the quality characteristics of the output slurry remain constant and at the required levels.

[0044] Soft sensors and regulation systems based on them extend the available automation solutions for electrode manufacturing. Soft sensors can be integrated into the quality monitoring of the manufacturing process to predict the slurry quality. Autonomous control techniques enhance process control and allow data-driven optimization to be integrated into existing automation solutions.

[0045] Information technology methods for training soft sensors and for learning control policies based on sensor data are known methods related to machine learning, i.e. artificial intelligence. However, their application to continuous ball milling and mixing is novel. In particular, the consideration of ball wear as part of the process characteristic variables, the multidimensional description of the slurry quality (viscosity, distribution rate of solids content, moisture content rate, ...) in a single soft sensor and the omission of specialized sensor systems in the operating environment are essential core aspects of the present invention.

Claims

1. 1. A process for the industrial-scale production of a suspension (SP) for batteries in a production facility, in which at least one input material (EM) is processed by ball milling in at least one rotating chamber with grinding balls (MK) of a ball mill grinding device (KM) and an output material (ZP) is discharged mixed with a number of other materials by a downstream processing unit (EX1), said ball milling being carried out as a continuous process with a continuously controlled input of at least an input material (EM) and a continuously controlled output of said processed output material (ZP) to a downstream processing unit (EX1), state parameters of said input material (EM) and a number of process parameters of said production facility are detected as first parameters during the production of said suspension (SP), the results of laboratory analyses relating to the state and quality of at least the suspension (SP) produced are detected as second parameters at least during a learning phase during production, In the learning phase, the first parameters and the second parameters are used to train a model to predict a state or a quality by machine learning; said ball milling device (KM) being controlled or adjusted by said first parameters and said trained model, at least outside said learning phase; The degree of wear of the grinding balls (MK) is used as one of the process parameters; A method comprising:

2. 2. The method of claim 1, wherein supervised learning or reinforcement learning is utilized for the machine learning.

3. 3. The method according to claim 2, characterized in that in the case of the reinforcement learning, a reward function is used which aims at rewarding as small as possible the deviation between the predicted state and the actual state or quality of at least the produced suspension (SP).

4. 2. The method according to claim 1, characterized in that the degree of wear is calculated from the usage history of the grinding balls (MK).

5. 2. The method according to claim 1, characterized in that the rotation speed of the ball mill grinding device (KM) is used as one of the process parameters.

6. 2. The method according to claim 1, characterized in that the energy consumption or the required drive torque of the ball milling device (KM) is used as one of the process parameters.

7. 2. The method according to claim 1, characterized in that a context parameter from the ball mill grinding device (KM) is used as one of the process parameters.

8. 8. The method according to claim 7, characterized in that sound emissions from a mixing chamber of the ball mill grinding device (KM) are used as the context parameter.

9. 2. The method according to claim 1, characterized in that the results of a laboratory analysis of the intermediate product (ZP) at the output of the ball milling plant (KM) are also used as the second parameter.

10. 2. The method according to claim 1, characterized in that the value from a state sensor for the intermediate product (ZP) at the output side of the ball mill grinding device (KM) is also used as the first parameter.

11. 2. The method according to claim 1, characterized in that said suspension (SP) is shaped by said downstream processing unit (EX1) to form a blank for an electrode.

12. A device for the industrial-scale production of suspensions (SP) for batteries in a production facility, comprising a ball mill grinding device (KM) with grinding balls (MK) for processing at least one input material (EM) in at least one rotating chamber, said ball mill grinding apparatus (KM) being configured to provide a continuous process of continuously controlled input of said at least one input material and continuously controlled delivery of treated output material (ZP) to a downstream processing unit (EX1); a first sensor (S1) is provided for detecting as first parameters during the production of said suspension (SP), state parameters of said input material (EM) and a number of process parameters of said production installation, a second sensor and / or at least one analytical device is provided for detecting, at least during the learning phase, the state or quality of the suspension (SP) produced, In the learning phase, a computing device is provided for training a model to predict the state or the quality by machine learning based on the first parameter; a control device is provided that is configured for controlling or regulating the ball mill grinding device (KM) by means of the first parameters and the trained model, at least outside the learning phase; The degree of wear of the grinding balls (MK) is used as one of the process parameters; An apparatus comprising:

Images