Prediction system and prediction method for enlarging production scale

Abstract

The invention provides a prediction system for enlarging a production scale, which comprises the following parts: (a) a production scale enlargement theoretical part for predicting process parameters of factory enlargement production according to data obtained from a laboratory scale test; (b) a "data driving section" that adjusts an amplification parameter obtained from the "production scale amplification theoretical section"; and (c) an auxiliary decision-making part for outputting the corrected factory trial production parameters according to the data obtained from the data driving part. The invention also provides a prediction method for enlarging the production scale. The prediction system and the prediction method are used for predicting the enlargement of the production scale of a cathode active material (CAM) in a lithium ion battery.

Description

Technical Field

[0001] The present invention relates to the technical field of process amplification, and in particular to a prediction system for amplifying production scale and a corresponding prediction method. Background Art

[0002] The development of new cathode active materials (CAM) usually includes steps such as laboratory development and medium-scale trials (or factory trial production). From laboratory-scale trials to medium-scale trials or trial production, the expansion of production scale leads to differences in production conditions, such as differences in equipment process parameters, expert knowledge and many other aspects. In addition, the demand for expanding production scale usually comes from customers, and there is time pressure. There is often not enough time to repeatedly design and compare experiments to obtain the best technical route. Even if there is enough time, many rounds of testing are required to obtain satisfactory results. On the one hand, this increases the consumption of raw materials and the cost of product development. On the other hand, since the equipment on the production line needs to be used for testing, it will interfere with normal production activities. In addition, in actual work, the required expert knowledge is highly dependent on technicians, and expert knowledge is likely to be lost due to personnel turnover.

[0003] Therefore, there remains a need to provide a method and system for scaling up production from laboratory development to pilot-scale trials that can be performed quickly, cost-effectively, and fully utilize expert knowledge. Summary of the invention

[0004] In this document, any specific values ​​mentioned for a feature (including a series of specific values ​​as endpoints) can be re-combined into new ranges.

[0005] In this document, each defined aspect may be combined with any other aspect unless explicitly indicated to the contrary. In particular, any feature indicated as being preferred or advantageous may be combined with any other feature indicated as being preferred or advantageous.

[0006] As used herein, "process parameters" means any and all parameters of a production process that can be displayed by any device or instrument, as well as the weight percentages of all raw material components that are mixed in a chemical reaction to obtain the final product.

[0007] In this article, "equipment parameters" means all equipment parameters used to produce the final product and add raw materials.

[0008] In this article, “product quality parameters” refer to all performance test results about the CAM final product.

[0009] In one aspect, the present invention provides a prediction system for scaling up production, comprising the following parts: (a) "Scale-up theory", which predicts the process parameters after scale-up based on the data obtained from laboratory-scale experiments; (b) a “data-driven part”, adjusting the parameters obtained from the “theoretical scale-up part”; and (c) The “decision-making support part” outputs parameters suitable for the next round of predictions for the factory trial production based on the data obtained from the “data-driven part”.

[0010] In another aspect, the present invention provides a method for predicting production scale expansion, comprising the following steps: (i) Input the initial values ​​of the laboratory scale test data into the "scale-up theory section" to predict the parameters after scale-up; (ii) using a “data-driven portion” to adjust predicted production scale-up parameters based on one or more machine learning algorithms; (iii) obtaining the correction direction and magnitude of the data obtained from step (ii) through the “decision-making assistance part”; (iv) applying the revised parameters obtained from step (iii) to the factory trial production and conducting the next round of predictions through steps (i) to (iv); and (v) Repeat the iterative process from steps (i) to (iv) until the final predicted result is consistent with the actual result.

[0011] By using the prediction system and prediction method of the present invention, the scale-up of CAM production can be achieved quickly, at low cost and with full use of professional knowledge.

[0012] The “production scale-up theory part” is based on various scale-up models, such as constant Reynolds number (Re), constant energy input per unit volume (P / V), constant linear velocity and constant Froude number.

[0013] In fluid dynamics, the Reynolds number (Re) is a dimensionless quantity that helps predict fluid flow patterns in different situations by measuring the ratio between inertial and viscous forces. The Reynolds number is defined as Re = uL / ν = ρuL / μ, where ρ is the fluid density, u is the flow velocity, L is the characteristic length, μ is the dynamic viscosity of the fluid, and ν is the kinematic viscosity of the fluid. In order to maintain similarity between two flow systems under the action of viscous forces, the ratio of the inertial force to the viscous force of the water flow must be equal, that is, the Reynolds number must be equal. At a constant tip speed, the impeller Reynolds number increases as the device size becomes larger. Scaling up the Reynolds number to keep the Reynolds number constant for pipe flow or impeller mixing will result in a decrease in velocity in larger diameter pipes or vessels. This reduces the flow rate and reduces the efficiency of the pipe or agitator. Other parameters need to be adjusted to offset the disadvantages.

[0014] The Froude number is a dimensionless number defined in continuum mechanics that represents the ratio of the flow inertia to the external field. The Froude number is defined based on the velocity-length ratio as Fr = u / (gL) 1 / 2 , where u is the local flow velocity, g is the local external field, and L is a characteristic length.

[0015] For different devices, any of the above-mentioned amplification models or a combination of a plurality of amplification models may be used. In some preferred embodiments, constant Reynolds number, constant power / volume ratio, constant blade tip speed and constant Froude number theoretical models are used simultaneously.

[0016] The data input for scale-up comes from laboratory scale experiments, including process data, equipment parameters and product quality parameters. In some embodiments, the equipment parameters include equipment type, equipment size (such as diameter, height and volume, etc.), and internal equipment components such as agitators and baffles. Process data include raw materials, temperature, pressure, stirring speed and stirring time. Product quality parameters include particle size distribution (PSD), residual lithium content (%), pH value, capacity, C value and cycle performance.

[0017] The predicted scale-up process parameters obtained from the "production scale-up theory part" can be input into the "data-driven part" to further correct the coefficients. The "data-driven part" is based on machine learning and uses one or more of the Boost algorithm, SVM (support vector machine) algorithm, neural network algorithm, Bayesian algorithm, decision tree algorithm, LSH (local sensitive hashing) algorithm and KNN (K nearest neighbor) algorithm. Through the output data from the "production scale-up theory part", combined with the information of raw materials and laboratory and production line historical data, the closest predicted value can be found through similarity and distance measurement. The data obtained by the "data-driven part" is ready to be input into the "assisted decision-making part".

[0018] The "decision-making support part" is based on a knowledge tree constructed through interviews with experts in the field of technology, and its structure follows the flow chart for preparing cathode active materials for use in batteries. After the data obtained in the "data-driven part" is input into the "decision-making support part", combined with the historical data of the production line, the direction and magnitude of the parameters can be adjusted through calculations through the branches of the knowledge tree to obtain the best solution.

[0019] The output data of the "decision-making support part" can be used for the next round of prediction. After multiple rounds of iterative calculations, the data obtained are getting closer and closer to the actual situation. The iterative calculation continues until the parameter error falls within the acceptable range. Whether the final prediction result is consistent with the actual parameter will be judged based on the determination coefficient R2, and the calculation formula is as follows: R2 = 1 - Σ i (yi -y f ) 2 / Σ i (y i -y a ) 2 Formula 1 Among them, y i is the actual data point, y f is with y i The corresponding predicted data point, y a Yes i The prediction results are acceptable when R2 is in the range of 0.8 to 1, preferably between 0.9 and 1.

[0020] In some preferred embodiments, the method for predicting production scale-up comprises the following steps: (i) Input the initial data values ​​from the bench-scale experiments into the “Theoretical Scale-up Section” to obtain the predicted scale-up parameters; (ii) adjusting predicted amplification parameters based on the “data-driven portion” of one or more machine learning algorithms; (iii) obtaining the correction direction and magnitude of the data obtained from step (ii) through the “decision-making assistance part”; (iv) using the revised parameters for plant trial production obtained in step (iii) for the next round of predictions in steps (i) to (iv); and (v) Repeat the iterative process from step (i) to (iv) until the final prediction result is consistent with the result obtained from the actual situation. Through the production expansion prediction method based on the prediction system, the production scale of CAM can be quickly expanded at a low cost and expert knowledge can be fully utilized. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The diagram is a schematic diagram of production expansion from laboratory test scale to factory trial production scale based on the prediction system and prediction method of the present invention.

[0022] Figure 2 Schematic diagram of how the "Theoretical part of production scale-up" works.

[0023] Figure 3 A diagram to illustrate how the “data-driven part” works.

[0024] Figure 4 A schematic diagram to illustrate how the "decision-making support part" works. DETAILED DESCRIPTION

[0025] The invention can be implemented in many different ways, and several specific embodiments are shown below. Such disclosure is an example of the principles of the present invention and does not limit the present invention to only the specific embodiments described below.

[0026] Embodiment 1: A prediction system for scaling up production, comprising the following parts: (a) “Scale-up theory”, which predicts the process parameters for scale-up to industrial plants based on data obtained from bench-scale experiments; (b) a “data-driven part”, adjusting the scale-up parameters obtained from the “production scale-up theory part”; and (c) The “decision-making support part” outputs the revised parameters of the factory trial production based on the data obtained from the “data-driven part”.

[0027] Embodiment 2: According to the prediction system of embodiment 1, the "production scale expansion theoretical part" is predicted based on one or more theories selected from constant Reynolds number, constant energy input per unit volume, constant linear velocity and constant Froude number.

[0028] Embodiment 3: According to any prediction system in embodiments 1 to 2, the "data-driven part" is based on machine learning and uses at least one algorithm selected from the group consisting of Boost algorithm, SVM (support vector machine) algorithm, neural network algorithm, naive Bayes algorithm, decision tree algorithm, LSH (local sensitive hashing) algorithm and KNN (K nearest neighbor) algorithm.

[0029] Embodiment 4: According to any one of the prediction systems in Embodiments 1 to 2, the "decision-making assistance part" is based on a knowledge map constructed from expert knowledge, including a plurality of knowledge trees.

[0030] Embodiment 5: The prediction system according to any one of embodiments 1 to 2 is used to predict the scale-up of cathode active material (CAM) production in lithium-ion batteries.

[0031] Embodiment 6: A prediction system according to embodiment 5, wherein the input and output data include equipment parameters set for each step of preparing cathode active material (CAM), raw material formula, and quality analysis results of prepared CAM samples or products.

[0032] Embodiment 7: A prediction method for scaling up production, comprising the following steps: (i) Input the initial data from the bench-scale experiments into the “Theoretical Scale-up Section” to obtain the predicted scale-up parameters; (ii) using a “data-driven portion” to adjust predicted amplification parameters based on one or more machine learning algorithms; (iii) resolving the direction and magnitude of the correction of the data obtained from step (ii) through the “decision-making assistance part”; (iv) the revised pilot production parameters obtained from step (iii) for use in the prediction of the next round of steps (i) to (iv); and (v) Repeat steps (i) to (iv) until the final prediction result is consistent with the actual result.

[0033] The batch process parameters, batch equipment parameters and batch product quality parameters obtained from the laboratory test are input into the "production scale-up theory part" to obtain the predicted expansion parameters, which are input into the "data-driven part" for adjustment, and the adjusted predicted parameters are input into the "decision-making assistance part" to generate control parameters that can be used for factory trial production, and the batch process parameters, batch equipment parameters and batch product quality parameters obtained from the factory trial production can be used to further improve the parameter settings in the laboratory test. Such a process can be iterated multiple times to continuously optimize the prediction system for production scale-up.

[0034] Embodiment 8: According to the prediction method of embodiment 7, the "theoretical part of production scale expansion" is predicted based on one or more theories selected from constant Reynolds number, constant energy input per unit volume, constant linear velocity and constant Froude number.

[0035] The equipment parameters, process parameters and product quality parameters obtained from laboratory tests are transformed into parameters that can be used for factory trial production through the "production scale-up theory part", which is based on, but not limited to, constant Reynolds number (Re), constant energy input per unit volume (P / V), constant linear velocity and constant Froude number, etc. The subscript "B" represents "B sample" (i.e., the sample obtained from laboratory tests), and the subscript "C" represents "C sample" (i.e., the sample obtained from factory trial production).

[0036] Embodiment 9: According to any prediction method in embodiments 7 to 8, the "data-driven part" is based on machine learning, using one or more selected from enhancement algorithm, SVM (support vector machine) algorithm, neural network algorithm, Bayesian algorithm, decision tree algorithm, LSH (local sensitive hashing) algorithm and KNN (K nearest neighbor) algorithm.

[0037] The parameters obtained from the "production scale-up theory part", the test product quality parameters, the laboratory sample parameters and the quality test results of the laboratory samples are input into the "data-driven part" to obtain the prediction parameters for the factory trial production. The existing product database used for the "data-driven part" includes raw material information, process parameters and product quality information, and is combined with one or more algorithms such as Boost algorithm, SVM (support vector machine) algorithm, neural network algorithm, Bayesian algorithm, decision tree algorithm, LSH (local sensitive hashing) algorithm and KNN (K nearest neighbor) algorithm for data processing. The existing product database will be updated in real time to improve the accuracy of the simulation.

[0038] Embodiment 10: A prediction method according to any one of embodiments 7 to 8, wherein the "decision-making assistance part" is based on a knowledge map constructed from expert knowledge, including a plurality of knowledge trees.

[0039] The core part of the "decision-making support part" is the establishment of the knowledge tree. The target parameter is product quality. Once a certain product quality parameter is focused on, the knowledge tree will provide all the process parameters related to it, including process parameters with different degrees of positive and negative correlation. After finding the key process parameters, one or more control parameters will be determined for factory trial production. Here, the term "product quality parameter" means the product performance target to be achieved, the term "process parameter" means any factor that affects the product preparation process, and the term "control parameter" means any parameter that can be controlled or read in the preparation of the product. The process parameter can be achieved by one or more control parameters, and sometimes the process parameter is equal to the control parameter, such as temperature or time.

[0040] Implementation 11: According to any prediction method in Implementation 7 to 8, in step (v), it is determined whether the final prediction result conforms to the actual parameters according to Formula 1: R2 = 1 - Σ i (y i -y f ) 2 / Σ i (y i -y a ) 2 Formula 1 Among them, y i is a real data point, y f is with y i The corresponding predicted point, y a Yes i When the R2 ranges from 0.8 to 1, the prediction results are acceptable.

[0041] Embodiment 12: According to any prediction method in Embodiments 7 to 8, in step (v), it is determined whether the final prediction result conforms to the actual parameters according to Formula 1: R2 = 1 - Σ i (y i -y f ) 2 / Σ i (y i -y a ) 2 Formula 1 Among them, y i is a real data point, y f is with y i The corresponding predicted point, y a Yes i When R2 ranges from 0.9 to 1, the prediction results are acceptable.

[0042] Embodiment 13: The prediction method according to any one of embodiments 7 to 8 is used to predict the scale-up of cathode active material (CAM) production in lithium-ion batteries.

[0043] Embodiment 14: A prediction method according to embodiment 13, wherein the input and output data include equipment parameters set for each step of preparing cathode active material (CAM), raw material formula, and quality analysis results of prepared CAM samples or products.

[0044] The following describes an embodiment of a prediction system and prediction method for applying the present invention to a pilot production line of a cathode active material (CAM) water washing process to expand the production scale. Those skilled in the art may adjust the specific operations to achieve the same or nearly the same effect without departing from the spirit of the present invention. This embodiment is only an example and does not limit the scope of protection of the present invention.

[0045] In the production process of cathode active materials (CAM), due to the gap between laboratory and pilot production line equipment, the water washing process parameters of the pilot production line need to be quickly adjusted and continuously optimized for products of different specifications. Typical water washing process equipment includes (1) a metering box for CAM semi-finished products, equipped with a weight loss metering unit to measure the weight of the material; (2) a pure water storage tank with a cooling coil for providing cooling water to the washing kettle; (3) an online liquid flow meter located between the pure water storage tank and the washing kettle for checking the flow of pure water; and (4) a washing kettle connected to the metering box and the pure water storage tank, equipped with a temperature detection device to monitor the water washing temperature and a stirring device to control the stirring paddle speed.

[0046] The control parameters in this embodiment are temperature and pure water weight, washing material weight, flushing water weight and stirring frequency. The control parameters here are equivalent to process parameters. Product quality parameters include quality parameters of washed materials, such as humidity, specific surface area, residual alkali content and magnetic impurity content.

[0047] According to the "production scale-up theory part", different process steps need to be classified. Different operating parameters in each process step will affect the mechanism parameters and product attribute parameters in the theoretical model. According to the mechanism of a specific part and the process parameters of input B sample (i.e., the sample obtained from the laboratory scale test), the operating parameters corresponding to sample C (i.e., the sample obtained in the factory trial production) are calculated by the equivalent amplification principle, and proportional expansion is carried out according to the invariance principle of the mechanism parameters.

[0048] During the washing process, the equipment used in the laboratory is different from the equipment used in the factory pilot production. The constant Froude number theory is used to ensure that the mixing conditions remain unchanged from the laboratory equipment to the factory pilot production equipment. Given the stirring speed, the stirring paddle diameter for sample B and the stirring paddle diameter for sample C, the theoretical stirring speed value for sample C can be calculated using the constant Froude number theory.

[0049] To train the "data-driven part", data needs to be collected according to different product types, including the actual values ​​of all process parameters, the quality inspection results of samples B and C, and the theoretical values ​​of the process parameters of sample C obtained from the "production scale-up theory part" (collectively referred to as "historical data"). All historical data must be converted into a readable format suitable for machine learning algorithms (such as LSH, local sensitive hashing). The target parameter is the actual stirring speed in the factory trial production. Through algorithm iteration and adjustment of algorithm parameters, the predicted value converges to the actual stirring speed, that is, the determination coefficient R2 should be as close to 1 as possible.

[0050] The steps used by the production scale-up prediction system to predict stirring speed include: (1) Based on the actual stirring speed of sample B, the theoretical stirring speed of sample C is calculated using the “theoretical part of production scale expansion”; (2) The actual stirring speed of sample B, the actual test results of the product quality of sample B, the theoretical stirring speed of sample C, and the target product quality parameters of sample C are input into the "data-driven part" in a readable format for machine learning algorithm (such as LSH, local sensitive hashing) processing, and the optimized theoretical stirring speed of sample C is output; (3) Apply the optimized theoretical stirring speed of sample C to actual factory trial production to verify its reliability; (4) Repeat the above steps (1) to (3) until the error of the product quality test result is within the error range of ±5%.

[0051] When new products appear, the product database must be updated in time to train the “data-driven part” to ensure the accuracy of the data source.

[0052] The "decision-making support part" is based on the knowledge mapping formed by interviewing experts. According to the production process of CAM, control parameters, process parameters and product quality parameters are collected, and the correlation between these parameters is explored to form a knowledge tree, thereby building a complete decision-making support system.

[0053] After obtaining the theoretical stirring speed from the "data-driven part", it is necessary to verify the speed in the workshop trial production to identify the deviation between the theoretical value and the actual situation (such as high residual alkali), and enter the deviation into the "decision-making support part" for retrieval. Based on the process parameters related to the residual alkali index, the "decision-making support part" retrieves the process parameters that need to be adjusted on the production line on the branch of the knowledge tree and confirms the adjustment direction of the process parameters, such as reducing the stirring speed, reducing the weight of the intermediate, increasing the mixing time, etc. The workshop operator can decide what suggestions to take based on the actual situation of the production line.

Claims

1. A prediction system for scaling up production, comprising the following parts: (a) "Scale-up theory", which predicts the process parameters for scale-up to industrial plants based on data obtained from bench-scale experiments; (b) the "data-driven part", adjusting the scale-up parameters obtained from the "production scale-up theory part"; as well as (c) The "decision-making support part" outputs the revised parameters of the factory trial production based on the data obtained from the "data-driven part".

2. The prediction system according to claim 1, wherein the "production scale-up theoretical part" is predicted based on one or more theories selected from the group consisting of constant Reynolds number, constant energy input per unit volume, constant linear velocity and constant Froude number.

3. According to the prediction system described in any one of claims 1 to 2, the "data-driven part" is based on machine learning and adopts at least one algorithm selected from the group consisting of Boost algorithm, support vector machine algorithm, neural network algorithm, naive Bayes algorithm, decision tree algorithm, local sensitive hashing algorithm and K nearest neighbor algorithm.

4. The prediction system according to any one of claims 1 to 2, wherein the "decision-making assistance part" is based on a knowledge map constructed from expert knowledge, including a plurality of knowledge trees.

5. The prediction system according to any one of claims 1 to 2, used for predicting the scale-up of production of cathode active materials in lithium-ion batteries.

6. The prediction system according to claim 5, wherein the input and output data include equipment parameters set for each step of preparing cathode active materials, raw material formulas, and quality analysis results of prepared cathode active material samples or products.

7. A prediction method for scaling up production, comprising the following steps: (i) Input the initial data from the bench-scale experiments into the "Theoretical Scale-up Section" to obtain the predicted scale-up parameters; (ii) adjusting the predicted amplification parameters using a "data-driven portion" based on one or more machine learning algorithms; (iii) resolving the direction and magnitude of the correction of the data obtained from step (ii) through the "decision-making aid part"; (iv) the revised pilot production parameters obtained from step (iii) are used for prediction of the next round of steps (i) to (iv); as well as (v) Repeat steps (i) to (iv) until the final prediction result is consistent with the actual result.

8. The prediction method according to claim 7, wherein the "production scale-up theoretical part" is predicted based on one or more theories selected from the group consisting of constant Reynolds number, constant energy input per unit volume, constant linear velocity and constant Froude number.

9. The prediction method according to any one of claims 7 to 8, wherein the "data-driven part" is based on machine learning, using one or more selected from a boosting algorithm, a support vector machine algorithm, a neural network algorithm, a Bayesian algorithm, a decision tree algorithm, a local sensitive hashing algorithm, and a K nearest neighbor algorithm.

10. The prediction method according to any one of claims 7 to 8, wherein the "decision-making assistance part" is based on a knowledge map constructed from expert knowledge, including a plurality of knowledge trees.

11. The prediction method according to any one of claims 7 to 8, wherein step (v) comprises: Determine whether the predicted result is consistent with the actual result according to Formula 1: R2 = 1 - Σ i (y i -y f ) 2 / Σ i (y i -y a ) 2 Formula 1 Among them, y i is a real data point, y f is with y i The corresponding predicted point, y a Yes i When R2 ranges from 0.8 to 1, it is determined that the predicted results are consistent with the actual results.

12. The prediction method according to any one of claims 7 to 8, wherein step (v) comprises: Determine whether the predicted result is consistent with the actual result according to Formula 1: R2 = 1 - Σ i (y i -y f ) 2 / Σ i (y i -y a ) 2 Formula 1 Among them, y i is a real data point, y f is with y i The corresponding predicted point, y a Yes i When R2 ranges from 0.9 to 1, it is determined that the predicted results are consistent with the actual results.

13. The prediction method according to any one of claims 7 to 8, used for predicting the scale-up of production of cathode active materials in lithium-ion batteries.

14. The prediction method according to claim 13, wherein the input and output data include equipment parameters set for each step of preparing the cathode active material, raw material formula, and quality analysis results of the prepared cathode active material samples or products.

Images